Field Reports from Erik's Laboratory

We should talk about how the AI revolution is impacting our mental health

2026-04-10T00:00:00-04:00

How AI is impacting the mental health of everyone is not getting enough discussion, and I also don't know what we do about it. While AIs may be able to help us get more done faster, that increased productivity may not be sustainable and does not come as a reduction in either stress or anxiety. As I suspect this is impacting many of us, whether we acknowledge them or keep them buried, it is important for leaders to acknowledge this and talk about it so that people don’t feel alone with their fears.

I'm coming from a position of thinking that emerging AI technologies are highly disruptive and transformative in the same way that the internet was in the late 90's. This doesn't mean that it will solve every problem or that it is actually a good idea for everything, but it is here to stay. I think the overall comparisons to Blockchain/NFTs/VR are misleading (as those were technologies in search of problems in a way that LLMs are not). As someone who has often been a selective early adopter of technologies (I had a Linux web server in my dorm room in 1994, had voice controlled home automation in the 2004, and I was pushing the boundaries of ChatGPT within a week of its release), this is a technology I’m willing to experiment with and am finding increasingly useful for certain tasks. In particular, I’ve been finding LLMs like Claude especially useful for domains where I’m an expert and where it is easier to fully review output than to do everything from scratch. For example, using Claude as an augmented search tool to help me find primary sources is transformative in the same way that Altavista was transformative over Yahoo.

Within the work environment there are certainly compelling and transformative uses of AI and LLMs, and there are also many places where introducing it is counter-productive at this point and/or where the risks outweigh the benefits. A colleague of mine put it well on how to approach new uses of AI: “ok, fine, let's try AI for all of this, but at the end of the day I'm a scientist so how do we measure and how do we learn from this."

But back to my initial premise, all over the place I see AI technologies as really, really triggering the fight-or-flight threat instincts in most everyone. Regardless of how experienced and senior people are, there seems to be uniform reactions of "Can I keep up? Will this replace me career-wise? How do I manage the technical risks? What are the existential risks to our civilization? What are the ethical implications and the environmental costs? How will this impact my children’s future path?” and many more. Since “Code is a liability. Code’s capabilities are assets.” (- Cory Doctorow), those of us working at tech companies are also seeing a rapidly and unsustainably ballooning amount of code that we will need to maintain.

All of the things that our “Growth Mindset” training classes taught us about are all directly applicable here, where the perceived threats and the high rate of change are almost certainly having psychological, sociological, and physiological impacts on almost all of us. Secondary discussions around this are one of the top topics of conversation everywhere it seems, whether in social media, work chats, or at social gatherings.

Some people are responding with extreme skepticism as a natural reaction, and in the creative fields this may be highly warranted – while AI can write novels, is this something people actually want to read? But in other areas this is hazardous and may just result in increasing cognitive dissonance. If we take the comparison to the Internet bubble in the early 2000, there was certainly hype and over-investment, but that rapid rate of investment did give many companies the momentum they needed to build great things we rely on heavily today. Even things that there was huge skepticism around (who wants to do banking online?) have now turned out to be part of our daily lives.

Other people seem to be taking the Growth Mindset to the extreme, embracing AI evangelism and going after it all without skepticism. Yes, it is possible to do a huge amount if you have no family (or prioritize work/innovation above all else) and work a huge number of hours of the day, but it is neither equitable nor healthy to force everyone to make this choice. Given the rate of change here, I fear that even this is unsustainable, and many of the people on this path are also going to get burned badly by all of the security and other pitfalls present in even the best of these systems.

For those of us who think of ourselves as pragmatic innovators, we can try and take a middle ground but even that middle ground is extremely stressful. There is no way anyone can keep up with the rate of change, and new threats are showing up constantly. While many people might have the option of trying to ignore it, it is still important for everyone to be aware of all of the new threats from bad actors that are leaking into the real world. For those of us in technical roles, ignoring the transformative nature of AI isn’t an option — while there are lots of bad uses, there are also lots of compelling uses, and we need to have enough familiarity to be able to differentiate and to use the tools well.

Trying to keep up here while also dealing with the emerging threats and also trying to do our jobs is exhausting, and is feeling increasingly unsustainable. While offloading things to AI might seem like an attractive solution, it doesn’t really help and makes things worse as the only way to safely use AIs today is to put up enough guardrails to be able to be personally responsible for their outputs and actions. This means that while they can help us be more productive, that increased productivity does NOT come as a reduction in either stress or anxiety.

I’m not sure what we can do about this? The only thing that I can think of is empathy and acknowledging the levels of stress that our present-day future is forcing onto ourselves and everyone around us, along with appreciating that different people are processing it in different ways. Human connections may also help some people but make things worse for others. While this seems like an important topic for those of us in tech leadership roles to talk about and try and address, the solution is NOT to try and use AI to solve it. (My dystopian fear is bringing this to HR departments or management who then suggest that employees talk to chatbots for psychological support — if anyone seriously suggests this there is plenty of research saying that doing so would be criminally negligent.) But no matter what, try and be good to the humans around you.

Using DHCPv6-PD on Ubuntu 22.04 Jaunty with systemd-networkd to route multiple prefixes

2023-08-12T14:00:00-04:00

This provides an overview of how you can use the systemd-networkd service in Ubuntu 22.04 to obtain an IPv6 DHCPv6-PD prefix (eg, a /56) from an upstream ISP and subdivide it across local subnets, providing a /64 per subnet.

Note that this post assumes you already have extensive experience with Linux and IPv6. Most people are much better off using a distribution such as OpenWRT that handles all of this complexity for you. If you want to know why IPv6 matters, check out Akamai's blog post on "10 Years Since World IPv6 Launch".

Overview

The systemd-networkd service in Ubuntu 22.04 has enough support for DHCPv6-PD to obtain a prefix from an upstream ISP and subdivide it across local subnets, providing a /64 per subnet. For example, Comcast provides a /60 which has enough space for 16 subnets.

This includes substantial improvements since Ubuntu 20.04 (which I wrote about previously although the syntax for some of the configuration has changed. This is an updated version and assumes Ubuntu 22.04 Jaunty (or equivalent).

Because netplan doesn't have enough features to configure DHCPv6-PD yet (see https://bugs.launchpad.net/netplan/+bug/1771886), some of the configuration needs to live in systemd-networkd overrides.

Configuring the Interfaces

I am assuming starting with a netplan configuration like the following:

network:

  ethernets:

    eno1:    # ISP1
      dhcp4: true
      dhcp6: true
      accept-ra: true
      ipv6-privacy: false
      nameservers:
        search: [ example.org ]
      dhcp4-overrides:
        use-dns: false
        use-hostname: false
        use-ntp: false
      dhcp6-overrides:
        use-dns: false
        use-hostname: false
        use-ntp: false

    eno2:    # LAN
      dhcp4: false
      optional: true

  version: 2

  vlans:

    en-main:
      id: 6
      link: eno2
      dhcp4: no
      addresses:
        - 10.1.0.0/24
      optional: true

    en-guest:
      id: 7
      link: eno2
      dhcp4: no
      addresses:
        - 10.2.0.0/24
      optional: true

This has interfaces:

eno1 = WAN
eno2 = Primary LAN
en-main = Main VLAN
en-guest = Guest VLAN

While netplan and systemd-networkd somehow collaborate to put this netplan config into files such as /run/systemd/network/10-netplan-eno1.network, there is no way to get the rules needed for enabling DHCPv6-PD into those dynamically generated files in /run.

Thankfully, you can put overrides in that systemd-networkd will read. See the systemd.network(5) man page for more details on all of the options available for these files.

To configure a DHCPv6-PD client on the WAN interface (eno1 in my case), create /etc/systemd/network/10-netplan-eno1.network.d/override.conf :

[Match]
Name=eno1

[DHCPv6]
PrefixDelegationHint=::/56

You can specify a smaller or larger value for PrefixDelegationHint if needed (eg, /60 for Comcast Xfinity Residential).

For each of the LAN interfaces, create files such as /etc/systemd/network/10-netplan-en-guest.network.d/override.conf and /etc/systemd/network/10-netplan-en-main.network.d/override.conf with the following. Note that the filename prefixes must match those in /run/systemd/network/ (ie, don't change the "10-").

[Match]
Name=en-main

[Network]
IPv6PrefixDelegation=dhcpv6
IPv6DuplicateAddressDetection=1
LinkLocalAddressing=ipv6

[DHCPv6PrefixDelegation]
## Must be unique per subnet
SubnetId=6

The "SubnetId" here must be unique per subnet and is used to construct the address. The allowed values depend on what size prefix was delegated by DHCPv6-PD. If you have a /60, this can be a unique hex value in the range 0-f. If you have a /56, this can be a unique hex vale in the range of 0-ff.

For example, if you have 2001:db8:abcd:3400::/56 and have SubnetId of "7" then the subnet will be assigned the prefix 2001:db8:abcd:3407::/56.

It is critical to have this here as otherwise these will be assigned in different orders each time systemd-networkd restarts. It doesn't keep track of what prefixes it has advertised to links, so hosts on various links/subnets could end up with stale prefix advertisements for the wrong subnets and broken connectivity.

As a variation on the above, the following also: * Assigns a ULA prefix on the subnet which remains up even if you lack external network connectivity. Make sure to replace the "XX:XXXX:XXXX" part of "fdXX:XXXX:XXXX:YYYY::/64" with a sufficiently random value, and then the "YYYY" may wish to be the same as the SubnetId as above (but does not need to be). * Has the router also assign itself an address on the subnet with a token of "::7777", so it would assign itself 2001:db8:abcd:3407::7777.

[Match]
Name=en-main

[Network]
## use both DHCPv6 and static ULA prefixes ("yes")
IPv6PrefixDelegation=yes
IPv6DuplicateAddressDetection=1
LinkLocalAddressing=ipv6

[DHCPv6PrefixDelegation]
## Must be unique per subnet
SubnetId=6
Token=::7777
Assign=true

[IPv6Prefix]
Prefix=fdXX:XXXX:XXXX:YYYY::/64

Note the filename prefixes in /etc/systemd/network/ need to match those in /run/systemd/network/.

For more details on the options and their meanings, see the man page systemd.network(5).

Loading Changes

You can either reboot, or you can reapply netplan and reload systemd-networkd configuration:

netplan apply
networkctl reload

Sometimes it can help to also restart restart systemd-networkd with:

service systemd-networkd restart

Caveats and notes

Some additional notes:

The IPv6 address on the WAN link (eg, eno1) comes from an individual RA ND or DHCPv6 assignment, not a DHCPv6-PD prefix.

Testing and Debugging

There's plenty more information than existed in older versions, especially by running networkctl status. This will show the various assigned addresses as well as some recent log entries. For example, you can see which prefix was obtained via DHCPv6-PD and what was assigned to various links:

eno1: DHCPv6 address 2001:db8:501:b9:1111:2222:3333:4444/128 (valid for 3d 40min 1s, preferred for 3d 40min 1s)
eno1: DHCP6: received PD Prefix 2001:db8:1234:950::/60
en-main: DHCPv6-PD address 2001:db8:1234:956::7777/64 (valid for 3d 40min 1s, preferred for 3d 40min 1s)
en-guest: DHCPv6-PD address 2001:db8:1234:957::7777/64 (valid for 3d 40min 1s, preferred for 3d 40min 1s)

Here the DHCPv6-PD prefix of 2001:db8:1234:950::/60 was assigned, and then 2001:db8:1234:956::/64 was assigned out to the link with SubnetId "6".

Some things that may be helpful:

ip -6 route list will list the prefixes assigned to the interfaces
rdisc6 en-guest will do router discovery on that interface (eg, en-guest in this case) and will show the assigned prefix
Doing a tcpdump -s 1500 -vvv -i eno1 port 546 or port 547 and then a systemctl restart systemd-networkd will show the DHCPv6-PD lease including prefix length received from the WAN ISP
Doing journalctl -xeu systemd-networkd will give some logging information from systemd-networkd but I have yet to figure out how to get the most relevant parts logged.
Doing radvdump will show the RAs showing up on an interface.

More network logging

Doing systemctl edit systemd-networkd to add in:

[Service]
Environment=SYSTEMD_LOG_LEVEL=debug

And restart systemd-networkd service and looking at logs:

systemctl restart systemd-networkd
journalctl -b -u systemd-networkd

Multi-Homing and clearing bad prefixes

I've also started to play more with Primary/Backup failover between two links. Using the approach from RFC8475 - Using Conditional Router Advertisements for Enterprise Multihoming seems to work, although I additionally configure NAT66 for cases where traffic tries to use the wrong source address for the wrong link.

One thing that is useful is to use the ra6 tool from ipv6tools to set the preferred lifetime down to "0" to deprecate a prefix on a link. For example, this sets the preferred lifetime to "0" for an old prefix "2001:db8:5050:553::/64" which had been on a link with interface "en-guest", sending it to the all-hosts multicast (ff02::1):

ra6 -v -i en-guest -e -P 2001:db8:5050:553::/64#AL#7200#0 --lifetime 300 -r 0 -x 0 -c 0 -p -1 -d ff02::1

It is possible not all of these arguments are strictly needed.

You may need to send this periodically (for the valid lifetime that had been set previously) as hosts may be offline or may be rate-limiting RAs.

This can be useful for misconfigurations (eg, if changing SubnetId), cleaning up from botched ISP renumberings, and for primary/backup failovers.

Future Work

It would be nice if systemd-networkd kept track of IPv6 prefixes it had assigned to links so that it could deprecate those if they were no longer in-use (eg, when an ISP does a renumbering), and to also support Conditional RAs for primary/backup failovers between multiple uplinks.

It would also be nice if all of this could be configured via netplan rather than with overrides.

I also plan to setup an IPv6-only subnet using NAT64+DNS64 and will have a follow-on blog post on the details there.

Updates

2023-08-26: removed "ForceDHCPv6PDOtherInformation" which has been deprecated

Creating Mastodon account identity aliases with Apache

2022-11-26T14:00:00-05:00

Update 2022-11-30: added a RewriteMap to unescape URI-encoding, per @jered@convivian.com's comment and added a pointer to webfinger.net.

Update 2022-12-16: fixed a syntax error for host-meta XML and added in the proper content-type

I've long operated email forwarding (eg, of the form "username@personaldomain.example") which forwards on to my actual email service. This means I can switch backend email service providers without needing to tell everyone my new address. While Mastodon is federated and allows migrating accounts and followers between servers (as long as the server you are migrating from is still available), it wasn't immediately obvious how to setup a stable "identity" on my own domain without running my own server. (While I considered running one, I really don't want to have to be the SRE for yet another personal service, and hence joined the well-operated Hachyderm server.)

It is possible, as described below to, easily setup forwarding of a fediverse account name from a personal domain/server to a third-party operated fediverse (eg, Mastodon) server.

While this allows for discovery (eg, you can put "@username@personaldomain.example" into a Mastodon search field to find your account), this is still just a redirection. Followers and posts are still associated with the underlying Mastodon server.

Background

Mastodon and other "fediverse" services provide federated social networking and microblogging built on top of the ActivityPub and WebFinger and HTTP protocols. While many people have been used to centralized social network services such as Twitter and Facebook, the fediverse is much closer to email and websites which have always been federated (even if there are popular big operators).

Naming in the fediverse leverages the DNS (Domain Name System), again the same as email and websites. The security model also relies on PKIX (the Public Key infrastructure underlying HTTPS where Certificate Authorities validate ownership over DNS names and sign keyparts used to authenticate TLS connections).

Mastodon/fediverse/ActivityPub/WebFinger use "acct:" scheme URIs to name accounts. The WebFinger protocol (RFC 7033) is used to map from an acct: URI to the various ActivityPub service endpoints by making an HTTPS request to a well-known URI.

Warnings/Caveats

One warning before we start: this breaks the abstraction barrier of the target server you are forwarding to. They usually return the WebFinger configuration listing their endpoints, and by copying and modifying them it means things could break if the target server adds new endpoints or needs to change its URIs.

The forwarding also only works initially and subsequent interactions are with the target server. This could be confusing for users, especially if the username changes substantially.

At the same time, you can add additional WebFinger information in if you wanted to, such as phone numbers or similar. I'm curious what real use-cases for this might be.

WebFinger forwarding with Apache

The easiest way to setup WebFinger forwarding is to use mod_rewrite. If it's not enabled:

a2enmod rewrite

Then in the VirtualHost stanza for the Apache2 configuration for your domain include:

   RewriteEngine on
   RewriteMap unescape int:unescape
   RewriteCond ${unescape:%{QUERY_STRING}} ^(rel=[^&]*&)*resource=acct:([a-z0-9-]+)@example.org(&rel=[^&]*)*$
   RewriteRule ^/.well-known/webfinger$ /.well-known/_webfinger/%2.json [L,QSD]

   <Location "/.well-known/host-meta">
      ForceType application/xrd+xml
   </Location>

Replace "example.org" with your personal domain name (eg, "personaldomain.example"). This says to rewrite URLs of the form "https://example.org/.well-known/webfinger?resource=acct:username@example.org" to instead serve files from "/.well-known/_webfinger/username". (It also ignores any rel= query string attributes, per the RFC.)

Note: the RewriteMap to unescape URI-encoding is needed for some clients. Thanks to @jered@convivian.com for pointing this out.

Next, create a file in /.well-known/_webfinger/username.json (replacing "username" with yours) containing:

{
   "aliases" : [
      "https://mastodon-server.example/@username",
      "https://mastodon-server.example/users/username",
      "acct:username@mastodon-server.example"
   ],
   "links" : [
      {
         "href" : "https://mastodon-server.example/@username",
         "rel" : "http://webfinger.net/rel/profile-page",
         "type" : "text/html"
      },
      {
         "href" : "https://mastodon-server.example/users/username",
         "rel" : "self",
         "type" : "application/activity+json"
      },
      {
         "rel" : "http://ostatus.org/schema/1.0/subscribe",
         "template" : "https://mastodon-server.example/authorize_interaction?uri={uri}"
      }
   ],
   "subject" : "acct:username@personaldomain.example"
}

In that, replace "mastodon-server.example" with the URL of your mastodon server, "username" with your username, and "username@personaldomain.example" with your full vanity domain account thatyou want to redirect over.

Note that the above corresponds to at least one Mastodon 4.0 server. You may need to periodically update this to match the configuration for your mastodon server. For example, fetch with:

curl https://mastodon-server.example/.well-known/webfinger?resource=acct:username@mastodon-server.example | json_pp

and start with that instead.

Purely optional: If you want, you can add additional link relations with additional metadata under "links". See here for a registry with other link relations.

You may also need to add a file /.well-known/host-meta with a URI template for some webfinger clients to use, also replacing "personaldomain.example" with your domain name:

<?xml version="1.0" encoding="UTF-8"?>
<XRD xmlns="http://docs.oasis-open.org/ns/xri/xrd-1.0">
  <Link rel="lrdd" template="https://personaldomain.example/.well-known/webfinger?resource={uri}" type="application/xrd+xml" />
</XRD>

Once these are in-place, reload your apache configuration (eg, "apache2ctl graceful") and then test:

curl 'https://personaldomain.example/.well-known/webfinger?resource=acct:username@personaldomain.example'

which should return the json document.

You can also test by entering in your account name at webfinger.net.

You should be able to then follow "@username@personaldomain.example" into a search box in a Mastodon client would then bring up "@username@mastodon-server.example".

As an example, "@erik@nygren.org" is setup this way and points to "@nygren@hachyderm.io".

Other Implementations

Some other implementations that may be helpful:

Acknowledgements

Some useful posts which pointed me in the right direction:

Setting up an ActivityPub feed as a static site
Better Mastodon aliases
Image generated with OpenAI DALL-E 2

(Obselete) Using DHCPv6-PD on Ubuntu 20.04 with systemd-networkd to route multiple prefixes

2022-06-18T14:00:00-04:00

Update: Ubuntu 22.04 has significant improvements in the systemd-networkd service that fix a number of issues from this blog post. The syntax has changed. See Using DHCPv6-PD on Ubuntu 22.04 Jaunty with systemd-networkd to route multiple prefixes for an updated version of this.

Overview

The systemd-networkd service in Ubuntu 20.04 has enough support for DHCPv6-PD to obtain a prefix from an upstream ISP and subdivide it across local subnets, providing a /64 per subnet. For example, Comcast provides a /60 which has enough space for 16 subnets.

As far as I can tell, the ability to debug what is going on here or obtain state is almost entirely lacking. This makes it very hard to use troubleshoot issues or even tell what is going on.

Because netplan doesn't have enough features to configure DHCPv6-PD yet (see https://bugs.launchpad.net/netplan/+bug/1771886), some of the configuration needs to live in systemd-networkd overrides.

Configuring the Interfaces

I am assuming starting with a netplan configuration like the following:

network:

  ethernets:

    eno1:    # ISP1
      dhcp4: true
      dhcp6: true
      accept-ra: true
      ipv6-privacy: false
      nameservers:
        search: [ example.org ]
      dhcp4-overrides:
        use-dns: false
        use-hostname: false
        use-ntp: false
      dhcp6-overrides:
        use-dns: false
        use-hostname: false
        use-ntp: false

    eno2:    # LAN
      dhcp4: false
      optional: true

  version: 2

  vlans:

    en-main:
      id: 6
      link: eno2
      dhcp4: no
      addresses:
        - 10.1.0.0/24
      optional: true

    en-guest:
      id: 7
      link: eno2
      dhcp4: no
      addresses:
        - 10.2.0.0/24
      optional: true

This has interfaces:

eno1 = WAN
eno2 = Primary LAN
en-main = Main VLAN
en-guest = Guest VLAN

Thankfully, you can put overrides in that systemd-networkd will read. See the systemd.network(5) man page for more details on all of the options available for these files.

To configure a DHCPv6-PD client on the WAN interface (eno1 in my case), create /etc/systemd/network/10-netplan-eno1.network.d/override.conf :

[Network]
IPv6PrivacyExtensions=no

[DHCPv6]
ForceDHCPv6PDOtherInformation=yes
PrefixDelegationHint=::/56

You can specify a smaller value for PrefixDelegationHint if needed (eg, /60 for Comcast Residential).

For each of the LAN interfaces, create files such as /etc/systemd/network/10-netplan-en-guest.network.d/override.conf and /etc/systemd/network/10-netplan-en-main.network.d/override.conf with:

[Network]
IPv6PrefixDelegation=dhcpv6
IPv6DuplicateAddressDetection=1
IPv6PrivacyExtensions=no
LinkLocalAddressing=ipv6

[IPv6PrefixDelegation]
RouterLifetimeSec=1800

Note that if you want to use both DHCPv6 provided addresses and your own (eg, to mix in ULA) this may be possible by setting IPv6PrefixDelegation=true, but I have not tried this. (This did not appear to work in Ubuntu 20.04.)

Note the filename prefixes in /etc/systemd/network/ need to match those in /run/systemd/network/.

Loading Changes

You can either reboot, or you can restart systemd-networkd with:

systemctl restart systemd-networkd

Caveats

Some big caveats I've noticed:

The IPv6 address on the WAN link (eg, eno1) comes from an RA ND assignment, not a DHCPv6 prefix.
The individual LANs/VLANs don't appear to get IPv6 addresses assigned. I haven't yet figured out how to get on-link addresses. As such, ip addr won't show addresses on those links. (At least un Ubuntu 20.04)
- This seems fixed (or at least fixable) on Ubuntu 22.04.
A maximum of two interfaces is supported for delegating to on Ubuntu 20.04, but this is fixed in the version in 22.04.

Testing and Debugging

As of the systemd-networkd in Ubuntu 20.04 there aren't good ways to tell what is going on. In particular, networkctl status provides either no information or misleading information. (In Ubuntu 22.04 there is more debugging information.)

Some things that may be helpful:

ip -6 route list will list the prefixes assigned to the interfaces
rdisc6 en-guest will do router discovery on that interface (eg, en-guest in this case) and will show the assigned prefix
Doing a tcpdump -s 1500 -vvv -i eno1 port 546 or port 547 and then a systemctl restart systemd-networkd will show the DHCPv6-PD lease including prefix length received from the WAN ISP
Doing journalctl -xeu systemd-networkd will give some logging information from systemd-networkd but I have yet to figure out how to get the most relevant parts logged.

More network logging

Doing systemctl edit systemd-networkd to add in:

[Service]
Environment=SYSTEMD_LOG_LEVEL=debug

And restart systemd-networkd service and looking at logs:

systemctl restart systemd-networkd
journalctl -b -u systemd-networkd

Shows a little more info, but still not some of the things that I'd hope to see from looking at the systemd-networkd code.

Future

I'll update this when I upgrade to Ubuntu 22.04 as hopefully it has some more features. If you have any suggestions for better ways of doing things above, they are welcome. I'm unfortunately unable to provide tech support for anything in the above, however...

10 Years Since World IPv6 Launch

2022-06-06T12:04:00-04:00

See my work blog post:

10 Years Since World IPv6 Launch

Excerpt:

The World IPv6 Launch event on June 6, 2012, was where Akamai, major web content providers, and internet service providers (ISPs) collaborated to overcome the inertia that had been holding back the widespread deployment of IPv6. IPv6 is the current generation of the Internet Protocol (IP) and it addresses the scalability issues inherent in legacy IPv4. Over the past decade, Akamai has witnessed massive growth in IPv6 adoption, bringing it from a narrowly used protocol to a default one that carries a majority of traffic for some customers and for some significant end-user populations.

In the spirit of the 10th anniversary, this blog post will highlight 10 things that have improved with IPv6 deployment over these 10 years, as well as list another 10 areas where work still remains.

As a general trend, the past decade has been about enabling IPv6 to live alongside IPv4 by dual-stacking networks and services, with some use of IPv6-centric networks that still have IPv4 access. Increasingly, we are seeing trends appear around enabling new capabilities and decreasing complexity through the use of IPv6-only networking.

[... more ...]

Designing for multi-color 3D printing with OpenSCAD

2019-01-02T22:30:00-05:00

This post describes a way to make multi-color (multi-material / multi-extruder) objects in OpenSCAD suitable for 3D printing with a Prusa i3 MK3 MMU2. It assumes basic familiarity with OpenSCAD, Slic3r, and using the Prusa MK3. The snowman ornament shown here (thingiverse:3319509) was designed alongside with my five-year-old son and then I used it to experiment with this approach to generate a multi-color ornament.

Background

OpenSCAD allows you design and specify 3D models in a programming language, and allows you to output one part at a time (such as into an STL or AMF file). Slic3r Prusa Edition is a slicer used to do multi-color printing for the Prusa i3 MK3 MMU2. It can import either a multi-part AMF file or can load in a set of STL files, one for each color/extruder.

While OpenSCAD has support for applying colors to elements within an object, it wasn't immediately clear how to leverage this to print a multi-color object. OpenSCAD issue 1608 aims to add some more direct support, but it isn't resolved.

The approach that I've taken here is define and embed a "multicolor()" operator in the OpenSCAD object design which changes its behavior based on a global variable. I then export three separate STL files, one per color, and then import these together into Slic3r.

The full OpenSCAD file for the snowman are available here, and I scaled this down by 60% before printing.

For example, here is a rendering of the multi-color object in OpenSCAD:

Designing multi-color in OpenSCAD

I use this function to define a "multicolor" module:

/* Pick a color below for STL export, or "ALL" to show all colors. */
current_color = "ALL";
//current_color = "black";
//current_color = "white";
//current_color = "red";

/* Similar to the color function, but can be used for generating multi-color models for printing.
 * The global current_color variable indicates the color to print.
 */
module multicolor(color) {
    if (current_color != "ALL" && current_color != color) { 
        // ignore our children.
        // (I originally used "scale([0,0,0])" which also works but isn't needed.) 
    } else {
        color(color)
        children();
    }        
}

This can then be used to allow different additive solids to be rendered in different colors. As a very basic use-case:

multicolor("red") cube(20, center=true);
multicolor("green") translate([0,0,10]) sphere(10);

Which when rendered with current_color = "ALL"; results in a multi-color object:

In this above case, you could switch to each of current_color = "red"; and current_color = "green"; one at a time, and then for each do a Design->Render (F6) followed by an export (File->Export->Export as STL) as two separate files.

This module works by only applying a color and rendering the children when current_color is "ALL" or when it equals the color passed to multicolor(). Otherwise it ignores the tree of elements underneath.

Note that this only works well when additive primitives (such as union) are used to combine multicolor objects. You can have subtractive primitives (difference/intersection) as children of multicolor(), but with this approach you can't directly use subtractive primitives on multicolor() objects. (The color() function will make things appear like they might work, but the resulting rendered parts won't be proper.) I found that for many use-cases this limitation wasn't a problem.

Note that this approach relies on Slic3r being OK with overlapping parts within an object. It has an ordering such that parts higher in the list take precedence over parts lower in the list for an object.

Once the object is designed, for each color:

Comment out all other colors
Do a a Design->Render (F6)
Export the STL file (File->Export->Export as STL)

For example, with the snowman example above here is what just red looks like when rendered:

I found it helpful to export the parts with filenames that can be sorted in the order of the desired extruders. For example with the snowman I might export out three separate STLs with:

This is because I wanted black and red to have precedence over white when I load the parts into Slic3r.

Importing into Slic3r

Once you have STL exports, you'll want to import them into Slic3r to get them ready to print. Some initial steps:

Make sure you have the MMU2 Printer selected (such that multiple Filaments are available).
Go to the Printer Settings tab and for each Extruder on the side set the Extruder Preview Color (eg, to black, red, and white).
Update the Purging Volumes on the main page. This will open up a table. Especially for a color such as white, you'll need to greatly increase the volume to purge while loading. (Even upping from the default 70 to 150 still results in a pinkish or greyish hue to white on layers that also included red or black.)
Click Add... and select all of the STL files, and say Yes when it asks to load them as multiple parts of a single object.
Scale and rotate the object as desired. I scaled this snowman ornament down by 60%.

Once the object is loaded and positioned and scaled, double click on the loaded object. This should open a panel like this:

For each part confirm you have the correct extruder selected. Also make sure the parts are listed in the correct order, using the up and down arrow buttons if you need to move the parts. Again, higher parts in the list take precedence when there is overlap.

Next double-check you've selected the right type of filament (eg, PLA or PET) and then slice. When you go and preview, you should be able to see the object in the various colors (and if you scroll through object layers you can confirm what interior parts will be which colors. For example, this is what the full sliced object looks like:

And this is what the an area with multiple colors in a single layer looks like when adjusting to show the cross-section.

(Side-note: I love using the gyroid infill for many reasons, but it really does look a little like brains in the cross-section of heads --- and I learned the hard way that it thus looks way to creepy when used in translucent colors for figurine toys.)

Printing

Finally, make sure you load the filaments into the correct extruder order for the printer. I've sometimes found it helps to even include the color order in the gcode filename so I don't forget.

Then run your print. I've unfortunately had bad luck getting the MMU2 to successfully complete prints.

The more tool changes (ie, switching between colors on a layer), the more chances of failure. For example, my first attempt to print this ornament had the red fail to load for the top layers in a way that wasn't detected by the filament sensors. I've also found poorly wound filament to also be more likely to jam and result in failures with the MMU2 than without it. I've also had adhesion failures with the purge tower cause failed prints, to make sure its first layer is laid down well and restart the print if it doesn't look like this happened.

But here is a successful print (60% scale in PLA, Fillamentum Extrafill Vertigo Grey, Fillamentum Pearl Ruby Red, and Hatchbox White):

Other notes

Some other design hints from my experimenting with designing for multi-color printing:

Tool changes (ie, switching colors within a layer) takes LOTS of time with the MMU2. Because of this, printing a bunch of objects takes not much more time than printing one. If printing an ornament, print a few and give the others away!
Optimize for layer changes (ie, design to minimize the number of colors within a layer). Think about object internals as well, and review in Preview to look for ways to reduce the number of colors in an object, such as by shrinking how much of an part is inside another part, or by reordering the parts.
I've anecdotally seen poorer adhesion between different colors/parts within a layer. It looks like Slic3r puts perimeters around each of the parts so with some design approaches you are relying purely on how well the perimeter layers adhere. As such, having parts recessed some into other parts (such as by having the red and black details in the snowman take precedent over the white) can result in stronger structures.

Final Result

Here is the end-result of the snowman ornament hanging in our tree!

Coming next...

Soon to come: a blog post on how to use this technique to design and print a multi-color puzzle.

Cherry wood bookcase

2018-12-31T20:41:00-05:00

Project finished a few weeks ago: scratch-made cherry wood bookcase for our bedroom. We were getting tired of the piles of books accumulating on the floor from while to the older child after the earlier one had gone to sleep, plus this also gives us a place for the too-read-next and partially finished books that otherwise accumulate on our nightstands. Plus you can never have too many bookcases. I've now made four bookcases in the house from scratch, but the previous three were painted. This is now one of the nicer pieces of furniture we own. :-)

If I'd planned further in advance I could have better hidden the nails used to hold on the top-molding, but a little brown sharpie causes them to disappear.

Detail of top molding

I got most of the wood from [Anderson & McQuaid])http://www.andersonmcquaid.com/) in Cambridge, and the backings are 1/4" birch plywood from Rockler.

Prior to staining and finishing

I've been using Kreg Jigs to do most of the construction. Ana White's blog got me started with using the Kreg Jig to make bookcases and other furniture, which makes assembly of sturdy structures a breeze. (This helping tower I built previously used a similar construction technique).

I stained the bookcase in OCS-116 Harvest Cherry from Woodwright Finish and ordered it directly from them. The theory here is that if I order some Amish furniture down the road, I should be able to get it in this same finish. I might have gone a little lighter in color in retrospect to show the grain better, but I'm happy with how it turned out. I put on three top-coat layers of General Finishes Oil-baed Gel Topcoat which I rubbed on with a cloth, working outside of course.

It took over a month for everything to outgas enough for us to be willing to move the bookcase up to our bedroom, but now it is happily full of books.

Six years since World IPv6 Launch: entering the majority phases

2018-06-06T12:04:00-04:00

See my work blog post:

Six years since World IPv6 Launch: entering the majority phases

Excerpt:

As reported in an ISOC report last year, IPv6 adoption is now solidly in the "early majority phase" of the technology adoption life cycle by many metrics (well past "innovators" and "early adopters"), with progress beyond that in some areas. Akamai continues to see solid growth in IPv6, including significant traffic peaks during large events, and we continue to find cases where IPv6 outperforms IPv4.

Driven by the growth of Internet end-users outpacing the supply of IPv4 addresses needed to number their devices, and a desire to plan for a future IPv6-based Internet, many networks around the globe are shifting to provisioning their end-users with IPv6 connectivity. Currently, IPv6 deployment in 7 of the top ten global economies has reached a point where Akamai typically sees over 18% of requests to IPv6-enabled sites using IPv6, with India (56%), U.S. (46%), Germany (35%), Japan (28%), and Brazil (27%) being well beyond this. We have come a long ways since World IPv6 Launch when we reported seeing only around 1% of U.S. end-users using IPv6. With around 80% of U.S. mobile end-users accessing IPv6-enabled Akamai sites over IPv6 (and 90% from some carriers including Reliance Jio in India), the adoption of IPv6 in the mobile space is well through the "late majority" phase.

Over the past year IPv6 has been enabled on many of the largest online streaming events on the Internet, as delivered by Akamai, such as numerous major sporting events including the VIVO IPL and some major broadcasters of the Olympics. During these we've often seen between roughly a quarter (¼) and two-thirds (⅔) of the global event traffic delivered over IPv6, depending on the audience distribution. We often see a considerably higher average throughput to IPv6 end-users than to IPv4 end-users during dual-stacked events (although some of this may be due to sample bias). We've even seen multiple recent cases where massive IPv4-only events caused problems for CGNAT middle-boxes but IPv6 traffic continued to flow unimpeded.

For dual-stacked hostnames we typically see higher average estimated throughput over IPv6 than over IPv4. Some of this may be due to IPv6-connected users being correlated with better connectivity, but over half of dual-stacked hostnames (weighted by daily bytes delivered) have IPv6 estimated throughput at least 50% faster than IPv4, and 90% of these hostnames have the IPv6 estimated throughput at least 10% faster than IPv4. Individual sites and network providers will vary (with IPv6 performing worse than IPv4 in some cases). However the overall trend aligns with past results from LinkedIn, Facebook, Akamai, and others that show IPv6 as delivering significantly better performance than IPv4, especially in some mobile networks. As reported in an ISOC report last year, IPv6 adoption is now solidly in the "early majority phase" of the technology adoption life cycle by many metrics (well past "innovators" and "early adopters"), with progress beyond that in some areas. Akamai continues to see solid growth in IPv6, including significant traffic peaks during large events, and we continue to find cases where IPv6 outperforms IPv4.

Driven by the growth of Internet end-users outpacing the supply of IPv4 addresses needed to number their devices, and a desire to plan for a future IPv6-based Internet, many networks around the globe are shifting to provisioning their end-users with IPv6 connectivity. Currently, IPv6 deployment in 7 of the top ten global economies has reached a point where Akamai typically sees over 18% of requests to IPv6-enabled sites using IPv6, with India (56%), U.S. (46%), Germany (35%), Japan (28%), and Brazil (27%) being well beyond this. We have come a long ways since World IPv6 Launch when we reported seeing only around 1% of U.S. end-users using IPv6. With around 80% of U.S. mobile end-users accessing IPv6-enabled Akamai sites over IPv6 (and 90% from some carriers including Reliance Jio in India), the adoption of IPv6 in the mobile space is well through the "late majority" phase.

Over the past year IPv6 has been enabled on many of the largest online streaming events on the Internet, as delivered by Akamai, such as numerous major sporting events including the VIVO IPL and some major broadcasters of the Olympics. During these we've often seen between roughly a quarter (¼) and two-thirds (⅔) of the global event traffic delivered over IPv6, depending on the audience distribution. We often see a considerably higher average throughput to IPv6 end-users than to IPv4 end-users during dual-stacked events (although some of this may be due to sample bias). We've even seen multiple recent cases where massive IPv4-only events caused problems for CGNAT middle-boxes but IPv6 traffic continued to flow unimpeded.

For dual-stacked hostnames we typically see higher average estimated throughput over IPv6 than over IPv4. Some of this may be due to IPv6-connected users being correlated with better connectivity, but over half of dual-stacked hostnames (weighted by daily bytes delivered) have IPv6 estimated throughput at least 50% faster than IPv4, and 90% of these hostnames have the IPv6 estimated throughput at least 10% faster than IPv4. Individual sites and network providers will vary (with IPv6 performing worse than IPv4 in some cases). However the overall trend aligns with past results from LinkedIn, Facebook, Akamai, and others that show IPv6 as delivering significantly better performance than IPv4, especially in some mobile networks.

[...more...]

Using a Logitech Touchpad as a light switch for Philips Hue light bulbs

2016-03-25T22:52:00-04:00

Our nursery lacked any light switches but did have some wall sconces. As part of redecorating the room prior to making it a nursery back in 2013, I replaced the old sconces with some newer ones from Rejuvenation and installed a set of three Philips Hue light bulbs. As Philips did not yet have a good remote lightswitch (which they've recently released), I had to hack together something to allow the bulbs to be controlled by a switch near the door.

My first attempt was to use an X10 wireless RF wall switch through an Insteon PowerLinc USB gateway to some custom python software, but this was quite flakey as I've had bad luck getting the PowerLinc to reliably send me signals for extended periods of time, plus this relies on X10 which is fairly unreliable. (This used the crazy control path of a button push resulting in X10-over-RF to X10-over-powerline to USB to an HTTP-over-ethernet request to the Hue hub to ZigBee RF to the bulbs.)

Even from the beginning, I also setup an experiment (which has worked quite well) of having a cron job run every few minutes and update the color temperature of the lights based on the time of day --- warm lighting in the evenings and cool "daylight" in the mornings. This has worked quite well both for the Hue bulbs I have in my office as well as those in the nursery (and in the Lightstrip I've recently added above some windows in our living room).

Looking around for better ways to control the light bulbs, I had trouble finding any sorts of reasonably priced RF switches with long battery life and a robust way to get control information to a Linux machine (my preferred gateway to my home automation).

What I did find was the "Logitech Wireless Touchpad with Multi-Touch Navigation" which was $25 at the time:

I have software running on my Linux machine that reads this as a USB Input device and then sends out commands to the bulbs via the Hue hub. This gave me two mouse buttons (left and right) which I repurposed for "on" and "off". I was able to use the gestures as additional input control. A single finger up/down gesture can be used to dim the lights. Two fingers left/right adjust the color temperature of the lights (or the hue in color mode). Finally, a "three finger up" gesture puts the lights into "dim red mode" which is particularly good when getting munchkins ready for bed or when stumbling into the nursery in the middle of the night.

Source code is available here on github, although it is meant as a starting point for someone wanting to play with this, and is NOT an easy-to-use off the shelf solution for someone not used to home automation hacking in python. It relies on isaackelly's python-hue library.

The Rebel Alliance S-Wing "Mamaroo" Training Pod

2016-03-19T18:09:00-04:00

Shortly after our son was born back in 2013, we got a used Mamaroo swing for him. The Mamaroo's design looked enough like something out of a sci-fi movie that it inspired me to decorate it like a Star Wars X-Wing, cutting out the decorations using our Cricut.

I designed the decals in Inkscape and then used Sure-Cuts-A-Lot to cut them out of repositionable adhesive-backed vinyl. For the Rebel Alliance logo, I used an SVG off of the wikipedia page.

I cut some additional copies of the artwork and used it to decorate my laptop.

The svg and scut2 files are available here for anyone who is interested in using them.

Building a Little Helper Tower

2014-09-27T23:17:00-04:00

With our little munchkin now being very curious whenever we're working in the kitchen, I decided to build him a Helping Tower based on plans from Ana White. I haven't done many woodworking projects in awhile, and certainly none of a larger size where finish quality mattered. The end result came out quite nicely and is being enjoyed so far:

Given the amount of time that would be involved in making it, I decided to use maple which also better matches the other furniture we have in our kitchen. I ended up making a bunch of modifications to the plans:

Due to the stock available, the corner posts ended up being 2x2's and other dimensions were adjusted accordingly. I had to make adjustments to a few other pieces as well based on the stock that was available at Anderson McQuaid.
Based on someone else's idea, I also put in slats to allow the adjustable platform to be moved up to be a counter-height high chair. I will need to put in safety straps before using it this way, however.
I routed the top to give it a more finished look. I also made more decorative feet to match the feet I added to our kitchen nook table a few years back.
The anti-tip feed on the side facing the counter can fold inwards to allow it to get closet in to the island.

I used a Kreg Jig for the first time to create the pocket holes and it worked very nicely and made assembly fairly straight-forward. Clamping in-place prior to screwing in turned out to be critical. The end result is quite sturdy.

More photos are here.

The tower is finished with Zar's Honey Maple and then three coats of General Finishing's Toy Maker's Finish.

Close up of the folding feet on one side:

I put plugs into the upper pocket holes to make them look better:

Making a board book inspired by the Hobbit

2014-06-08T16:51:00-04:00

(For pictures of all of the pages, click here.)

Shortly after my son was born I found that I was spending large amounts of time singing tunes and reading poems out of "board books" for children. For the uninitiated, a board book is a book for babies printed on thick paperboard. This makes it easy for little fingers to turn, but also hard for them to tear. In addition to reading many old classics (such as Goodnight Moon and Sandra Boynton's The Going-To-Bed Book), I also looked for other types of "classics" to introduce. I ordered a copy of Poems from The Hobbit, a wonderful collection of Tolkein's poems from The Hobbit. However, its 3" square form factor with thin pages made it not the best book for a small baby. While local stores such as Magpie Kids had board books like the Cozy Classics version of Pride and Prejudice, they had little to no sci-fi or fantasy for babies. This inspired me to make a board book of Bilbo's Old Walking Song ("Roads Go Ever On") out of the Hobbit.

After working on it off-and-on for a few months, the finished copy finally arrived in the mail:

Each page has a few lines from the poem over an (often heavily touched-up) photo, generally of a road going somewhere. All of the photos are from my various travels and are of places including from Maine, Portugal, Croatia, Washington State, Hawaii, Russia, New Hampshire, and finally near my own childhood home in California. I'd considered doing some other form of artwork such as watercolor, but this seemed like a good use of photos that I've taken over the years and also means that I can tell a story with each one as my son gets older. I edited some of the photos to look a little more like illustrations, but I also didn't want to go over-the-top here. The cover is actually a composite of three separate photos. The full set of images on each of the pages is here if you'd like to see the full book. Here is a sample page spread:

I used Inkscape to do composition of the pages, gimp to do photo editing, the Ringerbearer font, and then exported them to pdf files for printing. I did change one word in the poem ("horror" to "darkness") as this seemed a little more appropriate for a young child's book. Pint Size Productions did the printing and did a great job. If you would like to print your own copy for private and non-commercial use, you can download this zip file of the PDFs and upload it to Pint Size Productions.

Other Pages

IPv6 by the numbers: strong growth in second half of 2013

2014-03-04T08:13:00-05:00

See my new work blog post. Excerpt:

No longer is IPv6 "just around the corner". It's here. In the half-year following when I last wrote about our measurements of IPv6 adoption, many of the metrics we were tracking have doubled. This is in large part due to increased adoption of IPv6 by residential broadband networks in the U.S.A. and Germany. As of December 2013, we were serving over 20 billion IPv6 requests per day, double the 10 billion per day delivered just six months prior.

[...more...]

Launching Forward with IPv6

2012-05-31T12:04:00-04:00

See my work blog post:

Launching Forward with IPv6

Excerpt...

With the era of freely available IPv4 addresses nearing its end, I'm pleased to see that 2012 appears to be the year when the IPv6 Internet will finally reach maturity and launch into wide-scale commercial use. For over a decade, the groundwork for the migration to version 6 of the Internet Protocol (IPv6) has been built, with changes to operating systems, client and server software, routers, and Internet backbone networks. To-date, however, the availability of IPv6 content and end-users has remained slim with few Web sites being available over IPv6 and with just over 0.5% of global Internet users having IPv6 connectivity that their machines will elect to use.

[...more...]

Norse plaque inspired by an old fireback

2012-05-14T00:25:00-04:00

Ever since my early childhood I have fond memories of going to Camp Happy, my great aunt and uncles summer home up in Door County, Wisconsin. The home was supposedly built as a replica of a hunting lodge from the Old World. Installed above the fireplace was a wonderful bronze relief with a Norse design. A viking ship tosses on a stormy sea below shields likely representing Thor and Odin, and two serpents form the border. I have many wonderful memories of family up at Camp Happy; watching the sunset over the great lakes, eating out on the screened-in porch, walking through the woods out to High Bluff, but especially relaxing and talking into the night in the fire-lit living room.

For years I've had a desire to make a replica of the bronze plaque above the fireplace mantle, so last year I started on the project. While I wanted to get an end result that was very close in style, I wasn't aiming to end up with an exact copy. After about six months of working on-and-off on the project in some spare time over the weekends, I was successful. Here's the final version installed next to my parents' fireplace in California:

The completed plaque installed on my parents' fireplace.

To give some context, below is the original plaque installed in Camp Happy's fireplace mantle. While I'm not sure of its origin story, one possibility is that it is a bronze casting of cast iron fireplace fireback.

The process I followed to create this...

Backing board for setup to pour wax from a double-boiler. Packing tape was used for the side-walls of the mold.

I started off by printing a photo of the original scaled to the desired size and overlain with a grid. From this, I took a piece of MDF and cut it to the right shape to make the backing. While I used 1/2" MDF, this ended up resulting in a much thicker casting in the end. I would have been better off using a much thinner backing board (such as 1/8").

Part way through pouring the wax. A digital thermometer was used to control the temperature of the double-boiler.

I leveled the backing board on a table, put nails into it to gauge height, and used sides of packing tape to make a mold. Donning appropriate protective gear (full face mask, respirator, long-sleeves, and working outside), I used a in improvised double-boiler to melt "Victory Brown" sculpture wax and then poured it onto the backing board to the desired depth. (If you try this, make sure to take appropriate safety precautions!)

Once it cooled, I cleaned up the wax surface and edges using various carving tools so that I would have a good blank to start from.

The wax blank after some cleanup and smoothing to make it flat.

Next, I used a loop tool to rough-in the design. In retrospect, I ended up carving deeper than I probably intended to, resulting a deeper relief. If I was to do this again I might start off carving more shallowly.

Roughing-in the design, primarily using ribbon tools.

I worked from the upper left corner down to the bottom right corner. I was primarily using wax carving tools, along with a metal loop tool. I also would sometimes use a piece of canvas or nylons to smooth our the wax (almost acting like sandpaper).

In-progress, but no cleanup yet. Working on the details from the upper-left down to the lower-right.

Below is the initial completed version before I refined it... One thing I changed was the symbol in the upper right shield. It seemed to be impossible to find a Norse symbol for Odin or the sun that hasn't been subverted so I finally just made something up that fit the space and still looked a little like a sun.

Initial version before refinements.

I took it over to the foundry and got a number of suggestions for things I needed to do differently before casting it. In particular, I needed to smooth the edges so the sand wouldn't catch, make sure that everything had enough draft (diagonal slope) and no undercuts, and then coat it with a hard lacquer so that the sand wouldn't stick to the wax. I also used the opportunity to make some improvements suggested by my mom: improving the leading edge of the sail and changing the position of the oars to give them a more definitive line.

The final wax, before being laquered and cast. Refinements included changing the oar location, the sail shape, and ensuring that there was a proper angle everywhere to get a good draft.

Close-up of the final wax.

With the wax prepared, I took it over to Mystic Valley Foundry in Somerville and had them make two aluminum sand castings. (Given the thickness due to the backing board, bronze would have been way too heavy.)

The aluminum sand castings, just back from Mystic Valley Foundry

The castings came back in about two weeks. Each of them had its own natural imperfections from the casting process. The next step was to take one of them and clean it up. After a day with a dremel tool and hand files, I had the piece cleaned up nicely.

One of the castings after being cleaned up with a dremel tool and hand files.

To prepare for finishing, I primed the plaque with an aluminum spray primer so that the top coat would bond better.

After priming with self-etching aluminum primer.

I then used an airbrush to apply Modern Masters reactive bronze paint. (I'd experimented with some other types of coatings to get a bronze finish, and I'd also tried applying this with a brush, but neither worked well enough.) As with all of the other steps (spray painting, using the dremel tool, etc), having adequate face protection is key. The 3M full-facepiece respirators work well as they don't fog up like separate googles and respirators do.

The result of the bronze spray painting turned nicely, and might not have been a bad place to stop, but it looks a little too "new".

After airbrushing with Modern Masters bronze paint.

I then applied the a second coat of paint and while it was still wet applied the Modern Masters blue patina aging solution. After it was on for a bit, I wiped it off to stop the reaction.

It turns out to be very fussy to get the right look out of the patina. I tried doing some touch-up, but it seemed like each time I tried I made things a little worse. I finally decided to stop. One lesson I learned a little too late is that the blue patina aging solution leaves some white copper salts behind. After the paint has dried you can remove them with a damp towl for a better end result.

After applying patina

To install the plaque, I mounted some brackets in the back and used those to hang it on the fireplace. I'm quite pleased with the finished result. We also setup a spotlight with an MR16 GU10 LED bulb for illumination at light.

Now that it's done I'll need to figure out what project to work on next --- after we finish our kitchen remodel. At some point I should also finish up the second one, as well as see how well I can make a resin casting (which might be much cheaper and lighter than the aluminum casting, albeit not as nifty).

Me with the finished plaque in-front of my parents' fireplace.

Experimenting with Cricut-assisted leather tooling

2011-05-15T16:07:00-04:00

Recently, our friends Ariel and Andy inspired me to try using the Cricut (a 2D cutting plotter that can be controlled with third-party software called Sure-Cuts-A-Lot) to assist in leather-tooling and provided some leather scraps to try with. (I'd tried with the wrong type of leader before without useful effect. Ariel suggested that it is necessary to use tooling leather.)

Gears design, cut on the cricut and tooled.

Of the various leather scraps I tried, it appears necessary to use reasonably thin "tooling leather" with the deep-cut blade holder. The normal blade holder doesn't provide enough clearance (this seems to be one major difference between them). Setting the blade to a depth of around 4-5 seems to work fine (I haven't extensively tried other depths). In addition to using the deep-cut blade, I also tried using a modified bit that I made (from sawing down a nail of the same diameter of the Cricut blades) and it seems to score the leather somewhat in a way that can come out better with a wash.

Fleur-de-lis pattern before coloring.

I tried two patterns to experiment with: for the first one I used a fleur-de-lis pattern built into the Cricut. For the second pattern, I used a Steampunk-style gears design that I put together in Inkscape. For the very small patterns, just holding the leather to the Cricut's cutting mat worked ok. For the larger pattern, the leather wouldn't hold effectively. After trying a few glues (rubber cement, superglue) which didn't work, I tried stapling the leather to a worn cutting mat. That seems to actually work reasonably well, although it's important to ensure the staple doesn't get caught in the Cricut's mechanism. The biggest problem that I encountered was that the Cricut appears to lose registration for complex leather patterns, possibly due to the thickness of the leather interfering with the in-and-out rolling distance movement assumptions, or due to the mat shifting due to the leather scrap not having uniform thickness. I will need to experiment a few more times, although I'll need to get more large scraps first.

Following cutting the designs into the leather, I used a mixture of tools to work the leather after spraying it with water. In addition to the basic Tandy leather tools, I also found that jewelry screwdrivers worked extremely well for tooling smaller parts of the design. As this is my first attempt (and as the Cricut mat slipped a few times with the gears design), some parts didn't work out as well as I'd hoped.

Following the cutting and the tooling, I then tried to color them. For the fluer-de-lis pattern, I used a mixture of acrylic paint (Jo Sonja Burnt Umber) with antiquing medium and then wiped it off immediately. I really like the result:

Fleur-de-lis after coloring. The pattern in the upper-left wasn't actually cut, but was embossed into the leather with my modified "nail bit" on the Cricut.

For the gears design, I then painted it using acrylic paints, also using some antiquing medium (and rubbing some of the darker color off). Note that some of the gears pattern is off as the Cricut slipped and cut some things in the wrong places.

I'm going to have to play with making a few more other things with this technique. I will need to find ways to minimize the impact of the Cricut slipping on the material. In general, I think the more organic/natural/older designs may work better than something industrial such as gears. At some point I'll have to try making some larger pieces as well, although I have a few other projects that I need to work more on first.

Note: We seem to be using our Cricut for plenty of crafty projects, including many wedding projects. I've recommended the Cricut that we have to a few people, but given that they are now suing the makers of the Sure-Cuts-A-Lot software that makes it useful to us, I can no longer recommend the Cricut. It looks like there are a number of new entrants to this market since we got ours that are hopefully less litigious against people who make their devices actually useful, so we may explore getting one of those at some point.

Our wedding software

2011-04-30T19:40:00-04:00

When Ksusha and I got married, I decided to use it as an opportunity to write some software to manage all aspects of our wedding, from tracking guests and RSVPs to generating place cards for printing. Given that half of the fun of wedding preparation (for us at least) was DIY projects, what better than to mix in a DIY software project?

After requests from a few friends, I finally got around to packaging up and posting this software as a starting-point for others (and you can download it here). I'm releasing it under the MIT License, meaning you can pretty much do whatever you want with it as long as you don't sue me, even if some bug somehow ruins your perfect day.

The djWed package is personal web-based wedding management software written in the Python Django web framework. This is intended entirely as a starting point of code to develop from, and is not intended to be something that works out-of-the-box. Rather, it is software that you can hack to do what you want.

When writing software for your own wedding, one initially encounters a pleasant design constraint that is rarely encountered in software engineering: code designed to be used exactly once. In fact, I intend to get married exactly once (now complete) and thus never use djWed ever again.

This meant that I took many short-cuts that I might not have otherwise taken, but which made sense for expediency. For example, generating mass emails or PDF/PS output from SVG templates is fairly awkward. However, it didn't make sense for me to develop an elegant user experience for functions that would be executed exactly once by just one person. Similarly, I hard-coded in that I'd have two wedding and reception venues ("MA" and "CA") as that
simplified some code. I was also much worse about documenting the code or developing regression or unit tests than I would have been if I was developing production software.

However, now that our wedding is finished, some friends have asked to use this code. As such, these friends may need to live with the consequences of these decisions if they wish to start from this code base. They will encounter many bugs, hacks, quirks, and perhaps even a few squirrels. If you are one of these people, you are welcome to do the same as I did and adapt the code for your single use, or you can do cleanup to generalize this software so that it could more readily be used by others. If you do clean it up and post it, let me know so I can link to it from here.

Some of the features included are:

Track guests and invitees (where each invitee may consist of multiple guests).
Each invitee has an "invite code" that they can use to login.
Invitees can update their contact information and leave comments.
Invitees can RSVP and select meals and other options.
Send mass emails to invitees with form-filling.
Web-based reports on RSVPs, meal selections, and more.
Spreadsheet exports of reports.
Gift and thank-you note tracking.
Generate mailing labels for invitees.
Table assignment tracking (but no automated table-assignments).
Multiple venue support (currently hard-coded at two venues in some places).
Integrated photo gallery using a modified version of Photologue that lets guests upload their photos.
Web-based editing of informational pages (HTML "page snippets").
Initial import of guest information from a CSV spreadsheet.
Generation of printed materials from SVG templates, including:
- Save-the-date cards with RSVP invite codes.
- Personalized place cards (even including food selection)
- Personalized RSVP cards
- Personalized envelopes with artwork

Now that this is done, perhaps I'll never need to look at this code again, although I have already reused some snippets from it for other projects.

At some point in the future, I may post some examples of printed materials generated, as well as some of the artwork we created. (We have plenty of wedding DIY projects that I should go back and write-up here.)

Note that subsequently Chris Shabsin and Dana Scott forked this software and used it for their wedding, as has at least one other person.

Paskha mold: version 1.0

2011-04-30T15:08:00-04:00

Early Spring in our house is a time for cooking and crafts projects as we celebrate both Passover and (Russian) Easter. Both of these holidays are great excuses to cook lots of tasty recipes that we learned from our parents and grandparents, and then to enjoy this food surrounded by friends and family. This year was no exception, and was also the first year in over a decade when we were blessed to have my parents in-town for both holidays.

Back in Russia, one of Ksusha's family traditions was to make Paskha and Kulich, and she has been making them here for Russian Easter for as long as I remember. Both are very tasty! Pascha is made from tvarog (similar to farmers cheese) with various other items such as brandied raisins mixed in. Kulich is a cake similar to panettone (and even tastier from my biased viewpoint, causing me to pretend that Passover has ended a few days earlier than it actually ends). To make the Paskha, Ksusha makes her own curd through a process involving whole milk, a bit of sour cream, a slow cooker, and then draining out the whey through cheese cloth. To make the final Paskha, the curd (and other flavorings) is pressed into a specially-made box lined with cheese-cloth.

Alas, Ksusha's family Paskha molds ("пасочница" or pasochnitsa) are back in Russia. (A google image search shows some examples of some.) As such, I've been meaning to make one for the past few years, so this year was my first try.

I used a mixture of 1/2" and 1/4" white ash from Rockler (both 5" wide), as that's what they had available. I played with a few different design ideas initially, and then I made templates in qcad and then went down to the shop to make the parts.

In the end, I had two of the sides (with the routed grooves) be 1/2" thick and the other two sides (with the tabs) were 1/4" thick. I then used a dremel tool to carve designs into the wood. One of the traditional designs is an Eastern Orthodox cross, and the other "XB" stands for "Christ has Risen". The resulting side panels:

The various pieces fit together by putting the tabs through the slots and then using the wood pins to hold everything together. (One of the tabs broke when drilling a hole for a pin, but I was able to improvise.) When assembled the box looks like:

After pressing the curd in the mold overnight, we removed it from the mold and while the Paskha was a little soft, it held its shape and the design were clearly visible.

In retrospect, I would have been better off making everything from 1/2" thick pieces as the moisture caused the 1/4" parts to warp (although I was able to flatten them back out by pressing them while wet under a heavy pot overnight). I would also have been better off having the carvings higher on the pieces so that the press didn't push the curd too close to the design. We would also have been better off using a slightly more firm version of the curd, as this one was fairly liquid and required too many layers of cheese cloth (impacting the visibility of the patterns).

Next year we'll have another Easter and I'll make another version.

The final result of this effort was this year's Paskha:

(Many thanks to my parents for taking some of the photos above while we were busy cooking and crafting away...)

Pumpkin Carving: Halloween 2008

2008-10-30T22:46:00-04:00

I just finished my first pumpkin for the year. I probably ended up with bits of pumpkin in my hair.

I just completed pumpkin number two. I'm not sure that I like it as much as the first one.

Since a few people asked how to carve pumpkins like this, I took some pictures this time. I primarily use potato peelers, ceramic ribbon/loop tools (the sort with blades), wood/wax carving tools, and a microplane grater.

(This post is recovered from the vaults.)

Power optimization

2007-12-08T22:54:00-05:00

One of my household projects the past few weeks has been to profile my home power usage with the goal of seeing where I can reduce it with small investments of time and money. The $20 Kill-a-Watt tool has been invaluable in this exercise. So far, I've managed to reduce my estimated power usage by around 10% ($190/year) with another 3% of quick-wins yet to do as well as some future projects that could yield another 20-30% savings. For the full details, here's the spreadsheet that I've been using.

Here are some interesting facts that I've learned along the way:

Massachusetts has some of the most expensive electricity in the contintental US.
In my house, just the three computers (office desktop, router, mediacenter pc) and three TiVos that were turned on 24/7 account for 425W 24x7 (and that's not counting laptops, and even with the monitors and 170W Windows box turned off). This is around half my electric bill! I suspect that many of you are in similar situations... Over time, I think that I'll start to replace some of these with low-powered Via Eden or AMD LX800 boxes with solid-state-drives. Even when buying new desktops, low-power CPUs will likely pay for themselves in MA.
Computer processors use up lots of power, but so do hard drives. Across my desktop computers and TiVos, I have 12 hard drives in the house. That's around 100W just by themselves! Configuring drives to spin-down when not in use can save 4-10W per drive. Taking the second CPU out of my router was also a quick-win.
For TiVos and other similar devices, the power costs can be a significant portion of the cost-of-ownership. For example, my Series3 TiVo draws 40W which costs about $5.50/month or $200 over three years. This is about half the cost of the service subscription!
Those power bricks do add up (at 500mW-2W each), but a few individual turned-off-but-plugged-in devices may dwarf those at 10-15W each. Finding those and putting together mitigation strategies for those few devices are faster wins than dealing with power bricks. For example, my subwoofers use 10-20W even when in "power-saving mode". Two Yamaha stereos that I have from \~1993 use 16W even when turned off. By contrast, my newer Yamaha receiver in my home theater only uses 1W when turned off.
There are some dimmable compact florescents that play nicely with X10/SmartHome dimmer switches. In particular, the Philips Marathon Alto bulbs seem to work ok so far, although it takes them a little time to warm up.
Plasma TVs use *lots* of power when they're turned on. My 42" 480p display uses around 220W when turned on, but 1080p plasmas are rated to use 2-3x as much. I guess this means that I should get an LCD when I eventually upgrade to 1080p.

These are the highlights of the audit along with example actions taken (or planned) to save power:

	Approx Pct	Monthly Kwh	Average (W)	Example Savings (W)	Actions to achieve savings
Home office computers	21%	176	240	14	Remove unused power bricks, configure drives to spin-down when not in-use, set Windows machine to go into standby when idle, eventually migrate apps off of desktop server to lower-powered machine
Lights (total)	15%	128	175	54	Replace selected lights with compact florescents
Home theater (total, with tivo+computer)	14%	116	159	19	Used trigger cable from receiver to power supply to shutoff subwoofer and DVD player when not in-use
Electric clothes dryer	11%	89	122		Planned: replace with energy-efficient gas dryer
Home router/server	10%	82	112	18	Removed a CPU, configured drives to spin themselves down when not in-use
Refrigerator	7%	64	87	33	Could save \~33W by replacing with more efficient fridge
Bedroom A/C (amortized)	6%	49	67
Air purifier	6%	47	64	50	Don't keep always-on
Kitchen lights	4%	36	49	18	Replaced two 90W R40 bulbs with dimmable 20W Philips Marathon CF bulbs
Halogen torchiere in office	3%	29	40	28	Planning to replace with compact florescent
TiVo Series3	3%	29	40
TiVo Series1 w/ 2 drives	2%	20	28	28	Retired since I don't really need three TiVos
TiVo Series2	2%	15	20

On a related note, is anyone interested in my old Hacked Series1 TiVo (with ethernet and extra storage) before I try to sell it? I also have a bunch of incandescent R30 bulbs for ceiling canisters that are left-over from my old house. I also have an old CRT TV stand from Ikea that I'd like to get rid of to make room for a breakfast nook table.

The Earth, modeled in chocolate

2006-02-13T22:46:00-05:00

Ksusha introduced me to Ferrero Rocher chocolates a few days ago. My first reaction after biting into it was "this would make a great model of the Earth!" (Well, maybe that was my second reaction. My first reaction was "tasty!")

Thus, it was necessary to slice one in half so that I could take a photo for labeling:

Alas, the crust-to-mantle thickness ratio isn't quite right, and there isn't a clear differentiation between the upper and lower mantle. Now to go eat what remains of my Earth model. Hopefully it's not a voodoo Earth...

Plush Sawzall

2004-09-28T23:01:00-04:00

The plush sawzall that I made in the spring for the child of two friends of mine has finally been presented. The parents were greatly amused. The child wasn't quite sure how to operate it. Some people on the dinner trip suggested that he needed plush safety goggles first.

Either way, this means I can finally post pictures... :-)

And here are some more pictures of Kids With Plush Power Tools:

A few people have asked for the pattern. At some point I may get around to scanning it in and posting it, just to inspire more chaos in the world.