LCA2010: Day 2, Wellington, New Zealand
January 19, 2010
I already covered Tuesday’s highlight in another post, noting that Weta Digital used Ubuntu on their 35,000 core to render the digital effects and graphics for the recent blockbuster, Avatar.
I had lunch with Paul Gunn, the gentleman who gave the morning presentation, and he introduced me to several of his colleagues at the end of the day, retiring to The Malt House for a few local ales, and hearing much more about their impressive data center and heavy using of Ubuntu throughout their infrastructure.
The day started with the keynote from Gabiella Coleman, an anthropology researcher who studies the free and open source culture. She started by quoting Dickens, “These are the Best of Times and these are the Worst of Times…” She talked about her introduction to Linux, a college classmate raving about a Slackware Linux CD. As she started her university research, that same classmate introduced her to the Copy Left movement. Her speech was interesting, very academic. She read a lot of quotes, and provided a fairly comprehensive history of Free and Open Source software.
Next I attended Paul Gunn’s talk in the SysAdmin Track on Weta Digital’s effects rendering data center, here in Wellington. Very early in the talk, he noted that 90% of their artists run the Ubuntu Desktop, with the remaining 10% using Windows. Adobe Photoshop is one of the few remaining apps that still need Windows for. He went on to talk more about their server render farm, noting that all of their 35,000 cores are also running Ubuntu. In the talk, he called these “Ubuntu Servers”, which to me meant that they were installed from the Ubuntu Server seed. Actually, after having talked extensively with Paul and his colleagues, these nodes are actually imaged with the same Ubuntu Desktop image that the artists used. This consistency is critical for their operation — a single image across thousands of systems. The size of the data center is just incredible. Somewhere around 5,000 HP (and some IBM) blades, mostly 4x or 8x Intel processors, for a total of about 35,000 cores (or render slots, as they call them). They have 2 petabytes of network storage, carved into hundreds of multi-terabyte NFS mounts. 10 Gigabit networking through. 10 years ago (October 2000), their render farm consisted of 13 machines, each were dual core 733MHz systems, with 100mbps networking. As you can tell, they have grown quite a bit. The rest of the talk was about heat and energy efficiency — something that is quite important
Next, I attended a session in the Data Storage track on Sheepdog, a shared network storage solution for KVM. I’ve followed the discussions about Sheepdog on the qemu-devel@ and kvm@ mailing lists, so I’m fairly familiar with its promises. Admittedly, I haven’t used it yet. This talk was a technical discussion of the design, which I found quite interesting. It’s not a general filesystem, but instead designed specifically for QEMU. Volumes are divided into 4MB objects, each replicated to multiple nodes, for high availability. Their goal is to become the de facto standard for cloud storage. Ambitious. I’ll keep my eye on it.
Back in the SysAdmin track, I attended a session on RAID. It was a lengthy narrative about this system administrator’s quest to solve some strange behavior with his Linux software RAID. His talk walked through all the successive testing and debugging he did (perhaps interesting to those less expert in RAID administration). As it turned out, he used two Western Digital disks, one supporting some obscure error correction scheme and the other not. This minor difference was causing some annoying behavior.
I popped into the Launchpad track for James Westby’s talk on Ubuntu distributed development. The audience (and in particular, a few Debian developers) were most interested in bzr’s “send” command. Other audience members seemed to be impressed with bzr being able to handle branching from cvs, svn, and git repositories (with the appropriate plugin).
Next, I attended a short session on Linux Containers, by Sam Vilain. He noted the difference and situations where you might choose Xen/KVM virtualization (need full hardware emulation, different kernels), and where you might choose containers (need maximum performance). In truth, he noted, many situations would benefit from a combination of both. I tend to agree with this, and would like to see Ubuntu fully support both KVM and LXC in appropriate situations. He used “sudo apt-get install lxc” on Ubuntu in his examples and demonstration.
I attended a session in the Storage Track by Arjen Lentz on Multi-Master Replication with MMM that was completely over my head.
And I finished my day in the Storage Track with an excellent talk (and demo!) by Ben Balbo, on his roll-your-own-dropbox implementation. He gave a bit about the design of Dropbox, reasons why you might use Dropbox, and reasons why you might not (cost, privacy, bandwidth). Particularly here in New Zealand and Australia, latency to the US or Amazon can really be a problem. So he demonstrated his solution, which is built on top of git. I must say, it looked very impressive. There was no mention of Ubuntu One, though it was certainly in my mind.
Today’s contest is from the Cuba Quadrant. Here are my submissions: