First, the important things: The conference bag was quite nice, and quite thoughtful. A carry on sized duffel bag. Mostly, this will be filled with t-shirts and non computer things for the trip back home. I got in late, so didnâ€™t get to as many sessions as Iâ€™d have liked, the two I did manage to get into however, were excellent.
Operational Swift (Not sure the official title, it was given by Ryan of Rackspace in the Lunch Hall) – All and all an excellent session. Some key takeaways:
- Get the design right the first time, or you will be destined to be chasing ghosts in the machine. It pays to bring in some outside expertise here.
- Use Swift for instance snapshots. Either that, or you will need to use Cinder or glance to tie into something massive.
- Network Segregation – keep your storage traffic segregated from your inter-instance traffic. Depending on how much traffic, you may also want to break out your management networks as well.
- Design for your IO patterns. 20-40 instances per host will mean lots of oddball IO patterns that could be difficult to get the full cache benefit from. Keep that in mind when designing swift, or plan for Cinder.
- At Petabyte and Petabyte plus sizes for Swift, one of the best things you can do is to break out the networking and save yourself the headache.
- At the end of the day you are still running physical infrastructure. Keep that in mind when designing for redundancy and such in the physical plant.
Quantum API Improvements Session – Also a good session, even if 80% of it was over my head (always a good sign). My notes on the session however are kind of crummy compared to what is here.
Hallway Track – Hung out with Tom McSweeny of EMC and Razor fame, as well as a few Rackers to discuss some goodies for the Wednesday Paul Sims session.