A few fellow bloggers have been writing about Callisto,
the simultaneous release of 10 Eclipse projects (including a shiny new version of Eclipse itself). We webmasters have been gearing up for the event as well. It will be a major event that will draw millions of people to our site all at once, and we need to make sure the infrastructure can handle it. Here's an outline of what we're doing:
- Mirrors. We'll start looking for new mirrors -- big ones -- and we'll allow our mirror sites to get exclusive access to the Callisto files for at least 24 hours. This will also allow some of our mirrors, such as iBiblio
, to set up some permaseed torrents
- Torrents. As outlined above, we will kindly ask iBiblio to provide a torrents with permaseed for the Callisto files. This should help ease the bandwidth requirements for everyone.
- Bandwidth. We're expecting everybody and their uncle Bob to want to get Callisto files the minute they're available. We're bumping up bandwidth from our usual 60Mbps to 300Mbps for the duration of the rush. Matt and I asked to for either a Porsche
each, or more bandwidth. They approved the bandwidth.
- Servers. We had some big iron idling around, so we put it to good use in the cluster
. This one box, an IBM P5 550
, is a quad-CPU PPC64 box with 16GB of RAM. We expect it to be able to handle about 40% of all the website hits all by itself.
- Stability. We've already started locking down parts of our server configs until Callisto is over. No quick-and-dirty Apache reconfigs for a while.
- Efficiency. We'll be examining a bunch of PHP code to make sure we don't have any resource hogs. We're expecting anywhere from 3 million to 6 million daily page views during the first days of Callisto, and if some of those pages misbehave, we could have problems.
It's going to be big. We're going to get a bizillion hits. But darn it, this time, we'll be [better] prepared.