Monday, June 12, 2006

More bandwidth for Callisto

Callisto is approaching fast, and we received our new 100 Mbps network connections last week. To survive the Big Event, we'll be removing the 60 Mbps cap on our existing 100 Mbps and use these two 100 Mbps connections dedicated to downloads. The net sum of usable bandwidth should be about 280 Mbps, routed through three different Cisco 7200 routers for maximum performance/failover.

On the server side, the eclipse.org cluster will handle the usual web pages, CVS, e-mail and RSYNC using its beefed up connection, while downloads will be redirected to the two dedicated 100 Mbps connections, served by an IBM eSeries P5 550 server with 16 GB of RAM. We managed to get 3,500 simultaneous web connections to this IBM server alone, with plenty of room to spare. All tolled, we successfully tested our infrastructure to support 11,000 concurrent web connections, and if need be, we can balloon to 15,000 (our daily average usually floats between 400 and 800.)

All this new bandwidth isn't available to us -- yet. We still need to do some testing first. Sometime next week we'll host a "Mirror Site Appreciation Day" where, for the duration of a day, eclipse.org will attempt to host *all* its download traffic. The idea here is to load our systems with massive amounts of traffic to test our server performance and routing setup. Stay tuned...

0 Comments:

Post a Comment

<< Home