We just did a test with the new top-secret project I'm working on. We had 11865 users at peak, split between 2 servers: dual 1G P4, 1G RAM linux boxes running Redhat 7.2, Apache 1.3.26 and a custom apache module we wrote. They were doing 1 request per user every 5 seconds, so 2373 requests per second, or around 1200 requests per second per box. At that point - they choked, HARD. As in, stopped responding to pings, and everything. Woot. :)
However, they did not die - they just stopped responding to any input for a while, then recovered. Not bad!
[root@ online]# uptime
1:33pm up 9 days, 46 min, 4 users, load average: 0.23, 1.55, 5.12
I think I'm going to post some stuff around, I need to know if anyone has any suggestions for tuning a redhat 7.2 box to handle even more connections/requests before it dies. spot? dormando? Anyone?
Update: Check this out:
Aug 15 13:23:43 kernel: eth0: card reports no resources.
Aug 15 13:23:43 kernel: eth1: card reports no resources.
Aug 15 13:23:47 last message repeated 6 times
Aug 15 13:23:48 kernel: eth0: card reports no resources.
Aug 15 13:23:48 kernel: eth1: card reports no resources.
Aug 15 13:23:50 last message repeated 5 times