Re: TCP Listeners: Preferences?

Posted in 2007

Art S. Kagel — Thu, 05 Apr 2007 19:31:48 -0400 — source: Usenet: comp.databases.informix

Tom Girsch wrote: > Keith Simmons wrote: > >> Tom >> >> TCP connections should always be run on NET VPs and SHM on CPU. There >> are many threads on this from about 2002/3 explaining exactly why and >> how performance will plummet if you mix them up !! >> > I just want to verify that this still applies even when you're not > running SHM listeners at all. We'll be running ONLY TCP listeners. The rule still applies. The CPU VPs are can't block on the sockets and are too busy to poll effectively. Also the first few CPU VPs tend to take all the work on themselves so the other VPs will tend to be quiescent and will take time to awaken and swap back into memory if needed quickly when load ramps up. TCP listeners in CPU VP only works well with no SHM connections and a single CPU VP, since there's nothing to gain from the NET VP in that limited circumstance. Art S. Kagel

↪ replying to Art S. Kagel

Thomas J. Girsch — Fri, 06 Apr 2007 09:42:50 -0500 — source: Usenet: comp.databases.informix

Art S. Kagel wrote: > The rule still applies. The CPU VPs are can't block on the sockets and > are too busy to poll effectively. Also the first few CPU VPs tend to > take all the work on themselves so the other VPs will tend to be > quiescent and will take time to awaken and swap back into memory if > needed quickly when load ramps up. > > TCP listeners in CPU VP only works well with no SHM connections and a > single CPU VP, since there's nothing to gain from the NET VP in that > limited circumstance. Thanks, Art. That's what I needed.