On Tue, 22 Jun 2010, Lucas Zinato Carraro wrote: > Thanks for all answers, > > " As Wes points out, proxyd process churn is relatively expensive." > > Any metric to "sizing" a solution ? > ex: proxyd memory and cpu by connection > > I expect "20.000" simultaneous connections in frontends ( 3 or 5 machines ) > 10.000 using IMAPs and 10.000 using IMAP. > > An maximum attachment set to 10 Mb. We have 3 frontend servers here, each with 2 dual-core AMD 280 cpus and 4GB of RAM. Our peak usage was about 4500 proxyd processes (1500 per server). These are a mix of IMAP and IMAPS (I don't know the breakdown). During those peak times, the load average was around 0.6 to 0.8. Right now there are 800 proxyd processes consuming about 1GB of RAM. There is almost zero memory used by proxyd that is not shared memory. If I had to guess at the peak capacity of my frontend servers, I'd say we could handle about 2000 proxyd processes each. Today, you could buy an 8 core, 8GB server and handle double the number of processes. For your requirements, I'd buy 5 or 6 of those servers. Andy ---- Cyrus Home Page: http://cyrusimap.web.cmu.edu/ Cyrus Wiki/FAQ: http://cyrusimap.web.cmu.edu/twiki List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
