On Fri, Jun 7, 2013 at 4:08 PM, Loïc BLOT <loic.b...@unix-experience.fr>wrote:
> Hello misc, > I have serious problems with my Dell R320 servers (6 servers but i use > Intel em i350 cards). Before i was under OpenBSD 5.3 but problem also > occurs on 5.2 (today it occurs). > Sometimes (too often) the totally freeze. Nothing responds (but ICMP > ping works...) > > Since 1 month i search to resolve this problem but i don't found where > is the problem. I have tried all you mentioned about DDB without > success. > > * Four systems handle 700 clients, and two have squid 3.2.5 compiled > with small options (pf-transparent, ssl), and use also named, dhcpd and > openOSPFd > * The last two system uses only PF+openBGPd+openOSPFd. > > Before, i think it was related to LACP agregates, but two of the squid > systems doesn't use agregates. > I also think it's related to CARP, but only 4 four uses CARP... > > On each system, only 10% of bw is used (10-50Mbps), RAM is correct (each > system has too many ram, 16Gbit and use only 8G for squid servers and > 200M for other servers). > > One month ago i have this network problem each 30 minutes on one server. > I thinked about too many connections on my proxy. > In UNIX logic, a network connection is a file. Then i have increased > kern.maxfiles to 16K, openfiles-cur to 8k and openfiles-max to 16k. > > Since this moment, i haven't have crashed since today. Then i have > increased all to 36K. Is this the good way ? Is there anything else to > check ?? Must i set openfiles to infinity ?? > > At this time, here is the current open files on the main squid router on > the production: > kern.nfiles=4701. > > Thanks for advance. If you need more details please tell me. > Good start will be to provide 'netstat -m, netstat -s, vmstat -i, vmstat -s,dmesg,systat' outputs > > -- > Best regards, > Loïc BLOT, > UNIX systems, security and network expert > http://www.unix-experience.fr
