[Users] kernel: TCP: time wait bucket table overflow

Konstantin A. Lepikhov lakostis at altlinux.org
Mon Oct 2 15:42:29 EDT 2006


After updating to test018 OpenVZ patch my syslog full of those messages.
/proc/user_beancounters shows nothing unusual comparing with previous
kernel. Various docs says that this message indicates low memory situation
or simple DoS attack and consider increase net.ipv4.tcp_max_tw_buckets but
it didn't help. I'm found another sysctl - net.ipv4.tcp_max_tw_buckets_ub,
but it undocumentend in ovz docs and looking in code shows that it's
internal switch for ubc counters. So what I can do now and how ovz kernel
affects this? :) Okay, it's heavy loaded system with 2 Gigs of RAM/1.4 P4
CPU and 1x100Mbit e100 NIC. According nginx internal status there is not
so many connections:

Active connections: 1067                                                                                              
server accepts handled requests                                                                                       
 6494 6494 9525                                                                                                       
 Reading: 1 Writing: 6 Waiting: 1060

~# uptime          
23:36:22 up 2 days, 10:25,  1 user,  load average: 0.80, 2.06, 2.51

~# free
     	     total      used       free     shared    buffers     cached
Mem:       2072844   1825712     247132          0     128736     951484
-/+ buffers/cache:    745492    1327352
Swap:       995896      3780     992116

PS I can attach kernel buildconfig and dmesg if needed.

WBR et al.

More information about the Users mailing list