Hello!
We are launched new LVS directors:
2x HP DL360 G7 / 16 GB RAM / 2x Six-Core X5675 @ 3.07GHz / 2x Intel 82599EB
10G (bonding)
320 servers behind load balancer in ipip mode
3,5 Gbit/s Full Duplex Traffic (one-arm balancer)
60k conn/sec
Connections synchronized between two balancers. For this to work kernel
needs to be patched
diff -U0 -r a/net/netfilter/ipvs/ip_vs_sync.c
b/net/netfilter/ipvs/ip_vs_sync.c
--- a/net/netfilter/ipvs/ip_vs_sync.c 2011-08-03 23:43:28.000000000
+0400
+++ b/net/netfilter/ipvs/ip_vs_sync.c 2011-11-24 17:16:08.000000000
+0400
@@ -1473 +1473 @@
- schedule_timeout_interruptible(HZ);
+ schedule_timeout_interruptible(HZ/10);
Servers are directly connected through 1G NICs for connections
synchronization.
Output of ipvsadm -L --persistent-conn shows exact number of connections on
both load balancers in ActiveConn and InActConn state. But number of
PersistConn on Slave node are approx twice lower.
Do you have any ideas why it happens?
Sysctl parameters:
net.core.optmem_max = 16777216
net.core.rmem_max = 16777216
net.core.rmem_default = 16777216
net.core.wmem_max = 16777216
net.core.wmem_default = 16777216
net.ipv4.tcp_rmem = 4096 16777216 16777216
net.ipv4.tcp_wmem = 4096 16777216 16777216
netstat -s
Udp:
43168615 packets received
22 packets to unknown port received.
0 packet receive errors
153851 packets sent
RcvbufErrors: 0
SndbufErrors: 0
Best regards,
Aleksey
_______________________________________________
Please read the documentation before posting - it's available at:
http://www.linuxvirtualserver.org/
LinuxVirtualServer.org mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
or go to http://lists.graemef.net/mailman/listinfo/lvs-users
|