[lvs-users] Upgraded my Load Balancers, Performance Went to Sh*t

Eric Robinson eric.robinson at psmnv.com
Fri Sep 18 16:47:27 BST 2015


Formerly both of my load balancers were running RHEL 5.2 and everything has been awesome for years. I reformatted them and installed RHEL 6.3 (because that's what's on the rest of my servers) and now we're seeing application pauses and freezes. If we bypass the load balancer and point the client application directly  at one of the tomcat servers, it performs like a dream. If we re-point the application back at the load balancer, the pauses and freezes return.

Connectivity from the client to the load balancer is solid. Using ping -f (ping flooding) or tcping, we can send half a million packets with zero drops and no high latency. Likewise, connectivity from the load balancer to the tomcat realservers is just as solid. However, we have observed TCP retransmissions in the communication between the load balancer and the realservers, which may correspond to application freezes.

Aside from the upgrade to RHEL 6.3, we also added two additional NICs and enabled bonding (mode 1). The clients connect to the load balancer on bond1. The load balancer forwards the packets to the realservers on bond0.

Is anyone aware of issues with LVS and bonding? Or problems with LVS and RHEL 6.3?

--Eric Robinson





More information about the lvs-users mailing list