We have 2 integration servers sitting behind an F5 load balancer.
When the clients invoke webservices hoisted on the integration server directly, there is no latency. The round-trip is completed in sub-seconds.
However, when the traffic is routed through the F5 load balancer, we see latency as high as 9 seconds.
Logs reveal that the application hoisted on the IS is processing the request in sub-seconds. Hence the latency is not induced by the application.
Checking the network logs (TCP/IP packets capture) revealed following pattern:

The client sends request to F5.
F5 forwards the request to the IS box.
IS box acknowledges the packet.
IS box sits idle for couple of seconds (8-9 secs).
IS box sends a TCP/IP window resize request and immidiately IS box sends back the response.

Interstingly, we dont see this behaviour when the clients connect to IS directly.
Also F5 is a pass through and it does nothing but modify the IP in the packet to route it to one of the ISs.
All the a parties involved are within the enterprise. Nothing is exposed to external world.
Environment: Integration Server 6.5 on Solaris

Has anybody witnessed such a behaviour

• Prithvi