Forum Discussion
NimbostratusViprion congestion settings with MPLS
We have a Viprion F5 acting as Layer 7 Loadbalancer towards an Apache webserver. We observed download times from our remote locations of 1MB/s (connected with MPLS at 65MB/s, rtt=5ms). Downloading directly from the Apache, our remote sites reach 50MB/s download times. Local (LAN 125MB/s, rtt=2ms) we reach 22MB/s via loadbalancer and 77MB directly from Apache. Even with tcp receive window with auto (130kB), proxy window auto (200kB low, 260kB high) and send window auto (260kB), we did not reach reasonable download times. As workaround we implemented the following irule: if { [TCP::mss] < 1460 } { TCP::congestion bbr } elseif { [TCP::rtt] > 0 && [TCP::rtt] < 100 } { TCP::congestion highspeed } elseif { [IP::ttl] > 250 } { TCP::congestion highspeed } else { TCP::congestion bbr } }
With that, local and remote are now around 20-25MB/s
Is this normal or do we suffer from some other issues around the F5?
Why do we need those congestion settings whereas a standard Apache is in all cases better?
1 Reply
- Jeff_Granieri
Employee
Hi F5UserMaxMisinger ,
Can you advise what tcp profiles are applied on client/server side? As a basic starting point if not already done I would recommend tcp_progressive applied on the Client Side and tcp_lan_optimized on the server side. This assumes that your clients are located globally and will have different RTT, the tcp_progressive profile will autonegotiate window scaling on a per client connection basis. Are these typical file downloads ? If so enable Nagle on both sides of the F5 TCP profiles + ApacheServer, if this is not just file downloads make sure NAGLE is disabled on the F5 profiles as well as the Apache end. what is your baseline on expected performance?
