Forum Discussion
Nimbostratustcp window sizes and auto scaling
Ok so we're testing a virtual server of two different types; standard and fastl4. On the back end is centos 6 and we're testing from ec2 instance by doing a wget to a bin file.
When the server type is set to standard, we see 3.5M/s. When the server type is set to fastl4 we see 40M/s. The server/client profiles are tcp optimized lan and wan. They have the default window size of 65535. When we increase this we do a little better performance. But no where near the 40M/s.
So in fastl4 are the linux servers auto scaling up to 4M windows and that's the perf increase? Does the f5 not have auto scaling abilities? I found an option in the tcp profile for congestion control -> scaling But that didn't help. Even still we're seeing 15ms latency from aws to us.
According to articles like this: http://bradhedlund.com/2008/12/19/h...nce-links/ At 64K and 15ms latency we should still be seeing 30M/s.
Am I missing something here?
1 Reply
cheezus_115419Opened support case. Support said it's expected behavior due to the full proxy of waiting on client side and collecting, then opening server side and sending. Did a tcp dump with support and confirmed nothing wrong there. Not sure what else to say.
