Run report
The results are still quite divergent in subsequent runs, seeing lots of timeouts.
Investigating.
I was having problems with timeouts too. Where are you seeing them? I was OK running in parallel with 8 threads against a local CSS and against pod-compat for ESS but had to drop to 1 thread for NSS (inrupt.net). We monitored the server and could see it struggled if I even increased to 2 threads. What behavioural characteristics of different servers have you discovered and had to work around? @bourgeoa said the wait for the ACL cache after setting an ACL was no longer necessary on NSS but I can see lots of other waits in the tests.
Yes but the wait for acl's are not all removed from the tests.
I just realised I was thinking about HTTP timeouts connecting to the server. That is one class of issue but I realise that this was probably an issue of the tests timing out. So presumably removing the ACL waits and the post-setup waits would help in this case. However, that may then hit the issues of overdriving the server and getting HTTP timeouts.