One other thought: If we do go for a higher concurrency level than eight, we should expect more time increasing various timeouts in our tests, at the least. In the runs I've done today on the hyperthreaded 16 core machine, I've seen three new problems caused, I expect, by the system struggling to keep up with demands from 32 concurrent test runs (e.g., RabbitMQ not starting up within the timeout, and needing 2.5 minutes before all 32 LXC containers have assigned IPs).
In general, whatever machines we get, we should expect to have to spend a bit more time tweaking/fixing tests. Gary -- Mailing list: https://launchpad.net/~yellow Post to : [email protected] Unsubscribe : https://launchpad.net/~yellow More help : https://help.launchpad.net/ListHelp
