On 15/06/2011 09:29, Simon Kelley wrote: > Jan's valgrind suggestion is a good one, as is using netstat to look at > the size of socket receive queues.
Although if he can show that a rack reboot works with 200 leases and not in an identical setup with 7,000 leases then it might point to the software? I wonder how one might build a test case for this? How many machines in your rack? Is it possible to setup linux with 10s to 100s of network card aliases and simultaneously run a bunch of dhcp clients against them? What does this rack of machines do once they get their DHCP allocation? If they subsequently hammer the network then it would seem very likely that you might max out some switches and see packet loss? Curious problem - I'm mostly interested in how to reproduce without having a rack of equipment to keep pulling the plug on? Ed W
