fre 2010-02-19 klockan 12:54 -0800 skrev Andy Litzinger: > We run with stock kernels from CentOS/RHEL so I guess I meant in those the kernel and shell fd limits are way higher. Are you sure they are by default? It's easy to configure anyhow. > > On must systems the default is whatever the ulimit is set to when you > > run configure. > > Great, thanks. Is there any way to confirm this on a compiled squid, or is it best practice to define the value upon compilation? Unfortunately not. > I'm not sure I understand what you mean here. How/where does squid > get this value? And I suppose I should have said checking/increasing > the kernel file descriptor limits (/proc/sys/fs/file-max) and the > shell file descriptor limits (ulimit -n). ulimit is the primary limit. file-max is related, but the system global limit. Should be bigger than ulimit obviously. Squid reports it's current limit in cache.log at startup. > I understand that TIME_WAIT and ephemeral port increases are not > usually needed, but I am concerned with the case of reverse proxying > thousands of very short lived requests per second. I suppose it's > likely for the service to die long before I exhaust available > resources, but at least I'll know I won't be bottlenecking anything. Again it depends on the traffic pattern. The important number for TIME_WAIT & ports is the number of connections Squid makes to the web servers, not really the number of connections it is receiving. > I appreciate your feedback! I do think it would be valuable for this > type of qualified information to make it into the wiki somewhere. > I'll look for the process to do so, but if you have any hints as to > where this info should live I would love to hear them. Instructions how to contribute to the wiki is given on the wiki first page, second paragraph. http://wiki.squid-cache.org/ Regards Henrik
