On Wed, 15 Oct 2008 16:21:02 -0400 "J. Bruce Fields" <bfields@xxxxxxxxxxxx> wrote: > On Wed, Oct 15, 2008 at 08:14:57AM -0400, Jeff Layton wrote: > > From the descriptions in those commits, it looks like the check > > in svc_check_conn_limits was intended to prevent messages from too > > many sockets overloading the receive buffers. > > Aren't the receive buffers per-connection? > > --b. > Oof...you're quite right. I misunderstood the code earlier. Thinking this through a bit more... I suppose the main worry is that we have a service with too few threads and a ton of sockets. We end up in a situation where the receive buffers aren't getting processed quickly enough and connections start stalling out. I suppose the only real remedy for that is to increase the number of threads, but that's not an option for services like lockd. So maybe ignoring this check on single-threaded services is the way to go after all? I don't see a way to make this auto-tuning based on memory since that doesn't seem to be what the check is intended to help... -- Jeff Layton <jlayton@xxxxxxxxxx> -- To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
