On Tue, 2 Dec 2014 13:24:09 -0500 Jeff Layton <jlayton@xxxxxxxxxxxxxxx> wrote: > tl;dr: this code works and is much simpler than the dedicated thread > pool, but there are some latencies in the workqueue code that > seem to keep it from being as fast as it could be. > > This patchset is a little skunkworks project that I've been poking at > for the last few weeks. Currently nfsd uses a dedicated thread pool to > handle RPCs, but that requires maintaining a rather large swath of > "fiddly" code to handle the threads and transports. > > This patchset represents an alternative approach, which makes nfsd use > workqueues to do its bidding rather than a dedicated thread pool. When a > transport needs to do work, we simply queue it to the workqueue in > softirq context and let it service the transport. > > The current draft is runtime-switchable via a new sunrpc pool_mode > module parameter setting. When that's set to "workqueue", nfsd will use > a workqueue-based service. One of the goals of this patchset was to > *not* need to change any userland code, so starting it up using rpc.nfsd > still works as expected. The only real difference is that the nfsdfs > "threads" file is reinterpreted as the "max_active" value for the > workqueue. Hi Jeff, I haven't looked very closely at the code, but in principal I think this is an excellent idea. Having to set a number of threads manually was never nice as it is impossible to give sensible guidance on what an appropriate number is. Tying max_active to "threads" doesn't really make sense I think. "max_active" is a per-cpu number and I think the only meaningful numbers are "1" (meaning concurrent works might mutually deadlock) or infinity (which is approximated as 512). I would just ignore the "threads" number when workqueues are used.... or maybe enable workqueues when "auto" is written to "threads"?? Using a shrinker to manage the allocation and freeing of svc_rqst is a really good idea. It will put pressure on the effective number of threads when needed, but will not artificially constrain things. The combination of workqueue and shrinker seems like a perfect match for nfsd. I hope you can work out the latency issues! Thanks, NeilBrown
Attachment:
pgpV7FfGwlKI9.pgp
Description: OpenPGP digital signature
