From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Date: Mon, 22 Jan 2018 17:16:29 -0700 From: Keith Busch To: Jens Axboe Cc: linux-nvme@lists.infradead.org, linux-block@vger.kernel.org, Christoph Hellwig , Sagi Grimberg Subject: Re: [PATCH 1/3] nvme/pci: Start request after doorbell ring Message-ID: <20180123001628.GT12043@localhost.localdomain> References: <20171221204636.2924-1-keith.busch@intel.com> <20171221204636.2924-2-keith.busch@intel.com> <45a16b4c-fbc0-106d-df37-439374c7b5dc@kernel.dk> <118beb1f-54eb-c65d-1c9c-4775ebde1fa8@kernel.dk> <20171221210250.GA2975@localhost.localdomain> <6ffc074a-3c5b-c76d-d213-6606a77626e0@kernel.dk> <20180103202105.GC11172@localhost.localdomain> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii In-Reply-To: <20180103202105.GC11172@localhost.localdomain> List-ID: On Wed, Jan 03, 2018 at 01:21:05PM -0700, Keith Busch wrote: > > I've removed the submission side poll in a local build, and amazingly I > am observing a not insignificant increase in latency without it on some > workloads with certain hardware. I will have to delay recommending/posting > removal of this pending further investigation. This took longer to narrow than I hoped. I had been getting very wild results, and there were two things to blame: a kernel irq issue[*], and a platform bios causing a lot of CPU thermal frequency throttling. Now that's sorted out, I am back to tuning this and the opprotunistic nvme submission side polling once again does not produce a measurable difference. My world is sane again. Will try to send an update for consideration this week after some more testing. * https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit?id=a0c9259dc4