On Tue, Oct 27 2020 at 8:55am -0400, Mike Snitzer <snitzer@xxxxxxxxxx> wrote: > On Thu, Oct 22 2020 at 8:27pm -0400, > Mike Christie <michael.christie@xxxxxxxxxx> wrote: > > > This patch adds a path selector that selects paths based on a CPU to > > path mapping the user passes in and what CPU we are executing on. The > > primary user for this PS is where the app is optimized to use specific > > CPUs so other PSs undo the apps handy work, and the storage and it's > > transport are not a bottlneck. > > > > For these io-affinity PS setups a path's transport/interconnect > > perf is not going to flucuate a lot and there is no major differences > > between paths, so QL/HST smarts do not help and RR always messes up > > what the app is trying to do. > > > > On a system with 16 cores, where you have a job per CPU: > > > > fio --filename=/dev/dm-0 --direct=1 --rw=randrw --bs=4k \ > > --ioengine=libaio --iodepth=128 --numjobs=16 > > > > and a dm-multipath device setup where each CPU is mapped to one path: > > > > // When in mq mode I had to set dm_mq_nr_hw_queues=$NUM_PATHS. > > OK, the modparam was/is a means to an end but the default of 1 is very > limiting (especially in that it becomes one-size-fits-all, which isn't > true, for all dm-multipath devices in the system). > > If you have any ideas for what a sane heuristic would be for > dm_mq_nr_hw_queues I'm open to suggestions. But DM target <-> DM core > <-> early block core interface coordination is "fun". ;) > > > // Bio mode also showed similar results. > > 0 16777216 multipath 0 0 1 1 io-affinity 0 16 1 8:16 1 8:32 2 8:64 4 > > 8:48 8 8:80 10 8:96 20 8:112 40 8:128 80 8:144 100 8:160 200 8:176 > > 400 8:192 800 8:208 1000 8:224 2000 8:240 4000 65:0 8000 > > > > we can see a IOPs increase of 25%. > > Great. What utility/code are you using to extract the path:cpu affinity? > Is it array specific? Which hardware pins IO like this? > > Will you, or others, be enhancing multipath-tools to allow passing such > io-affinity DM multipath tables? > > > The percent increase depends on the device and interconnect. For a > > slower/medium speed path/device that can do around 180K IOPs a path > > if you ran that fio command to it directly we saw a 25% increase like > > above. Slower path'd devices that could do around 90K per path showed > > maybe around a 2 - 5% increase. If you use something like null_blk or > > scsi_debug which can multi-million IOPs and hack it up so each device > > they export shows up as a path then you see 50%+ increases. > > > > Signed-off-by: Mike Christie <michael.christie@xxxxxxxxxx> > > --- > > drivers/md/Kconfig | 9 ++ > > drivers/md/Makefile | 1 + > > drivers/md/dm-io-affinity.c | 272 ++++++++++++++++++++++++++++++++++++++++++++ > > 3 files changed, 282 insertions(+) > > create mode 100644 drivers/md/dm-io-affinity.c > > > > diff --git a/drivers/md/Kconfig b/drivers/md/Kconfig > > index 30ba357..c82d8b6 100644 > > --- a/drivers/md/Kconfig > > +++ b/drivers/md/Kconfig > > @@ -463,6 +463,15 @@ config DM_MULTIPATH_HST > > > > If unsure, say N. > > > > +config DM_MULTIPATH_IOA > > + tristate "I/O Path Selector based on CPU submission" > > + depends on DM_MULTIPATH > > + help > > + This path selector selects the path based on the CPU the IO is > > + executed on and the CPU to path mapping setup at path addition time. > > + > > + If unsure, say N. > > + > > config DM_DELAY > > tristate "I/O delaying target" > > depends on BLK_DEV_DM > > diff --git a/drivers/md/Makefile b/drivers/md/Makefile > > index 6d3e234..4f95f33 100644 > > --- a/drivers/md/Makefile > > +++ b/drivers/md/Makefile > > @@ -59,6 +59,7 @@ obj-$(CONFIG_DM_MULTIPATH) += dm-multipath.o dm-round-robin.o > > obj-$(CONFIG_DM_MULTIPATH_QL) += dm-queue-length.o > > obj-$(CONFIG_DM_MULTIPATH_ST) += dm-service-time.o > > obj-$(CONFIG_DM_MULTIPATH_HST) += dm-historical-service-time.o > > +obj-$(CONFIG_DM_MULTIPATH_IOA) += dm-io-affinity.o > > obj-$(CONFIG_DM_SWITCH) += dm-switch.o > > obj-$(CONFIG_DM_SNAPSHOT) += dm-snapshot.o > > obj-$(CONFIG_DM_PERSISTENT_DATA) += persistent-data/ > > Thinking about renaming all PS files to have a dm-ps prefix... > > Fact that we have dm-io.c makes dm-io-affinity.c all the more confusing. > > Can you rename to dm-ps-io-affinity.c and post v2? You know what, I'll take care of it when I deal with the other path selectors. We'll need entries like this in Makefile: dm-io-affinity-y += dm-ps-io-affinity.o Otherwise the module won't be name dm-io-affinity and "io-affinity" on DM table wouldn't autoload the kernel module. Alternatively, MODALIAS can be used... might prefer that as it'd reduce Makefile complexity for each path-selector. Mike -- dm-devel mailing list dm-devel@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/dm-devel
