Messages in this thread | | | Date | Fri, 18 Dec 2009 22:03:05 +0100 | Subject | Re: [PATCH] cfq-iosched: reduce write depth only if sync was delayed | From | Corrado Zoccolo <> |
| |
Thanks Jeff. Your tests show that with the patch, low_latency doesn't penalize writeback any more on fast hardware, so the goal of the patch is fulfilled. Lowering slice_async doesn't change the picture (I was expecting an improvement in seq write bandwidth). I'm still puzzled of why with 2.6.29 your SAN could achieve 300 MB/s sequential writes, and it can't achieve it any more. Can you try lowering slice_idle, trying e.g. 2ms?
Thanks Corrado
On Fri, Dec 18, 2009 at 4:32 PM, Jeff Moyer <jmoyer@redhat.com> wrote: > Corrado Zoccolo <czoccolo@gmail.com> writes: > >> Hi Jeff, >> On Wed, Dec 9, 2009 at 7:09 PM, Jeff Moyer <jmoyer@redhat.com> wrote: >>> Corrado Zoccolo <czoccolo@gmail.com> writes: >>> >>>> The numbers look good. Now, there is no penalty in having low_latency >>>> set for sequential writes, and just a small penalty for random ones. >>>> The fact that random reads are faster with low_latency set is interesting. >>>> Is the test is running with your patched tiobench (so that the number >>>> of random operations is comparable with sequential ones)? >>> >>> No, I forgot all about that. The number of random operations defaults >>> to 4000, which is pretty low. I'll re-run the tests with a number >>> comparable to the sequential runs. Sorry about that. >>> >> N.P. >> if you have time, can you also re-run the test changing: >> iosched/fifo_expire_async to 8 ? >> I hope that reducing the expire_async, will make cfq quicker at switching >> between the different threads, allowing more parallelism for seq >> writers on your hw. >> If this is the case, I think I can try to estimate the >> fifo_expire_async in the autotuning patch. > > Sorry this took so long. I've been rather busy of late. > > Cheers, > Jeff > > low_latency=1, fifo_expire_async=8 > > Unit information > ================ > File size = megabytes > Blk Size = bytes > Rate = megabytes per second > CPU% = percentage of CPU used during the test > Latency = milliseconds > Lat% = percent of requests that took longer than X seconds > CPU Eff = Rate divided by CPU% - throughput per cpu load > > Sequential Reads > File Blk Num Avg Maximum Lat% Lat% CPU > Identifier Size Size Thr Rate (CPU%) Latency Latency >2s >10s Eff > ---------------------------- ------ ----- --- ------ ------ --------- ----------- -------- -------- ----- > 2.6.32 8192 65536 8 88.39 89.74% 16.388 2032.62 0.00000 0.00000 98 > 2.6.32 8192 65536 16 90.77 185.3% 32.213 2175.99 0.00000 0.00000 49 > > Random Reads > File Blk Num Avg Maximum Lat% Lat% CPU > Identifier Size Size Thr Rate (CPU%) Latency Latency >2s >10s Eff > ---------------------------- ------ ----- --- ------ ------ --------- ----------- -------- -------- ----- > 2.6.32 8192 65536 8 19.62 25.74% 71.827 3397.26 0.00000 0.00000 76 > 2.6.32 8192 65536 16 23.82 55.01% 103.361 4075.53 0.00000 0.00000 43 > > Sequential Writes > File Blk Num Avg Maximum Lat% Lat% CPU > Identifier Size Size Thr Rate (CPU%) Latency Latency >2s >10s Eff > ---------------------------- ------ ----- --- ------ ------ --------- ----------- -------- -------- ----- > 2.6.32 8192 65536 8 108.28 1007.% 12.984 5643.55 0.00076 0.00000 11 > 2.6.32 8192 65536 16 112.40 2014.% 25.430 8592.98 0.00839 0.00000 6 > > Random Writes > File Blk Num Avg Maximum Lat% Lat% CPU > Identifier Size Size Thr Rate (CPU%) Latency Latency >2s >10s Eff > ---------------------------- ------ ----- --- ------ ------ --------- ----------- -------- -------- ----- > 2.6.32 8192 65536 8 63.94 337.7% 22.885 6047.22 0.00076 0.00000 19 > 2.6.32 8192 65536 16 61.94 662.5% 46.997 12759.69 0.15411 0.00000 9 > > > low_latency=0, fifo_expire_async=8 > > Unit information > ================ > File size = megabytes > Blk Size = bytes > Rate = megabytes per second > CPU% = percentage of CPU used during the test > Latency = milliseconds > Lat% = percent of requests that took longer than X seconds > CPU Eff = Rate divided by CPU% - throughput per cpu load > > Sequential Reads > File Blk Num Avg Maximum Lat% Lat% CPU > Identifier Size Size Thr Rate (CPU%) Latency Latency >2s >10s Eff > ---------------------------- ------ ----- --- ------ ------ --------- ----------- -------- -------- ----- > 2.6.32 8192 65536 8 89.13 88.30% 15.872 3101.39 0.00000 0.00000 101 > 2.6.32 8192 65536 16 86.78 161.7% 30.794 4909.02 0.00000 0.00000 54 > > Random Reads > File Blk Num Avg Maximum Lat% Lat% CPU > Identifier Size Size Thr Rate (CPU%) Latency Latency >2s >10s Eff > ---------------------------- ------ ----- --- ------ ------ --------- ----------- -------- -------- ----- > 2.6.32 8192 65536 8 20.21 26.64% 69.863 4285.42 0.00000 0.00000 76 > 2.6.32 8192 65536 16 20.10 52.75% 139.761 5986.94 0.00076 0.00000 38 > > Sequential Writes > File Blk Num Avg Maximum Lat% Lat% CPU > Identifier Size Size Thr Rate (CPU%) Latency Latency >2s >10s Eff > ---------------------------- ------ ----- --- ------ ------ --------- ----------- -------- -------- ----- > 2.6.32 8192 65536 8 108.74 1020.% 13.070 5331.78 0.00076 0.00000 11 > 2.6.32 8192 65536 16 112.18 2020.% 25.559 7903.16 0.00992 0.00000 6 > > Random Writes > File Blk Num Avg Maximum Lat% Lat% CPU > Identifier Size Size Thr Rate (CPU%) Latency Latency >2s >10s Eff > ---------------------------- ------ ----- --- ------ ------ --------- ----------- -------- -------- ----- > 2.6.32 8192 65536 8 64.53 337.8% 22.671 5388.77 0.00000 0.00000 19 > 2.6.32 8192 65536 16 61.75 668.9% 47.265 13271.37 0.12894 0.00000 9 >
-- __________________________________________________________________________
dott. Corrado Zoccolo mailto:czoccolo@gmail.com PhD - Department of Computer Science - University of Pisa, Italy -------------------------------------------------------------------------- The self-confidence of a warrior is not the self-confidence of the average man. The average man seeks certainty in the eyes of the onlooker and calls that self-confidence. The warrior seeks impeccability in his own eyes and calls that humbleness. Tales of Power - C. Castaneda
| |