>
> I'm not sure how representative this is of real workloads, but it does
> look rather fatal for zswap. The differences are so large, I wonder if
> it's just some silly bug or config issue.
>
In my observation, zswap_pool_pages always close to zswap_stored_pages
in this testing.
I think it means that the fragmentation of zswap is heavy.
Since in idea state number of zswap pool pages should be half of stored pages.
The reason may be this workload is not suitable for compression.
The data of it can't be compressed to a low percent.
It can only compressed to around 70% percent(not exactly but at least
above 50%).
I made a simple patch to limit the fragment of zswap to 70%.
The result can be better but still not positive.
v3.10-rc4 v3.10-rc4
2G-parallio-nozswap 2G-parallio-zswapdefrag
Ops memcachetest-0M 1041.00 ( 0.00%) 1058.00 ( 1.63%)
Ops memcachetest-198M 973.00 ( 0.00%) 1019.00 ( 4.73%)
Ops memcachetest-430M 892.00 ( 0.00%) 831.00 ( -6.84%)
Ops memcachetest-661M 819.00 ( 0.00%) 850.00 ( 3.79%)
Ops memcachetest-893M 775.00 ( 0.00%) 784.00 ( 1.16%)
Ops memcachetest-1125M 764.00 ( 0.00%) 766.00 ( 0.26%)
Ops memcachetest-1356M 749.00 ( 0.00%) 782.00 ( 4.41%)
Ops io-duration-0M 0.00 ( 0.00%) 0.00 ( 0.00%)
Ops io-duration-198M 21.00 ( 0.00%) 28.00 (-33.33%)
Ops io-duration-430M 29.00 ( 0.00%) 32.00 (-10.34%)
Ops io-duration-661M 34.00 ( 0.00%) 34.00 ( 0.00%)
Ops io-duration-893M 36.00 ( 0.00%) 42.00 (-16.67%)
Ops io-duration-1125M 43.00 ( 0.00%) 47.00 ( -9.30%)
Ops io-duration-1356M 50.00 ( 0.00%) 49.00 ( 2.00%)
Ops swaptotal-0M 469193.00 ( 0.00%) 461146.00 ( 1.72%)
Ops swaptotal-198M 496201.00 ( 0.00%) 495692.00 ( 0.10%)
Ops swaptotal-430M 520400.00 ( 0.00%) 520252.00 ( 0.03%)
Ops swaptotal-661M 538872.00 ( 0.00%) 513541.00 ( 4.70%)
Ops swaptotal-893M 522590.00 ( 0.00%) 532311.00 ( -1.86%)
Ops swaptotal-1125M 526934.00 ( 0.00%) 527089.00 ( -0.03%)
Ops swaptotal-1356M 525241.00 ( 0.00%) 525747.00 ( -0.10%)
Ops swapin-0M 251226.00 ( 0.00%) 248426.00 ( 1.11%)
Ops swapin-198M 236126.00 ( 0.00%) 239031.00 ( -1.23%)
Ops swapin-430M 265193.00 ( 0.00%) 266174.00 ( -0.37%)
Ops swapin-661M 280281.00 ( 0.00%) 263151.00 ( 6.11%)
Ops swapin-893M 263004.00 ( 0.00%) 268473.00 ( -2.08%)
Ops swapin-1125M 261962.00 ( 0.00%) 264925.00 ( -1.13%)
Ops swapin-1356M 262571.00 ( 0.00%) 266695.00 ( -1.57%)
Ops minorfaults-0M 625759.00 ( 0.00%) 620448.00 ( 0.85%)
Ops minorfaults-198M 769752.00 ( 0.00%) 703458.00 ( 8.61%)
Ops minorfaults-430M 678590.00 ( 0.00%) 686257.00 ( -1.13%)
Ops minorfaults-661M 669308.00 ( 0.00%) 668845.00 ( 0.07%)
Ops minorfaults-893M 656343.00 ( 0.00%) 680286.00 ( -3.65%)
Ops minorfaults-1125M 657954.00 ( 0.00%) 655280.00 ( 0.41%)
Ops minorfaults-1356M 672035.00 ( 0.00%) 659861.00 ( 1.81%)
Ops majorfaults-0M 39395.00 ( 0.00%) 38828.00 ( 1.44%)
Ops majorfaults-198M 38758.00 ( 0.00%) 42876.00 (-10.62%)
Ops majorfaults-430M 39819.00 ( 0.00%) 38668.00 ( 2.89%)
Ops majorfaults-661M 40171.00 ( 0.00%) 38443.00 ( 4.30%)
Ops majorfaults-893M 37576.00 ( 0.00%) 37664.00 ( -0.23%)
Ops majorfaults-1125M 36891.00 ( 0.00%) 37527.00 ( -1.72%)
Ops majorfaults-1356M 37123.00 ( 0.00%) 37779.00 ( -1.77%)
--
Regards,
--Bob
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>