The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

nous · 2013-06-03 18:10:57

graysky wrote:

nous wrote:
The BFS kernels are by definition a little slower than vanilla, which is the trade-off for the lower latency (i.e. better responsiveness) they provide by default. That been said, it's also possible to tweak the CFS scheduler towards lower latencies but my subjective perception judged it poorly compared to BFS.
Have to disagree with you here, nous. Three endpoints including time to compress, time to compile, and time to encode (x264) where evaluated and BFS patched kernels were faster on the whole than their non-bfs patched counterparts.
http://repo-ck.com/bench/cpu_schedulers_compared.pdf
"In addition to the primary design goals of the bfs, increased desktop interactivity and responsiveness, kernels patched with the ck1 patch set including the bfs outperformed the vanilla kernel using the cfs at nearly all the performance-based benchmarks tested. Further study with a larger test set could be conducted, but based on the small test set of 7 PCs evaluated, these increases in process queuing, efficiency/speed are, on the whole, independent of CPU type (mono, dual, quad, hyperthreaded, etc.), CPU architecture (32-bit and 64-bit), and of CPU multiplicity (mono or dual socket)."

First of all, you've done a great job in your presentation of the CFS vs BFS case. It may lack a few key factors that might significantly affect the end results (i.e. you only stated the hardware you benchmarked, not the software setup), but overall it is a good place to start and further explore. Which I did.

So, I took your bash script, modified it slightly and ran it on two of my boxen, an AMD Athlon II X2 @ 3GHz with 4GB of RAM running XFCE4 and an Intel Core 2 @ 2.60GHz with 6GB of RAM running Mate. I wanted to take your test one step further by defining the environment the benchmarks ran within. Thus, I ran 3 tests on each machine:
1. In single-user runlevel, at nice priority -19, allowing the test to squeeze every last CPU cycle out of the scheduler,
2. In graphical environment, at nice priority 0, with a few concurrent tasks running so as to provide an artificial load, emulating a normal user desktop, and
3. As #2, with the benchmarks running unnoticed at nice priority 19, to see how the scheduler would distribute the abundance of CPU power among the various other tasks.

The programs I chose to simulate the load in tests 2 and 3 were:
1. MPlayer, endlessly looping a x264/720p matroska, through vdpau which of course is very CPU-light on NVidias,
2. Firefox, with 1 tab at http://www.miniclip.com which has 2-3 small flash boxes that cause notoriously heavy loads for their size,
3. Clementine, looping through a playlist and fetching data and song information,
4. Dropbox.

In total, the CPU load of these programs was fluctuating between 20 and 40% in either box.

I disabled zram in both PCs, as my first batch exhibited remarkable deviation in run times and I just couldn't figure out why.

The 3 benchmarks that tested the schedulers were:
1. Compilation of a fully configured mplayer svn snapshot. For CFS with 'make -j3', for BFS with 'make -j2' as recommended by the author. An additional test was performed with 'make -j19', to stress the scheduler even more, although that's not a normal desktop use of make, but showed most interesting results.
2. Encoding with mencoder of a video into x264 (different one on each box, don't be alarmed by the results) to /dev/null (-o /dev/null). All CPU cores are utilized in each PC.
3. Compression of a tar image with xz (copy /usr/src in /tmp and compress). 1 core was used.
All tests pulled their files from and put them in a subdir in /tmp (tmpfs). The CFS kernel was the stock 3.9.3-ARCH. The BFS one was 3.9.3-pf-core2 on the Pentium and 3.9.4-ck-k10 on the Athlon. One might say that the CFS was at a disadvantage, running on a generic kernel vs the CPU optimized ones BFS ran on. But that wasn't even a focus of the test and I personally don't think it would make a great deal of difference at the results. Another compression test with lrzip utilizing all cores was performed on the Athlon II PC.

The tests ran 5 times each. The first run was discarded, as a warm-up.

I expected CFS to be generally faster. After all, the goal of BFS is not speed but rather low latency and desktop responsiveness. Which is precisely why I use it everywhere. I must be completely clear on this: I'm a die-hard BFS user and that's the reason I put linux-pf on the AUR. I don't give a dime if a video encoding or package compilation and compression finish 3 or 13 seconds later, when the entire process generally takes 15 minutes to complete. That having been said, let's go through the results. (Note: of course nice priorities officially range from -19 to 19, but I like to use -20 and 20. Oh, well.)

Pentium Core 2 E5300 @ 2.60 GHz

CFS, 3.9.3-ARCH, runlevel 1, nice -20
2,orion,3.9.3-1-ARCH,Compile,273.839647580,CFS,2013-05-27 23:29:55
3,orion,3.9.3-1-ARCH,Compile,273.843602936,CFS,2013-05-27 23:34:30
4,orion,3.9.3-1-ARCH,Compile,274.236322136,CFS,2013-05-27 23:39:04
5,orion,3.9.3-1-ARCH,Compile,273.855982238,CFS,2013-05-27 23:43:39
2,orion,3.9.3-1-ARCH,Encode,201.885036276,CFS,2013-05-27 23:51:38
3,orion,3.9.3-1-ARCH,Encode,201.566581356,CFS,2013-05-27 23:54:59
4,orion,3.9.3-1-ARCH,Encode,201.824596574,CFS,2013-05-27 23:58:21
5,orion,3.9.3-1-ARCH,Encode,202.312055204,CFS,2013-05-28 00:01:43
2,orion,3.9.3-1-ARCH,Compress,63.431066715,CFS,2013-05-28 00:07:13
3,orion,3.9.3-1-ARCH,Compress,63.219442267,CFS,2013-05-28 00:08:17
4,orion,3.9.3-1-ARCH,Compress,63.141987500,CFS,2013-05-28 00:09:20
5,orion,3.9.3-1-ARCH,Compress,63.053064024,CFS,2013-05-28 00:10:23

BFS, 3.9.3-pf, runlevel 1, nice -20
2,orion,3.9.3-pf,Compile,276.540509634,BFS,2013-05-27 22:14:01
3,orion,3.9.3-pf,Compile,276.559740635,BFS,2013-05-27 22:18:38
4,orion,3.9.3-pf,Compile,276.623804033,BFS,2013-05-27 22:23:16
5,orion,3.9.3-pf,Compile,276.670173959,BFS,2013-05-27 22:27:53
2,orion,3.9.3-pf,Encode,218.721420533,BFS,2013-05-27 22:36:09
3,orion,3.9.3-pf,Encode,218.718366992,BFS,2013-05-27 22:39:48
4,orion,3.9.3-pf,Encode,219.098066792,BFS,2013-05-27 22:43:27
5,orion,3.9.3-pf,Encode,219.428052253,BFS,2013-05-27 22:47:06
2,orion,3.9.3-pf,Compress,65.718380083,BFS,2013-05-27 22:52:54
3,orion,3.9.3-pf,Compress,65.708016001,BFS,2013-05-27 22:54:00
4,orion,3.9.3-pf,Compress,67.057308194,BFS,2013-05-27 22:55:05
5,orion,3.9.3-pf,Compress,68.082945332,BFS,2013-05-27 22:56:12

Comment: very small deviation from each test in sequence, BFS is slightly slower, a little more at encoding.

CFS, 3.9.3-ARCH, runlevel 5, nice 0, (firefox@www.miniclip.com, clementine, mplayer-vdpau looping a 720p matroska)
2,orion,3.9.3-1-ARCH,Compile,365.567715384,CFS,2013-05-28 17:21:45
3,orion,3.9.3-1-ARCH,Compile,366.068062106,CFS,2013-05-28 17:27:51
4,orion,3.9.3-1-ARCH,Compile,364.927825720,CFS,2013-05-28 17:33:57
5,orion,3.9.3-1-ARCH,Compile,364.643736914,CFS,2013-05-28 17:40:03
2,orion,3.9.3-1-ARCH,Encode,251.631909633,CFS,2013-05-28 17:50:21
3,orion,3.9.3-1-ARCH,Encode,250.505403723,CFS,2013-05-28 17:54:33
4,orion,3.9.3-1-ARCH,Encode,249.035681744,CFS,2013-05-28 17:58:43
5,orion,3.9.3-1-ARCH,Encode,251.696632829,CFS,2013-05-28 18:02:52
2,orion,3.9.3-1-ARCH,Compress,67.067290353,CFS,2013-05-28 18:09:18
3,orion,3.9.3-1-ARCH,Compress,67.082080789,CFS,2013-05-28 18:10:25
4,orion,3.9.3-1-ARCH,Compress,67.113858750,CFS,2013-05-28 18:11:32
5,orion,3.9.3-1-ARCH,Compress,66.925152513,CFS,2013-05-28 18:12:39

BFS, 3.9.3-pf, runlevel 5, nice 0, (firefox@www.miniclip.com, clementine, mplayer-vdpau looping a 720p matroska)
2,orion,3.9.3-pf,Compile,426.575523194,BFS,2013-05-28 07:31:26
3,orion,3.9.3-pf,Compile,430.236799620,BFS,2013-05-28 07:38:34
4,orion,3.9.3-pf,Compile,433.073104447,BFS,2013-05-28 07:45:44
5,orion,3.9.3-pf,Compile,430.106568760,BFS,2013-05-28 07:52:58
2,orion,3.9.3-pf,Encode,324.458309562,BFS,2013-05-28 08:05:36
3,orion,3.9.3-pf,Encode,302.165068992,BFS,2013-05-28 08:11:00
4,orion,3.9.3-pf,Encode,301.069478269,BFS,2013-05-28 08:16:02
5,orion,3.9.3-pf,Encode,300.225844082,BFS,2013-05-28 08:21:03
2,orion,3.9.3-pf,Compress,75.056115296,BFS,2013-05-28 08:28:08
3,orion,3.9.3-pf,Compress,75.643331354,BFS,2013-05-28 08:29:23
4,orion,3.9.3-pf,Compress,74.788901007,BFS,2013-05-28 08:30:38
5,orion,3.9.3-pf,Compress,75.268100192,BFS,2013-05-28 08:31:53

CFS is clearly a winner.

CFS, 3.9.3-ARCH, runlevel 5, nice 20, (firefox@www.miniclip.com, clementine, mplayer-vdpau looping a 720p matroska)
2,orion,3.9.3-1-ARCH,Compile,365.253588971,CFS,2013-05-28 18:19:53
3,orion,3.9.3-1-ARCH,Compile,364.527432364,CFS,2013-05-28 18:25:59
4,orion,3.9.3-1-ARCH,Compile,365.037900103,CFS,2013-05-28 18:32:04
5,orion,3.9.3-1-ARCH,Compile,364.299891538,CFS,2013-05-28 18:38:10
2,orion,3.9.3-1-ARCH,Encode,253.079271158,CFS,2013-05-28 18:48:26
3,orion,3.9.3-1-ARCH,Encode,251.327881445,CFS,2013-05-28 18:52:39
4,orion,3.9.3-1-ARCH,Encode,249.355326798,CFS,2013-05-28 18:56:50
5,orion,3.9.3-1-ARCH,Encode,252.519082940,CFS,2013-05-28 19:01:00
2,orion,3.9.3-1-ARCH,Compress,67.218847698,CFS,2013-05-28 19:06:21
3,orion,3.9.3-1-ARCH,Compress,66.889469713,CFS,2013-05-28 19:07:29
4,orion,3.9.3-1-ARCH,Compress,67.317543449,CFS,2013-05-28 19:08:36
5,orion,3.9.3-1-ARCH,Compress,67.086741037,CFS,2013-05-28 19:09:43

BFS, 3.9.3-pf, runlevel 5, nice 20, (firefox@www.miniclip.com, clementine, mplayer-vdpau looping a 720p matroska)
2,orion,3.9.3-pf,Compile,428.971103559,BFS,2013-05-28 10:23:13
3,orion,3.9.3-pf,Compile,437.677955196,BFS,2013-05-28 10:30:22
4,orion,3.9.3-pf,Compile,437.355290520,BFS,2013-05-28 10:37:41
5,orion,3.9.3-pf,Compile,431.780245537,BFS,2013-05-28 10:44:59
2,orion,3.9.3-pf,Encode,297.637168964,BFS,2013-05-28 10:57:11
3,orion,3.9.3-pf,Encode,305.917561292,BFS,2013-05-28 11:02:09
4,orion,3.9.3-pf,Encode,304.174386619,BFS,2013-05-28 11:07:15
5,orion,3.9.3-pf,Encode,306.060767895,BFS,2013-05-28 11:12:19
2,orion,3.9.3-pf,Compress,82.389801389,BFS,2013-05-28 11:19:34
3,orion,3.9.3-pf,Compress,82.018852159,BFS,2013-05-28 11:20:57
4,orion,3.9.3-pf,Compress,83.926561820,BFS,2013-05-28 11:22:19
5,orion,3.9.3-pf,Compress,82.359122859,BFS,2013-05-28 11:23:43

Here CFS is faster too.

AMD K10 Athlon II X2 @ 3 GHz

# CFS, runlevel 1, nice -20
2,hyperion,3.9.3-1-ARCH,Compile,274.149734379,CFS,2013-05-30 19:19:43
3,hyperion,3.9.3-1-ARCH,Compile,273.458213215,CFS,2013-05-30 19:24:18
4,hyperion,3.9.3-1-ARCH,Compile,274.001875997,CFS,2013-05-30 19:28:52
5,hyperion,3.9.3-1-ARCH,Compile,273.460219550,CFS,2013-05-30 19:33:27
2,hyperion,3.9.3-1-ARCH,Encode,139.160839603,CFS,2013-05-30 19:40:19
3,hyperion,3.9.3-1-ARCH,Encode,139.323758602,CFS,2013-05-30 19:42:38
4,hyperion,3.9.3-1-ARCH,Encode,138.904045727,CFS,2013-05-30 19:44:58
5,hyperion,3.9.3-1-ARCH,Encode,139.355157008,CFS,2013-05-30 19:47:17
2,hyperion,3.9.3-1-ARCH,Compress,76.648689022,CFS,2013-05-30 19:51:27
3,hyperion,3.9.3-1-ARCH,Compress,76.677580357,CFS,2013-05-30 19:52:43
4,hyperion,3.9.3-1-ARCH,Compress,76.869320645,CFS,2013-05-30 19:54:00
5,hyperion,3.9.3-1-ARCH,Compress,76.814949231,CFS,2013-05-30 19:55:17

# BFS, runlevel 1, nice -20
2,hyperion,3.9.4-1-ck,Compile,266.757315575,BFS,2013-05-31 07:17:36
3,hyperion,3.9.4-1-ck,Compile,266.670687605,BFS,2013-05-31 07:22:03
4,hyperion,3.9.4-1-ck,Compile,266.708117088,BFS,2013-05-31 07:26:30
5,hyperion,3.9.4-1-ck,Compile,266.870665671,BFS,2013-05-31 07:30:57
2,hyperion,3.9.4-1-ck,Encode,146.190383830,BFS,2013-05-31 07:37:50
3,hyperion,3.9.4-1-ck,Encode,145.925822635,BFS,2013-05-31 07:40:17
4,hyperion,3.9.4-1-ck,Encode,145.928657792,BFS,2013-05-31 07:42:43
5,hyperion,3.9.4-1-ck,Encode,146.119847753,BFS,2013-05-31 07:45:09
2,hyperion,3.9.4-1-ck,Compress,80.748200004,BFS,2013-05-31 07:48:57
3,hyperion,3.9.4-1-ck,Compress,81.451115119,BFS,2013-05-31 07:50:18
4,hyperion,3.9.4-1-ck,Compress,81.122056287,BFS,2013-05-31 07:51:39
5,hyperion,3.9.4-1-ck,Compress,80.957620501,BFS,2013-05-31 07:53:00

Comment: CFS is faster at encoding and compressing, slower at compiling.

# CFS, runlevel 5, nice 0 (firefox@www.miniclip.com, clementine, mplayer-vdpau looping a 720p matroska)
2,hyperion,3.9.3-1-ARCH,Compile,412.134148345,CFS,2013-05-30 22:00:14
3,hyperion,3.9.3-1-ARCH,Compile,412.700641938,CFS,2013-05-30 22:07:07
4,hyperion,3.9.3-1-ARCH,Compile,411.421222500,CFS,2013-05-30 22:14:00
5,hyperion,3.9.3-1-ARCH,Compile,411.628028696,CFS,2013-05-30 22:20:52
2,hyperion,3.9.3-1-ARCH,Encode,195.203460942,CFS,2013-05-30 22:31:01
3,hyperion,3.9.3-1-ARCH,Encode,194.683939815,CFS,2013-05-30 22:34:16
4,hyperion,3.9.3-1-ARCH,Encode,194.957419776,CFS,2013-05-30 22:37:31
5,hyperion,3.9.3-1-ARCH,Encode,195.152548690,CFS,2013-05-30 22:40:46
2,hyperion,3.9.3-1-ARCH,Compress,80.176535952,CFS,2013-05-30 22:45:32
3,hyperion,3.9.3-1-ARCH,Compress,79.764986904,CFS,2013-05-30 22:46:52
4,hyperion,3.9.3-1-ARCH,Compress,79.605422494,CFS,2013-05-30 22:48:12
5,hyperion,3.9.3-1-ARCH,Compress,79.505666696,CFS,2013-05-30 22:49:32

# BFS, runlevel 5, nice 0 (firefox@www.miniclip.com, clementine, mplayer-vdpau looping a 720p matroska)
2,hyperion,3.9.4-1-ck,Compile,411.757674376,BFS,2013-05-31 01:33:59
3,hyperion,3.9.4-1-ck,Compile,411.113806774,BFS,2013-05-31 01:40:52
4,hyperion,3.9.4-1-ck,Compile,411.243125970,BFS,2013-05-31 01:47:44
5,hyperion,3.9.4-1-ck,Compile,412.231301470,BFS,2013-05-31 01:54:36
2,hyperion,3.9.4-1-ck,Encode,209.944156882,BFS,2013-05-31 02:04:59
3,hyperion,3.9.4-1-ck,Encode,211.065329845,BFS,2013-05-31 02:08:29
4,hyperion,3.9.4-1-ck,Encode,208.464988899,BFS,2013-05-31 02:12:00
5,hyperion,3.9.4-1-ck,Encode,209.639097247,BFS,2013-05-31 02:15:29
2,hyperion,3.9.4-1-ck,Compress,100.522694957,BFS,2013-05-31 02:21:13
3,hyperion,3.9.4-1-ck,Compress,101.222192669,BFS,2013-05-31 02:22:53
4,hyperion,3.9.4-1-ck,Compress,100.942984762,BFS,2013-05-31 02:24:34
5,hyperion,3.9.4-1-ck,Compress,101.234929925,BFS,2013-05-31 02:26:15

Comment: same or better speed under CFS, especially at compressing.

# CFS, runlevel 5, nice 20 (firefox@www.miniclip.com, clementine, mplayer-vdpau looping a 720p matroska)
2,hyperion,3.9.3-1-ARCH,Compile,419.002663199,CFS,2013-05-30 23:01:28
3,hyperion,3.9.3-1-ARCH,Compile,420.936664837,CFS,2013-05-30 23:08:28
4,hyperion,3.9.3-1-ARCH,Compile,423.328528520,CFS,2013-05-30 23:15:30
5,hyperion,3.9.3-1-ARCH,Compile,421.893004523,CFS,2013-05-30 23:22:34
2,hyperion,3.9.3-1-ARCH,Encode,195.936355434,CFS,2013-05-30 23:32:53
3,hyperion,3.9.3-1-ARCH,Encode,195.846742536,CFS,2013-05-30 23:36:09
4,hyperion,3.9.3-1-ARCH,Encode,195.299812915,CFS,2013-05-30 23:39:24
5,hyperion,3.9.3-1-ARCH,Encode,195.642419985,CFS,2013-05-30 23:42:40
2,hyperion,3.9.3-1-ARCH,Compress,89.305732194,CFS,2013-05-30 23:47:29
3,hyperion,3.9.3-1-ARCH,Compress,89.013632651,CFS,2013-05-30 23:48:59
4,hyperion,3.9.3-1-ARCH,Compress,89.062349096,CFS,2013-05-30 23:50:28
5,hyperion,3.9.3-1-ARCH,Compress,89.074416846,CFS,2013-05-30 23:51:57

# BFS, runlevel 5, nice 20 (firefox@www.miniclip.com, clementine, mplayer-vdpau looping a 720p matroska)
2,hyperion,3.9.4-1-ck,Compile,412.311834995,BFS,2013-05-31 02:35:02
3,hyperion,3.9.4-1-ck,Compile,412.013986738,BFS,2013-05-31 02:41:55
4,hyperion,3.9.4-1-ck,Compile,411.341996708,BFS,2013-05-31 02:48:47
5,hyperion,3.9.4-1-ck,Compile,413.574672812,BFS,2013-05-31 02:55:40
2,hyperion,3.9.4-1-ck,Encode,211.293805955,BFS,2013-05-31 03:06:06
3,hyperion,3.9.4-1-ck,Encode,210.543653840,BFS,2013-05-31 03:09:37
4,hyperion,3.9.4-1-ck,Encode,210.645787071,BFS,2013-05-31 03:13:08
5,hyperion,3.9.4-1-ck,Encode,211.299153776,BFS,2013-05-31 03:16:38
2,hyperion,3.9.4-1-ck,Compress,106.071817687,BFS,2013-05-31 03:21:57
3,hyperion,3.9.4-1-ck,Compress,105.556187925,BFS,2013-05-31 03:23:44
4,hyperion,3.9.4-1-ck,Compress,105.137862807,BFS,2013-05-31 03:25:29
5,hyperion,3.9.4-1-ck,Compress,105.553909030,BFS,2013-05-31 03:27:14

Comment: BFS is almost as fast as at nice priority 0, and faster than CFS at compiling. The CFS-BFS margin is narrower here.

Bonus tests! Compiling with 'make -j19' and encoding with 'threads=8' in Athlon II:

# CFS, runlevel 5, nice 0, -j19
2,hyperion,3.9.3-1-ARCH,Compile,300.386296235,CFS,2013-06-01 14:41:18
3,hyperion,3.9.3-1-ARCH,Compile,300.264676918,CFS,2013-06-01 14:46:18
4,hyperion,3.9.3-1-ARCH,Compile,300.372279162,CFS,2013-06-01 14:51:19

# BFS, runlevel 5, nice 0, -j19
2,hyperion,3.9.4-1-ck,Compile,283.450907752,BFS,2013-06-01 15:08:18
3,hyperion,3.9.4-1-ck,Compile,283.386943115,BFS,2013-06-01 15:13:02
4,hyperion,3.9.4-1-ck,Compile,283.853832552,BFS,2013-06-01 15:17:46

Slower than the recommended -j3/-j2, but CFS suffers harder.

# BFS, runlevel 1, nice -20, -j4
2,hyperion,3.9.4-1-ck,Compile,283.836244742,BFS,2013-06-02 01:24:47
3,hyperion,3.9.4-1-ck,Compile,283.762850638,BFS,2013-06-02 01:29:31
4,hyperion,3.9.4-1-ck,Compile,283.584377288,BFS,2013-06-02 01:34:16
5,hyperion,3.9.4-1-ck,Compile,283.880673709,BFS,2013-06-02 01:39:00
# BFS, runlevel 1, nice -20, -j8
2,hyperion,3.9.4-1-ck,Compile,284.008915269,BFS,2013-06-02 01:48:31
3,hyperion,3.9.4-1-ck,Compile,284.165466309,BFS,2013-06-02 01:53:16
4,hyperion,3.9.4-1-ck,Compile,284.130012354,BFS,2013-06-02 01:58:01
5,hyperion,3.9.4-1-ck,Compile,284.129094977,BFS,2013-06-02 02:02:45
# BFS, runlevel 1, nice -20, -j16
2,hyperion,3.9.4-1-ck,Compile,283.754278961,BFS,2013-06-02 02:12:17
3,hyperion,3.9.4-1-ck,Compile,283.878814670,BFS,2013-06-02 02:17:01
4,hyperion,3.9.4-1-ck,Compile,283.997796005,BFS,2013-06-02 02:21:46
5,hyperion,3.9.4-1-ck,Compile,283.741360648,BFS,2013-06-02 02:26:30

Here it's clear that Con Kolivas is right about 'make -j2'. No significant speed gains are observed with -j4,8,16.

# CFS, runlevel 5, nice 0, -j19, threads=8 (firefox@www.miniclip.com, clementine, mplayer-vdpau looping a 720p matroska)
2,hyperion,3.9.3-1-ARCH,Compile,416.647583763,CFS,2013-06-01 18:01:07
3,hyperion,3.9.3-1-ARCH,Compile,417.432461893,CFS,2013-06-01 18:08:04
4,hyperion,3.9.3-1-ARCH,Compile,418.823885957,CFS,2013-06-01 18:15:02
5,hyperion,3.9.3-1-ARCH,Compile,418.352675897,CFS,2013-06-01 18:22:01
2,hyperion,3.9.3-1-ARCH,Encode,193.846426921,CFS,2013-06-01 20:46:44
3,hyperion,3.9.3-1-ARCH,Encode,193.249283162,CFS,2013-06-01 20:49:58
4,hyperion,3.9.3-1-ARCH,Encode,193.494891041,CFS,2013-06-01 20:53:11
5,hyperion,3.9.3-1-ARCH,Encode,193.283120252,CFS,2013-06-01 20:56:25

# BFS, runlevel 5, nice 0, -j19, threads=8 (firefox@www.miniclip.com, clementine, mplayer-vdpau looping a 720p matroska)
2,hyperion,3.9.4-1-ck,Compile,336.632319860,BFS,2013-06-01 15:45:32
3,hyperion,3.9.4-1-ck,Compile,334.341032402,BFS,2013-06-01 15:51:10
4,hyperion,3.9.4-1-ck,Compile,334.692631221,BFS,2013-06-01 15:56:45
5,hyperion,3.9.4-1-ck,Compile,333.298600626,BFS,2013-06-01 16:02:20
2,hyperion,3.9.4-1-ck,Encode,195.435245641,BFS,2013-06-01 16:11:07
3,hyperion,3.9.4-1-ck,Encode,192.836084036,BFS,2013-06-01 16:14:22
4,hyperion,3.9.4-1-ck,Encode,191.513742493,BFS,2013-06-01 16:17:35
5,hyperion,3.9.4-1-ck,Encode,193.469279378,BFS,2013-06-01 16:20:47

Huge speed gain at compiling with BFS here. Encoding with 8 threads is faster too. No gains whatsoever with these options under CFS.

# CFS, runlevel 5, nice 20, -j19, threads=8 (firefox@www.miniclip.com, clementine, mplayer-vdpau looping a 720p matroska)
2,hyperion,3.9.3-1-ARCH,Compile,440.035240070,CFS,2013-06-01 21:07:06
3,hyperion,3.9.3-1-ARCH,Compile,444.050347899,CFS,2013-06-01 21:14:26
4,hyperion,3.9.3-1-ARCH,Compile,444.461044865,CFS,2013-06-01 21:21:51
5,hyperion,3.9.3-1-ARCH,Compile,441.483990788,CFS,2013-06-01 21:29:16
2,hyperion,3.9.3-1-ARCH,Encode,193.048977781,CFS,2013-06-01 21:39:51
3,hyperion,3.9.3-1-ARCH,Encode,193.598163800,CFS,2013-06-01 21:43:04
4,hyperion,3.9.3-1-ARCH,Encode,192.642754742,CFS,2013-06-01 21:46:18
5,hyperion,3.9.3-1-ARCH,Encode,193.832351321,CFS,2013-06-01 21:49:30

# BFS, runlevel 5, nice 20, -j19, threads=8 (firefox@www.miniclip.com, clementine, mplayer-vdpau looping a 720p matroska)
2,hyperion,3.9.4-1-ck,Compile,423.903885077,BFS,2013-06-01 16:31:08
3,hyperion,3.9.4-1-ck,Compile,427.013678881,BFS,2013-06-01 16:38:12
4,hyperion,3.9.4-1-ck,Compile,424.031966082,BFS,2013-06-01 16:45:20
5,hyperion,3.9.4-1-ck,Compile,426.276298304,BFS,2013-06-01 16:52:25
2,hyperion,3.9.4-1-ck,Encode,195.566342134,BFS,2013-06-01 17:02:44
3,hyperion,3.9.4-1-ck,Encode,194.313523100,BFS,2013-06-01 17:05:59
4,hyperion,3.9.4-1-ck,Encode,196.014171431,BFS,2013-06-01 17:09:13
5,hyperion,3.9.4-1-ck,Encode,193.993581156,BFS,2013-06-01 17:12:30

At nice 20 BFS falls back to normal speeds.

Finally, multithreaded compression with lrztar:

# CFS, runlevel 5, nice 0 (firefox@www.miniclip.com, clementine, mplayer-vdpau looping a 720p matroska)
2,hyperion,3.9.3-1-ARCH,Compress,49.098348503,CFS,2013-06-01 23:26:47
3,hyperion,3.9.3-1-ARCH,Compress,47.854364915,CFS,2013-06-01 23:27:36
4,hyperion,3.9.3-1-ARCH,Compress,48.512722625,CFS,2013-06-01 23:28:24
5,hyperion,3.9.3-1-ARCH,Compress,48.181695154,CFS,2013-06-01 23:29:13

# CFS, runlevel 5, nice 20 (firefox@www.miniclip.com, clementine, mplayer-vdpau looping a 720p matroska)
2,hyperion,3.9.3-1-ARCH,Compress,48.977484727,CFS,2013-06-01 23:34:48
3,hyperion,3.9.3-1-ARCH,Compress,47.973123632,CFS,2013-06-01 23:35:37
4,hyperion,3.9.3-1-ARCH,Compress,49.047547925,CFS,2013-06-01 23:36:25
5,hyperion,3.9.3-1-ARCH,Compress,48.769420684,CFS,2013-06-01 23:37:14

# BFS, runlevel 5, nice 0 (firefox@www.miniclip.com, clementine, mplayer-vdpau looping a 720p matroska)
2,hyperion,3.9.4-1-ck,Compress,42.169158938,BFS,2013-06-01 23:47:53
3,hyperion,3.9.4-1-ck,Compress,41.604586265,BFS,2013-06-01 23:48:35
4,hyperion,3.9.4-1-ck,Compress,41.512915401,BFS,2013-06-01 23:49:17
5,hyperion,3.9.4-1-ck,Compress,43.137300803,BFS,2013-06-01 23:49:58

# BFS, runlevel 5, nice 20 (firefox@www.miniclip.com, clementine, mplayer-vdpau looping a 720p matroska)
2,hyperion,3.9.4-1-ck,Compress,41.468909304,BFS,2013-06-01 23:51:34
3,hyperion,3.9.4-1-ck,Compress,41.925650283,BFS,2013-06-01 23:52:16
4,hyperion,3.9.4-1-ck,Compress,41.601447697,BFS,2013-06-01 23:52:58
5,hyperion,3.9.4-1-ck,Compress,42.345945927,BFS,2013-06-01 23:53:39

BFS is clearly the winner here, unlike the single-threaded xz compression.

I must repeat here what I wrote earlier: the CFS runs were performed on the stock -ARCH kernel, which of course is generic. The Core 2 PC ran linux-pf-core2 and the Athlon II one ran linux-ck-k10, which (in theory) should give BFS an edge.

One might argue that very few repetitions of the tests were performed. True. But, from my point of view, microsecond accuracy is for places far away from home. Another valid argument is that the test environment wasn't strictly defined. Also true. But, (1) I don't care and/or have the time to devote to an issue that is solved for me since BFS appeared and, (2) one can't strictly set the confines of home/office environment which the majority of people use their computers (and BFS) for. I used what I usually use.

TL;DR: CFS is generally faster, as it should. BFS is snappier and I love it. Oh, and one of these days I'll make some graphs with gnuplot to make it shinier and easier to the eye (and also appear more authoritative).

Last edited by nous (2013-06-03 18:52:51)

Thaodan · 2013-06-10 19:23:37

Updated PKGBUILD, aufs3 is enabled again (url to gitrepo was wrong).
x86_64 repo packages are in progress.

Thaodan · 2013-06-10 19:55:56

Updated Repo, add Sandybridge packages for this version.

Thaodan · 2013-06-10 20:18:04

Updated Repo, add Ivry packages for this version.

gbj13 · 2013-06-13 17:32:42

The dropbox repo seems to be down, any alternatives?

Thaodan · 2013-06-13 17:49:50

gbj13 wrote:

The dropbox repo seems to be down, any alternatives?

You mean my repo? Its online see: https://dl.dropboxusercontent.com/u/172 … index.html

gbj13 · 2013-06-13 21:42:06

Yes I am sorry, user errror on my part. Thank you for your work

Thaodan · 2013-06-13 21:51:18

gbj13 wrote:

Yes I am sorry, user errror on my part. Thank you for your work

No problem.

Lockheed · 2013-06-20 18:47:12

.

Last edited by Lockheed (2013-06-21 15:07:56)

Thaodan · 2013-06-21 19:41:08

updated sandy- and ivybridge packages in my repo.

Thaodan · 2013-06-30 02:22:51

nvidia sandybridge drivers updated

Thaodan · 2013-06-30 02:48:38

nvidia ivybridge packages updated too

Thaodan · 2013-07-06 18:20:37

all packages are updated except nvidia-pf-ivybridge

Thaodan · 2013-07-18 22:34:08

updated packages
NOTE: I added patch from https://bugzilla.kernel.org/show_bug.cgi?id=60530 to fix the issue please report issues

jakob · 2013-08-11 09:25:44

Thanks Thaodan for your work!

Is anybody else also having issues resuming from hibernation? 3.10.1-pf doesn't seem to be able to find the swap partition (/dev/sda3 in my case) by itself anymore. I had to append resume=/dev/sda3 to my kernel line again after having removed it for more than half a year I believe. Otherwise, instead of resuming from hibernation Arch would boot up regularly saying could not find a resume device or something like that.

Anyone else having a similar experience?

//edit: I tried to hibernate on stock ARCH and got the error message "no swap on /dev/sda3, try swapon -a" which led me to the fact that there was something wrong with that and after fixing it, I can hibernate and resume with TOI again. Non-issue.

Last edited by jakob (2013-08-28 12:04:53)

Thaodan · 2013-08-17 15:16:16

nvidia-ivybridge packages updated.

again my repo is:
https://dl.dropboxusercontent.com/u/172 … -pf/x86_64

*to open it in your browser add /index.html to the url

Thaodan · 2013-09-09 15:16:22

updated PKGBUILD aufs3 inst ready, packages are compiling

Thaodan · 2013-09-09 19:20:43

packages are updated to 3.11, except nvidia-pf-ivybridge

nbvcxz · 2013-09-10 14:26:23

In pre-compiled packages the zswap is not set (pfkernel/linux-pf-k7 3.11.1-1 (base)):

# CONFIG_ZBUD is not set
# CONFIG_ZSWAP is not set

but they are enabled in core ArchLinux kernel. Could you enable them in linux-pf as zswap seems to be one of the most important news last time.

Thaodan · 2013-09-10 16:08:38

yes, I missed to check Archs configs more.

Last edited by Thaodan (2013-09-10 16:12:05)

Thaodan · 2013-09-10 16:33:55

Done

afader · 2013-09-12 14:10:48

The latest packages don't seem to have aufs? Or do I have to install it separately?

jakob · 2013-09-12 14:37:00

Thaodan wrote:

updated PKGBUILD aufs3 inst ready, packages are compiling

sic

Thaodan · 2013-09-12 17:10:40

there already some patches around but not in upstream.

Xemertix · 2013-09-14 23:24:47

With this kernel the vbox module doesn't compile because genksyms doesn't work.

# uname -r
3.11.1-pf

#  dkms install vboxhost/$(pacman -Q virtualbox|awk {'print $2'}|sed 's/\-.\+//') -k $(uname -rm|sed 's/\ /\//')

Kernel preparation unnecessary for this kernel.  Skipping...

Building module:
cleaning build area....
make KERNELRELEASE=3.11.1-pf -C /usr/lib/modules/3.11.1-pf/build M=/var/lib/dkms/vboxhost/4.2.18/build....(bad exit status: 2)
Error! Bad return status for module build on kernel: 3.11.1-pf (i686)
Consult /var/lib/dkms/vboxhost/4.2.18/build/make.log for more information.

DKMS make.log for vboxhost-4.2.18 for kernel 3.11.1-pf (i686)
Sun Sep 15 01:19:37 CEST 2013
make: Entering directory `/usr/src/linux-3.11.1-pf'
  LD      /var/lib/dkms/vboxhost/4.2.18/build/built-in.o
  LD      /var/lib/dkms/vboxhost/4.2.18/build/vboxdrv/built-in.o
  CC [M]  /var/lib/dkms/vboxhost/4.2.18/build/vboxdrv/linux/SUPDrv-linux.o
/bin/sh: scripts/genksyms/genksyms: cannot execute binary file
make[2]: *** [/var/lib/dkms/vboxhost/4.2.18/build/vboxdrv/linux/SUPDrv-linux.o] Error 126
make[1]: *** [/var/lib/dkms/vboxhost/4.2.18/build/vboxdrv] Error 2
make: *** [_module_/var/lib/dkms/vboxhost/4.2.18/build] Error 2
make: Leaving directory `/usr/src/linux-3.11.1-pf'

cd /usr/src/linux-3.11.1-pf/scripts/genksyms/
./genksyms
bash: ./genksyms: cannot execute binary file

Arch Linux

#726 2013-06-03 18:10:57

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#727 2013-06-10 19:23:37

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#728 2013-06-10 19:55:56

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#729 2013-06-10 20:18:04

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#730 2013-06-13 17:32:42

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#731 2013-06-13 17:49:50

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#732 2013-06-13 21:42:06

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#733 2013-06-13 21:51:18

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#734 2013-06-20 18:47:12

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#735 2013-06-21 19:41:08

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#736 2013-06-30 02:22:51

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#737 2013-06-30 02:48:38

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#738 2013-07-06 18:20:37

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#739 2013-07-18 22:34:08

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#740 2013-08-11 09:25:44

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#741 2013-08-17 15:16:16

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#742 2013-09-09 15:16:22

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#743 2013-09-09 19:20:43

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#744 2013-09-10 14:26:23

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#745 2013-09-10 16:08:38

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#746 2013-09-10 16:33:55

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#747 2013-09-12 14:10:48

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#748 2013-09-12 14:37:00

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#749 2013-09-12 17:10:40

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#750 2013-09-14 23:24:47

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

Board footer