| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-Id: <004f01c8bfed$640fa380$2c2eea80$@jp.nec.com>
Date: Tue, 27 May 2008 20:32:48 +0900
From: "Satoshi UCHIDA" <s-uchida@...jp.nec.com>
To: "'Ryo Tsuruta'" <ryov@...inux.co.jp>
Cc: <axboe@...nel.dk>, <vtaras@...nvz.org>,
<containers@...ts.linux-foundation.org>,
<tom-sugawara@...jp.nec.com>, <linux-kernel@...r.kernel.org>
Subject: RE: [RFC][v2][patch 0/12][CFQ-cgroup]Yet another I/O bandwidth controlling subsystem for CGroups based on CFQ
Hi, Tsuruta-san.
> I'm looking forward to your report.
>
I report my tests.
My test shows following features.
o The guaranteeing degrees are widely in write I/Os than in read I/Os for each environment.
o Vasily's scheduler can guarantee I/O control.
However, its guaranteeing degree is narrow.
(in particular, at low priority)
o Satoshi's scheduler can guarantee I/O control.
However, guaranteeing degree is too small in write only and low priority case.
o Write I/Os are faster than read I/Os.
And, CFQ scheduler controls I/Os by time slice.
So, guaranteeing degree is caused difference from estimating degree at requests level by the situation of read and write I/Os.
I'll continue testing variously.
I hope to improve I/O control scheduler through many tests.
Details of the tests are as follows:
Environment:
Linux version 2.6.25-rc5-mm1 based.
4 type:
kernel with Vasily's scheduler
kernel with Satoshi's scheduler
Native kernel
Native kernel and use ionice commands to each process.
CPU0: Intel(R) Core(TM)2 CPU 6700 @ 2.66GHz stepping 6
CPU1: Intel(R) Core(TM)2 CPU 6700 @ 2.66GHz stepping 6
Memory: 4060180k/5242880k available (2653k kernel code, 132264k reserved, 1412k data, 356k init)
scsi 3:0:0:0: Direct-Access ATA WDC WD2500JS-19N 10.0 PQ: 0 ANSI: 5
sd 3:0:0:0: [sdb] 488282256 512-byte hardware sectors (250001 MB)
sd 3:0:0:0: [sdb] Write Protect is off
sd 3:0:0:0: [sdb] Mode Sense: 00 3a 00 00
sd 3:0:0:0: [sdb] Write cache: enabled, read cache: enabled, doesn't support DPO or FUA
sd 3:0:0:0: [sdb] 488282256 512-byte hardware sectors (250001 MB)
sd 3:0:0:0: [sdb] Write Protect is off
sd 3:0:0:0: [sdb] Mode Sense: 00 3a 00 00
sd 3:0:0:0: [sdb] Write cache: enabled, read cache: enabled, doesn't support DPO or FUA
sdb: sdb1 sdb2 sdb3 sdb4 < sdb5 >
Test 1:
Procedures:
o Prepare 200 files which size is 250MB on 1 partition sdb3
o Create 3 groups with priority 0, 4 and 7.
Estimate peformance in Vasily's scheduler : group-1 57.1% group-2 35.7% group-3 7.2%
Estimate peformance in Satoshi's scheduler : group-1 61.5% group-2 30.8% group-3 7.7%
Estimate peformance in Native Linux CFQ scheduler with ionice command :
group-1 92.8% group-2 5.8% group-3 1.4%
o Run many processes issuing random direct I/O with 4KB data on each files in 3 groups.
#1 Run 25 processes issuing read I/O only per groups.
#2 Run 25 processes issuing write I/O only per groups.
#3 Run 15 processes issuing read I/O and 15 processes issuing writeI/O only per groups.
o Count up the number of I/Os which have done in 120 seconds.
o Measure at 5 times in each situation.
o Results is calculated by averatve of 5 times.
Results:
Vasily's scheduler
The number of I/Os (percentage to total I/Os)
-----------------------------------------------------------------------------------
| group | group 1 | group 2 | group 3 | total |
| priority | 7(highest) | 4 | 0(lowest) | I/Os |
|---------------+------------------+------------------+------------------|----------|
| #1 Read | 6077.2 (46.33%) | 3680.8 (28.06%) | 3360.2 (25.61%) | 13118.2 |
| #2 Write | 10291.2 (53.13%) | 5282.8 (27.27%) | 3796.2 (19.60%) | 19370.2 |
| #3 Read&Write | 7218.0 (49.86%) | 4273.0 (29.52%) | 2986.0 (20.63%) | 14477.0 |
| | (Read 45.52%) | (Read 51.96%) | (Read 52.63%)| (Read 48.89%)|
------------------------------------------------------------------------------------
Satoshi's scheduler
The number of I/Os (percentage to total I/O)
-----------------------------------------------------------------------------------
| group | group 1 | group 2 | group 3 | total |
| priority | 0(highest) | 4 | 7(lowest) | I/Os |
|---------------+------------------+------------------+------------------|----------|
| #1 Read | 9082.2 (60.90%) | 4403.0 (29.53%) | 1427.4 (9.57%) | 14912.6 |
| #2 Write | 15449.0 (68.74%) | 6144.2 (27.34%) | 881.8 (3.92%) | 22475.0 |
| #3 Read&Write | 11283.6 (65.35%) | 4699.0 (27.21%) | 1284.8 (7.44%) | 17267.4 |
| | (Read 41.08%) | (Read 47.84%) | (Read 57.07%)| (Read 44.11%)|
------------------------------------------------------------------------------------
Native Linux CFQ scheduler
The number of I/Os (percentage to total I/O)
-----------------------------------------------------------------------------------
| group | group 1 | group 2 | group 3 | total |
|---------------+------------------+------------------+------------------|----------|
| #1 Read | 4362.2 (34.94%) | 3864.4 (30.95%) | 4259.8 (34.12%) | 12486.4 |
| #2 Write | 6918.4 (37.23%) | 5894.0 (31.71%) | 5772.0 (31.06%) | 18584.4 |
| #3 Read&Write | 4701.2 (33.62%) | 4788.0 (34.24%) | 4496.0 (32.15%) | 13985.2 |
| | (Read 45.85%) | (Read 48.99%) | (Read 51.28%)| (Read 48.67%)|
------------------------------------------------------------------------------------
Native Linux CFQ scheduler with ionice command
The number of I/Os (percentage to total I/O)
-----------------------------------------------------------------------------------
| group | group 1 | group 2 | group 3 | total |
| priority | 0(highest) | 4 | 7(lowest) | I/Os |
|---------------+------------------+------------------+------------------|----------|
| #1 Read | 12844.2 (85.34%) | 1544.8 (10.26%) | 661.8 (4.40%) | 15050.8 |
| #2 Write | 24858.4 (92.44%) | 1568.4 ( 5.83%) | 463.4 (1.72%) | 26890.2 |
| #3 Read&Write | 16205.4 (85.53%) | 2016.8 (10.64%) | 725.6 (3.83%) | 18947.8 |
| | (Read 37.49%) | (Read 57.97%) | (Read 56.62%)| (Read 40.40%)|
------------------------------------------------------------------------------------
Test 2:
Procedures:
o Prepare 200 files which size is 250MB on 1 partition sdb3
o Create 3 groups with priority 0, 4 and 7.
o Run many processes issuing random direct I/O with 4KB data on each files in 3 groups.
#1 Run 25 processes issuing read I/O only in group 1 and group 2 and run 25 processes issuing write I/O only in group 3.
(This pattern is represent by "R-R-W".)
#2 Run 25 processes issuing read I/O only in group 1 and group 3 and run 25 processes issuing write I/O only in group 2.
(This pattern is represent by "R-W-R".)
#3 Run 25 processes issuing read I/O only in group 2 and group 3 and run 25 processes issuing write I/O only in group 1.
(This pattern is represent by "R-R-W".)
o Count up the number of I/Os which have done in 120 seconds.
o Measure at 5 times in each situation.
o Results is calculated by averatve of 5 times.
Results:
Vasily's scheduler
The number of I/Os (percentage to total I/Os)
-----------------------------------------------------------------------------------
| group | group 1 | group 2 | group 3 | total |
| priority | 7(highest) | 4 | 0(lowest) | I/Os |
|---------------+------------------+------------------+------------------|----------|
| #1 R-R-W | 8828.2 (52.46%) | 4372.2 (25.98%) | 3628.8 (21.56%) | 16829.2 |
| #2 R-W-R | 5510.2 (35.01%) | 6584.6 (41.83%) | 3646.0 (23.16%) | 15740.8 |
| #3 W-R-R | 6400.4 (41.91%) | 3856.4 (25.25%) | 5016.4 (32.84%) | 15273.2 |
------------------------------------------------------------------------------------
Results shows peculiar case in test #2.
I/O counts in group 2 are 5911, 4895, 8498, 9300 and 4319.
In third and fourth time, I/O counts are huge.
An average of first, second and fifth results is following.
-----------------------------------------------------------------------------------
| group | group 1 | group 2 | group 3 | total |
| priority | 7(highest) | 4 | 0(lowest) | I/Os |
|---------------+------------------+------------------+------------------|----------|
| #2 R-W-R | 6285.7 (41.82%) | 5041.7 (33.54%) | 3702.7 (24.64%) | 15030.0 |
-----------------------------------------------------------------------------------
Satoshi's scheduler
The number of I/Os (percentage to total I/O)
-----------------------------------------------------------------------------------
| group | group 1 | group 2 | group 3 | total |
| priority | 0(highest) | 4 | 7(lowest) | I/Os |
|---------------+------------------+------------------+------------------|----------|
| #1 R-R-W | 9178.6 (61.95%) | 4510.8 (30.44%) | 1127.4 (7.61%) | 14816.8 |
| #2 R-W-R | 9398.0 (56.29%) | 6152.2 (36.85%) | 1146.4 (6.87%) | 16696.6 |
| #3 W-R-R | 15527.0 (72.38%) | 4544.8 (21.19%) | 1380.0 (6.43%) | 21451.8 |
------------------------------------------------------------------------------------
Native Linux CFQ scheduler
The number of I/Os (percentage to total I/O)
-----------------------------------------------------------------------------------
| group | group 1 | group 2 | group 3 | total |
|---------------+------------------+------------------+------------------|----------|
| #1 R-R-W | 4622.4 (33.15%) | 4739.2 (33.98%) | 4583.4 (32.87%) | 13945.0 |
| #2 R-W-R | 4610.6 (31.72%) | 5502.2 (37.85%) | 4422.4 (30.43%) | 14535.2 |
| #3 W-R-R | 5518.0 (37.57%) | 4734.2 (32.24%) | 4433.2 (30.19%) | 14685.4 |
-----------------------------------------------------------------------------------
Native Linux CFQ scheduler with ionice
The number of I/Os (percentage to total I/O)
-----------------------------------------------------------------------------------
| group | group 1 | group 2 | group 3 | total |
| priority | 0(highest) | 4 | 7(lowest) | I/Os |
|---------------+------------------+------------------+------------------|----------|
| #1 Read | 12619.4 (84.49%) | 1537.4 (10.29%) | 779.2 (5.22%) | 14936.0 |
| #2 Write | 12724.4 (81.44%) | 2276.6 (14.57%) | 623.2 (3.99%) | 15624.2 |
| #3 Read&Write | 24442.8 (91.75%) | 1592.6 ( 5.98%) | 604.6 (2.27%) | 26640.0 |
-----------------------------------------------------------------------------------
Thanks,
Satoshi UCHIDA.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists