Re: [RFC][v2][patch 0/12][CFQ-cgroup]Yet another I/O bandwidthcontrolling subsystem for CGroups based on CFQ

[Ryo Tsuruta]

Hi Uchida-san,

I realized that the benchmark results which I posted on Apr 25 had
some problems with the testing environment.

  From: Ryo Tsuruta <ryov@xxxxxxxxxxxxx>
  Subject: Re: [RFC][v2][patch 0/12][CFQ-cgroup]Yet another I/O
  bandwidth controlling subsystem for CGroups based on CFQ
  Date: Fri, 25 Apr 2008 18:54:44 +0900 (JST)

Uchida-san said,

> In the test #2 and #3, did you use direct write?
> I guess you have used the non-direct write I/O (using cache).

I answered "Yes," but actually I did not use direct write I/O, because
I ran these tests on Xen-HVM. Xen-HVM backend driver doesn't use direct
I/O for actual disk operations even though guest OS uses direct I/O.

An another problem was that the CPU time was used up during the tests.

So, I retested with the new testing environment and got good results. 
The number of I/Os is proportioned according to the priority levels.

Details of the tests are as follows:

Envirionment:
  Linux version 2.6.25-rc2-mm1 based.
  CPU0: Intel(R) Core(TM)2 CPU          6600  @ 2.40GHz stepping 06
  CPU1: Intel(R) Core(TM)2 CPU          6600  @ 2.40GHz stepping 06
  Memory: 2063568k/2088576k available (2085k kernel code, 23684k
  reserved, 911k data, 240k init, 1171072k highmem)
  scsi 1:0:0:0: Direct-Access     ATA      WDC WD2500JS-55N 10.0 PQ: 0  ANSI: 5
  sd 1:0:0:0: [sdb] 488397168 512-byte hardware sectors (250059 MB)
  sd 1:0:0:0: [sdb] Write Protect is off
  sd 1:0:0:0: [sdb] Mode Sense: 00 3a 00 00
  sd 1:0:0:0: [sdb] Write cache: disabled, read cache: enabled,
  doesn't support DPO or FUA
  sdb: sdb1 sdb2 sdb3 sdb4 < sdb5 sdb6 sdb7 sdb8 sdb9 sdb10 sdb11
  sdb12 sdb13 sdb14 sdb15 >

Procedures:
  o Prepare 3 partitions sdb5, sdb6 and sdb7.
  o Run 100 processes issuing random direct I/O with 4KB data on each
    partitions.
  o Run 3 tests:
    #1 issuing read I/O only.
    #2 issuing write I/O only.
    #3 sdb5 and sdb6 are read, sdb7 is write.
  o Count up the number of I/Os which have done in 60 seconds.

Results:
                          Vasily's scheduler
               The number of I/Os (percentage to total I/Os)
   ---------------------------------------------------------------------
  | partition     |     sdb5     |     sdb6     |     sdb7     | total  |
  | priority      |  7(highest)  |      4       |  0(lowest)   |  I/Os  |
  |---------------+--------------+--------------+--------------|--------|
  | #1 read       |   3383(35%)  |   3164(33%)  |   3142(32%)  |  9689  |
  | #2 write      |   3017(42%)  |   2372(33%)  |   1851(26%)  |  7240  |
  | #3 read&write |   4300(36%)  |   3127(27%)  |   1521(17%)  |  8948  |
   ---------------------------------------------------------------------

                          Satoshi's scheduler
               The number of I/Os (percentage to total I/O)
   ---------------------------------------------------------------------
  | partition     |     sdb5     |     sdb6     |     sdb7     | total  |
  | priority      |  0(highest)  |      4       |  7(lowest)   |  I/Os  |
  |---------------+--------------+--------------+--------------|--------|
  | #1 read       |   3907(47%)  |   3126(38%)  |   1260(15%)  |  8293  |
  | #2 write      |   3389(41%)  |   3024(36%)  |   1901(23%)  |  8314  |
  | #3 read&write |   5028(53%)  |   3961(42%)  |    441( 5%)  |  9430  |
   ---------------------------------------------------------------------

Thanks,
Ryo Tsuruta
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/