Обсуждение: fsync tests

I ran test_fsync on a local ext 3 partition and on a raid jfs SSD array 8x32GB RAID10.  The local disk returned quickly and gave what looks like acceptable results.  However, the RAID looks pretty bad.  Any one have any idea why fsync would take so long on an SSD RAID 10.


LOCAL DISK
Simple write timing:
    write                    0.005460

Compare fsync times on write() and non-write() descriptor:
(If the times are similar, fsync() can sync data written
 on a different descriptor.)
    write, fsync, close      0.072429
    write, close, fsync      0.075627

Compare one o_sync write to two:
    (o_sync unavailable) 
    (o_dsync unavailable) 
    write, fdatasync         0.075907
    write, fsync,            0.076320

Compare file sync methods with 2 8k writes:
    (o_dsync unavailable) 
    write, fdatasync         0.108367
    write, fsync,            0.108659



RAID 10 SSD jfs
Simple write timing:
    write                    0.004843

Compare fsync times on write() and non-write() descriptor:
(If the times are similar, fsync() can sync data written
 on a different descriptor.)
    write, fsync, close     95.546382
    write, close, fsync     51.407276

Compare one o_sync write to two:
    (o_sync unavailable) 
    (o_dsync unavailable) 
    write, fdatasync        10.999764
    write, fsync,           80.696959

Compare file sync methods with 2 8k writes:
    (o_dsync unavailable) 
    write, fdatasync       111.931881
    write, fsync,           99.329298

RBN wrote:
> I ran test_fsync on a local ext 3 partition and on a raid jfs SSD array
> 8x32GB RAID10.  The local disk returned quickly and gave what looks like
> acceptable results.  However, the RAID looks pretty bad.  Any one have any
> idea why fsync would take so long on an SSD RAID 10.

As the original author of the test, all I can say is that those numbers
look bad.  :-(

---------------------------------------------------------------------------

>
>
> LOCAL DISK
> Simple write timing:
>     write                    0.005460
>
> Compare fsync times on write() and non-write() descriptor:
> (If the times are similar, fsync() can sync data written
>  on a different descriptor.)
>     write, fsync, close      0.072429
>     write, close, fsync      0.075627
>
> Compare one o_sync write to two:
>     (o_sync unavailable)
>     (o_dsync unavailable)
>     write, fdatasync         0.075907
>     write, fsync,            0.076320
>
> Compare file sync methods with 2 8k writes:
>     (o_dsync unavailable)
>     write, fdatasync         0.108367
>     write, fsync,            0.108659
>
>
>
> RAID 10 SSD jfs
> Simple write timing:
>     write                    0.004843
>
> Compare fsync times on write() and non-write() descriptor:
> (If the times are similar, fsync() can sync data written
>  on a different descriptor.)
>     write, fsync, close     95.546382
>     write, close, fsync     51.407276
>
> Compare one o_sync write to two:
>     (o_sync unavailable)
>     (o_dsync unavailable)
>     write, fdatasync        10.999764
>     write, fsync,           80.696959
>
> Compare file sync methods with 2 8k writes:
>     (o_dsync unavailable)
>     write, fdatasync       111.931881
>     write, fsync,           99.329298

--
  Bruce Momjian  <bruce@momjian.us>        http://momjian.us
  EnterpriseDB                             http://enterprisedb.com

  + None of us is going to be here forever. +

On Sun, Jun 6, 2010 at 5:07 PM, RBN <rbn.ggl@gmail.com> wrote:
>
>
> I ran test_fsync on a local ext 3 partition and on a raid jfs SSD array
> 8x32GB RAID10.  The local disk returned quickly and gave what looks like
> acceptable results.  However, the RAID looks pretty bad.  Any one have any
> idea why fsync would take so long on an SSD RAID 10.

Are you just running 1 loop?  Those numbers are worse than bad if
you're running one loop.

How are the SSD drives attached to your machine?  HW RAID, JBOD HW
Caching, direct HBA?