Re: [Iometer-user] Very low results

[Ming Z. <mi...@el...>]

On Mon, 2006-11-20 at 19:23 +0100, Sascha Effert wrote:
> Hello,
>=20
> I am using IOMeter to messure an iSCSI-Storage connected to several ser=
vers.=20
> In my test each server running the danymo got an own partition. I start=
et=20

i assumed each server will get own iscsi disk? are these server share
one piece of physical storage in u target?


> some tests, described below. I got very strange results, much low than=20
> expected. IOMeter told me that my system has a throughput of about 2 MB=
/s for=20
> sequential writing. Using the Version of 2004 I get results about 30 MB=
/s,=20
> what is much more belevable.

the difference between 2004 and 2006 is that 2006 now use O_DIRECT by
default. this eliminate the cache effect in the box running dynamo but
might show u lower result than before.


>=20
> Used Version: 2006-07-27 and 2006-07-27-rc3
> IOMeter GUI: Windows 2003 Server
> Dynamos: Redhat AS 4 (2.6.9-42.0.3.ELsmp Kernel), i686, Intel Core2 CPU=
 @ 2.66=20
> GHz, 4 GB RAM
>=20
>=20
> Tests:
> Worker pro Server   : 1
> Disk Target         : virtuelles Laufwerk =FCber iSCSI
>                       (Direkt, nicht gemounted)
> Sectors             : 6553600
> Starting Disk Sector: 0
> Outstanding IOs     : 32
> Network Targets     : keine
> Access Specification: SequentialWrite
> Run Time            : 10 Minuten
> Ramp Up             : 2 Minuten
> Record Results      : All
> Cycling Options     : Normal
>=20
> SequentialWrite means:
> RandomRead:
> Default Assignment                    : None
> Transfer Request Size                 : 4 KB
> Percent Access Specification          : 100%
> Percent Read/Write Distribution       : 100% Write
> Percent Random/Sequential Distribution: 100% Sequential
> Bustiness Transfer Delay              : 0ms
> Burstiness Burst Length               : 1 IO
> Align IOs on                          : Sector Boundaries
> Replay Size                           : No Reply
>=20
> Can anybody tell me what is going wrong? I would prefer using the 2006-=
Version=20
> of IOMeter, because it seems to be much more stable than the version of=
 2004.
>=20
> A second question: I remember there was a patched version of the 2004-0=
7-30 in=20
> 2005. I can not find it any more on sf.net. Is it anywhere available?

i hided it.


>=20
> tschau
>=20
>=20
> Sascha Effert
> fe...@do...
> Tel.:(0177) 6266652
>=20