[ofa-general] RE: [Scst-devel] SRP aggregate bandwidth decreasing as threads increase

[Chris Worley]

On Mon, May 18, 2009 at 1:18 PM, Bart Van Assche
<bart.vanassche at gmail.com> wrote:
> On Mon, May 18, 2009 at 8:40 PM, Chris Worley <worleys at gmail.com> wrote:
>> Any other ideas of things to try?
>
> Depends on the workload that is running on the initiators. Are the
> initiators performing linear I/O or block I/O ?

I'm not sure what "linear I/O" is.  It is block I/O at 56KB chunks;
using direct I/O.

> Which I/O scheduler is
> being used by the initiator systems, and how has it been configured ?

The noop scheduler is being used on the targets and initiators.  All
the standard schedulers performed worse.

> Which I/O scheduler has been configured on the target, and with which
> parameters ? As you probably know, you can find these parameters under
> /sys/class/block/sda/queue/{*,*/*}. Are you using scst_disk or
> scst_vdisk ?

scst_vdisk.

> And what is the kernel version of the target system ?

Ubuntu 8.10 with a 2.6.27 kernel, if my memory serves me correctly.

> By
> the way, an important I/O performance regression has been fixed in
> kernel 2.6.29 (see also http://lwn.net/Articles/325307/).

Thanks,  I'll try that.

Chris
>
> Bart.
>