Re: [Scst-devel] tgtd segfault with software RAID, hard resetting link

[Bart V. A. <bar...@gm...>]

On Wed, Apr 15, 2009 at 11:43 PM, Tomasz Chmielewski <ma...@wp...> wrote:
> Bart Van Assche schrieb:
>> The command "hdparm -Y ..." at least slows down the I/O on the target.
>> You have to make sure that the timeout settings on the initiator are
>> such that it can handle the slower I/O processing. See e.g. section
>> 8.2 in http://www.open-iscsi.org/docs/README.
>
> Yep, I know.
>
> I have such settings on the initiator:
>
> node.session.timeo.replacement_timeout = 1000000
> node.conn[0].timeo.noop_out_interval = 0
> node.conn[0].timeo.noop_out_timeout = 0
>
> Which, to my knowledge, should allow the initiator to survive any
> disconnections/timeouts shorter than ~11 days.
>
> With it, it survives when I do "hdparm -Y ..." with STGT (patched) as a
> target; it shows IO errors with SCST as a target.
>
> Unless there are some other setting I'm not familiar with.

The above timeout settings make the initiator more robust with regard
to network problems, but not with regard to slow execution of iSCSI
commands. You will have to modify these parameters too.

Bart.