Re: seagate disks on sil3114 - slow?

[Tejun Heo <htejun@gmail.com>]

Soeren Sonnenburg wrote:
> On Fri, 2007-10-26 at 11:32 +0900, Tejun Heo wrote:
>> Soeren Sonnenburg wrote:
>>> On Thu, 2007-10-18 at 01:07 -0400, Jeff Garzik wrote:
>>>> Soeren Sonnenburg wrote:
>>>>> Dear all,
>>>>>
>>>>> I just now realize that all disk i.o. on my machine is quite slow
>>>>> compared with the pata disks I had before. Furthermore I recognized that
>>>>> I only have seagates (ST3400832AS,ST3400620AS,ST3750640AS,ST3750640AS)
>>>>> connected to a sil3114 controller.
>>>>>
>>>>> Being on kernel 2.6.23.1 and stumbling across
>>>>> http://home-tj.org/wiki/index.php/Sil_m15w I am now wondering if this
>>>>> patch made it already in kernel (or when it will make it if it is not
>>>>> yet in/why never).
>>>> Are you sure you have a 3114?  the mod15write isn't applied to that chip.
>>>> Does your dmesg say "applying Seagate errata fix (mod15write workaround)" ?
>>> At least lspci / dmesg indicates I have a 3114 ... but maybe the problem
>>> is something else. Is there a raw device speed, readonly benchmark to
>>> check whether things are as expected?
> 
> looks like I did not fully reply to this one: yes I have a 3114 and
> based on Bernds patch to enable m15w support for sil3114 I added the
> first two drives to the blacklist and yes I saw the applying mod15write
> workaround messages for these disks afterwards. These disks attached to
> the internal promise seems to work stably now (uptime is a week now),
> with the occasional error that thanks to the new EH is successfully
> recovered from - good job!!
> 
>> "dd if=/dev/sdX of=/dev/null bs=1M"?  Also, does changing IO scheduler
>> to deadline make any difference?
> 
> OK I knew that one... about 50-60M/s - sounds reasonable, but if I
> understand the m15w problem correctly it can only be triggered on
> writes...

3114 -> no m15w problem and m15w doesn't slow down transfers.  It locks
up certain drives.  What slows down transfers is workaround for m15w.

-- 
tejun