Very inconsistent (read) speed on UFS2

[Lev Serebryakov]

Hello, Daniel.
You wrote 31 августа 2011 г., 12:48:32:

> On 31.08.11 11:36, Lev Serebryakov wrote:
>> device     r/s   w/s    kr/s    kw/s wait svc_t  %b
>> ada1     340.9 292.9 43138.8   146.5    0   1.2  42
>> ada2     340.9 293.9 43138.8   147.0    0   1.9  63
>> ada3     340.9 292.9 43044.7   146.5    0   1.5  57
>> ada4     341.9 292.9 43232.9   146.5    0   1.3  42
>> ada5     341.9 292.0 43138.8   146.0    2   1.3  40
>>
> Very interesting, this writes. You need to find out what is causing these.
  Yep. I've been very surprised by them.

> Just some random thoughts:

> This flapping may have something to do with the drives' internal caches.
> What are the drives?
  WD20EARS, it is WD Green 2Tb, advanced format. Yes, I know, that
 they are not best performers at all, when here are seeks. It is why I
 don't expect good performance in random or multi-threaded
 (multi-client) access patterns here.

   And, yes, I know about advanced format. Stripe size is 128Kb, and
 GEOM is built from raw drives, so all stripes are aligned. FS is
 created on raw GEOM, without any partitioning again, and block size
 is 32Kb, so everything should be aligned here too.

   Really, if all reading speeds were, say, 120MiB/s, but
 every time and consistent, I don't start this thread. In case I would
 blame HDDs and my parsimony, but not software :)

> SATA drives, unlike SAS have simplex communication with the host, that
> is, the drive cannot simultaneously read and write data and commands 
> (from/to host). There might be some, perhaps locking contention in 
> there? It is not contention for bandwidth obviously.
    Yep...

> In any case, you cannot measure read performance as long as it
> intermixes with writes, especially as you noted that your RAID5 code has
> some non-obvious write characteristics/optimizations.
  I understand. Now I should understand how to pin down these writes.

-- 
// Black Lion AKA Lev Serebryakov <lev at FreeBSD.org>