kern/91410: [ata] Promise SATA controller + striping : "SETFEATURES SET TRANSFER MODE semaphore timeout"

[John McLaughlin]

Hi Søren,

	I've created the array with atacontrol, and after about an hour or so
with a make -j 4 buildworld running, network activity and stressing from
running:

	dd if=/dev/zero of=test bs=1024 count=1000000

	in a couple of sessions to write files to the array, it seems to be
rock solid.

	I'm going to stress it a bit more to see to make sure, and I'll try
your other suggestion of disabling PREEMPTIVE tomorrow along with GEOM

	I still have the problem where writing to the striped array doesn't
perform any better than writing to a single disk (when the system is
quiet of course!). While running above dd command, I measured the
throughput with iostat. The first output is for the array, which is if
anything slower at writing than the individual disk ad4, which is shown
in the first part of the second run. The second part is doing the dd to
both disks at the same time, which would seem to imply that the bus
speed isn't entirely to blame. 

	I imagine the interleave value for atacontrol will have a bearing, but
since much googling didn't turn up much guidance for values, I tried
both 64 and 512. Both gave similar write speeds. the 18-19MB/s is also
in line with the tests I was doing with GEOM before things went pear
shaped. Am I missing anything?

root at jmcl# iostat -w 10 ar0
      tty             ar0             cpu
 tin tout  KB/t tps  MB/s  us ni sy in id
(writing to raid0 array)
   0   13 127.31 149 18.56   2  0 51  3 44
   0    4 127.54 147 18.37   2  0 49  4 45
   0    4 120.47 153 18.03   2  0 52  3 43
   0    4 127.38 148 18.46   2  0 50  3 44
   0    4 127.92 148 18.44   3  0 50  4 44

root at jmcl# iostat -w 10 ad4 ad6
      tty             ad4              ad6             cpu
 tin tout  KB/t tps  MB/s   KB/t tps  MB/s  us ni sy in id

(writing to one disk)
   1   19 127.06 130 16.18   0.00   0  0.00   2  0 41  2 55
   0    6 127.63 150 18.66   0.00   0  0.00   3  0 49  2 47
   0  139 125.75 149 18.33   0.00   0  0.00   3  0 51  1 45
   0    6 125.90 151 18.58   0.00   0  0.00   2  0 50  1 47
   0    6 127.55 149 18.61   0.00   0  0.00   3  0 48  1 47

(writing to both disks simultaneously)

   0    6 122.43 125 14.90  127.52 116 14.49   4  0 93  3  1
   0    6 127.34 122 15.23  127.63 121 15.10   4  0 94  2  0
   0    6 124.17 126 15.27  124.49 122 14.88   4  0 94  2  1
   0    6 127.63 122 15.22  127.54 121 15.02   4  0 93  2  1
   0    6 127.24 122 15.12  127.63 121 15.08   4  0 93  2  0

- John

On Tue, 2006-01-10 at 14:51 +0100, Søren Schmidt wrote:
> Gleb Smirnoff wrote:
> > Synopsis: [ata] Promise SATA controller + striping : "SETFEATURES SET TRANSFER MODE semaphore timeout"
> > 
> > Responsible-Changed-From-To: freebsd-bugs->sos
> > Responsible-Changed-By: glebius
> > Responsible-Changed-When: Tue Jan 10 13:28:31 UTC 2006
> > Responsible-Changed-Why: 
> > To ATA maintainer.
> > 
> > http://www.freebsd.org/cgi/query-pr.cgi?pr=91410
> 
> Does it fail if you use atacontrol to create a RAID instead of gsomething ?
> 
> Does it fail if you remove PREEMPTION from the kernel config ?
> 
> -Søren