Re: [storage-discuss] Seriously degraded SAS multipathing performance.

2010-07-02 Thread Marion Hakanson
[email protected] said:
> Very glad to hear you've made progress. I'm going to try the logical-block
> load balance setting on my workstation at home. It'll be an interesting
> experiment I'm sure! 

Multipath on your home workstation?  My teenager might say, "Dude!".
Well, if he was the least bit interested anyway (:-).

I've managed to throw together a little summary of my test sessions
from yesterday.  I found it was quite worthwhile to make a tiny little
filebench profile instead of messing around with "dd" and "time" on
the command-line, and trying to keep track of all the results.

http://acc.ohsu.edu/~hakansom/md1200_loadbal_bench.html

Regards,

Marion


___
storage-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/storage-discuss


Re: [storage-discuss] Seriously degraded SAS multipathing performance.

2010-07-02 Thread James C. McPherson

On  3/07/10 02:48 AM, Marion Hakanson wrote:

[email protected] said:

different driver -->  different bug.

Josh's bug is 6963122.

You should file a new one - bugs.opensolaris.org, Product Solaris, Category
kernel, Subcategory driver-mpt-x86.

I suggest looking at the public comments field for Josh's bug first, though -
the RE for that bug has some D you can run to help narrow down the
performance problem point.


If I interpret the bug report correctly, it's been closed, having been
diagnosed as a function of HBA or drive firmware, not a driver issue.


Yes, that's correct.


I went looking for an easier workaround than manually disabling every
drive's second path using "mpathadm disable path ...", and stumbled
across "load-balance=logical-block" as a possibility.

   http://sunsolve.sun.com/search/document.do?assetkey=1-61-206894-1
   http://sunsolve.sun.com/search/document.do?assetkey=1-61-202693-1

I gather that mpxio with "load-balance=round-robin" makes these drives
go slow because the drive waits for one revolution before it can switch
ports/paths to handle next command from the host.


That's correct.


Probably does away
with the drives command-queue, too.


Pretty sure that's not the case.


Spent a good chunk of yesterday testing a number of different region-size
settings with "load-balance=logical-block", finding that basically anything
is much better than the default "load-balance=round-robin", but nothing is
as good as "load-balance=none".


That's really good to know, thankyou.


Best performance of all was after manually assigning half the drives to
use one path, and half to use the other path.  I now find myself wishing
there was a "load-balance=" setting which would do this
automatically.


If you log an RFE for it (solaris/kernel/io-multipath) then there's
a decent chance it could get implemented :-)


I speculate that this is what the LSI RAID-controllers
must do to provide their redundant-path load-balance functionality.


No idea on this, sorry.


Very glad to hear you've made progress. I'm going to try the
logical-block load balance setting on my workstation at home.
It'll be an interesting experiment I'm sure!


best regards,
James
--
Senior Software Engineer, Solaris
Oracle
http://www.jmcp.homeunix.com/blog
___
storage-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/storage-discuss


Re: [storage-discuss] Seriously degraded SAS multipathing performance.

2010-07-02 Thread Marion Hakanson
[email protected] said:
> different driver --> different bug.
> 
> Josh's bug is 6963122.
> 
> You should file a new one - bugs.opensolaris.org, Product Solaris, Category
> kernel, Subcategory driver-mpt-x86.
> 
> I suggest looking at the public comments field for Josh's bug first, though -
> the RE for that bug has some D you can run to help narrow down the
> performance problem point. 

If I interpret the bug report correctly, it's been closed, having been
diagnosed as a function of HBA or drive firmware, not a driver issue.

I went looking for an easier workaround than manually disabling every
drive's second path using "mpathadm disable path ...", and stumbled
across "load-balance=logical-block" as a possibility.

  http://sunsolve.sun.com/search/document.do?assetkey=1-61-206894-1
  http://sunsolve.sun.com/search/document.do?assetkey=1-61-202693-1

I gather that mpxio with "load-balance=round-robin" makes these drives
go slow because the drive waits for one revolution before it can switch
ports/paths to handle next command from the host.  Probably does away
with the drives command-queue, too.

Spent a good chunk of yesterday testing a number of different region-size
settings with "load-balance=logical-block", finding that basically anything
is much better than the default "load-balance=round-robin", but nothing is
as good as "load-balance=none".

Best performance of all was after manually assigning half the drives to
use one path, and half to use the other path.  I now find myself wishing
there was a "load-balance=" setting which would do this
automatically.  I speculate that this is what the LSI RAID-controllers
must do to provide their redundant-path load-balance functionality.

Regards,

Marion


___
storage-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/storage-discuss


Re: [storage-discuss] Seriously degraded SAS multipathing performance.

2010-07-01 Thread James C. McPherson

On  2/07/10 06:49 AM, Marion Hakanson wrote:

On 06/21/2010 10:06 AM, Josh Simon wrote:

I'm seeing seriously degraded performance with round-robin SAS
multipathing. I'm hoping you guys can help me achieve full throughput
across both paths...


[email protected] said:

Hi Josh, I suggest logging a bug on bugs.opensolaris.org, using solaris /
driver / mptsas. Please include the info you sent in the earlier message too.


Has this bug turned up in the database yet?  I want to file a "me too".
Except for us, this is Solaris-10U8 (not snv_134), and it's the mpt driver
(not mpt_sas).


Hi Marion,
different driver --> different bug.

Josh's bug is 6963122.

You should file a new one - bugs.opensolaris.org, Product Solaris,
Category kernel, Subcategory driver-mpt-x86.

I suggest looking at the public comments field for Josh's bug first,
though - the RE for that bug has some D you can run to help narrow
down the performance problem point.


cheers,
James




After seeing this problem report, I tried turning off the 2nd path on each
of the 2TB "Near-Line SAS" drives in our Dell MD1200 array.  Sequential
write performance went up by 4.25 times.  Making the same change to
an identical server with a J4400 and 1TB SATA drives resulted in only an
incremental improvement in bandwidth:

filebench fileio, multistreamwrite-1M:

MD1200, 2TB SAS, mpxio round-robin:131 MB/sec
MD1200, 2TB SAS, mpxio static single:  557 MB/sec

J4400, 1TB SATA, mpxio round-robin:427 MB/sec
J4400, 1TB SATA, mpxio static single:  498 MB/sec


Note that on the MD1200, with round-robin in effect, the drives were
showing 100%-busy with service-times in the 300 msec range.  With the
static single-path in effect, service-times dropped to the 30 msec range.

The above numbers were observed on this type of pool layout (both systems):

 NAME   STATE READ WRITE CKSUM
 f1b_zp1ONLINE   0 0 0
   raidz2   ONLINE   0 0 0
 c5t5000C50014D641DDd0  ONLINE   0 0 0
 c5t5000C50014D55B13d0  ONLINE   0 0 0
 c5t5000C50014D5B9B4d0  ONLINE   0 0 0
 c5t5000C50014D6EF76d0  ONLINE   0 0 0
 c5t5000C50014D07BC2d0  ONLINE   0 0 0
 c5t5000C50014D6DBA5d0  ONLINE   0 0 0
   raidz2   ONLINE   0 0 0
 c5t5000C50014D6CCB2d0  ONLINE   0 0 0
 c5t5000C50014D6A126d0  ONLINE   0 0 0
 c5t5000C50014D641BBd0  ONLINE   0 0 0
 c5t5000C50014D65228d0  ONLINE   0 0 0
 c5t5000C50014D6C530d0  ONLINE   0 0 0
 c5t5000C50014D6E35Ed0  ONLINE   0 0 0


Regards,

Marion





--
James C. McPherson
--
Senior Software Engineer, Solaris
Oracle
http://www.jmcp.homeunix.com/blog
___
storage-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/storage-discuss


Re: [storage-discuss] Seriously degraded SAS multipathing performance.

2010-07-01 Thread Marion Hakanson
> On 06/21/2010 10:06 AM, Josh Simon wrote:
>> I'm seeing seriously degraded performance with round-robin SAS
>> multipathing. I'm hoping you guys can help me achieve full throughput
>> across both paths...

[email protected] said:
> Hi Josh, I suggest logging a bug on bugs.opensolaris.org, using solaris /
> driver / mptsas. Please include the info you sent in the earlier message too.

Has this bug turned up in the database yet?  I want to file a "me too".
Except for us, this is Solaris-10U8 (not snv_134), and it's the mpt driver
(not mpt_sas).

After seeing this problem report, I tried turning off the 2nd path on each
of the 2TB "Near-Line SAS" drives in our Dell MD1200 array.  Sequential
write performance went up by 4.25 times.  Making the same change to
an identical server with a J4400 and 1TB SATA drives resulted in only an
incremental improvement in bandwidth:

filebench fileio, multistreamwrite-1M:

MD1200, 2TB SAS, mpxio round-robin:131 MB/sec
MD1200, 2TB SAS, mpxio static single:  557 MB/sec

J4400, 1TB SATA, mpxio round-robin:427 MB/sec
J4400, 1TB SATA, mpxio static single:  498 MB/sec


Note that on the MD1200, with round-robin in effect, the drives were
showing 100%-busy with service-times in the 300 msec range.  With the
static single-path in effect, service-times dropped to the 30 msec range.

The above numbers were observed on this type of pool layout (both systems):

NAME   STATE READ WRITE CKSUM
f1b_zp1ONLINE   0 0 0
  raidz2   ONLINE   0 0 0
c5t5000C50014D641DDd0  ONLINE   0 0 0
c5t5000C50014D55B13d0  ONLINE   0 0 0
c5t5000C50014D5B9B4d0  ONLINE   0 0 0
c5t5000C50014D6EF76d0  ONLINE   0 0 0
c5t5000C50014D07BC2d0  ONLINE   0 0 0
c5t5000C50014D6DBA5d0  ONLINE   0 0 0
  raidz2   ONLINE   0 0 0
c5t5000C50014D6CCB2d0  ONLINE   0 0 0
c5t5000C50014D6A126d0  ONLINE   0 0 0
c5t5000C50014D641BBd0  ONLINE   0 0 0
c5t5000C50014D65228d0  ONLINE   0 0 0
c5t5000C50014D6C530d0  ONLINE   0 0 0
c5t5000C50014D6E35Ed0  ONLINE   0 0 0


Regards,

Marion


___
storage-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/storage-discuss


Re: [storage-discuss] Seriously degraded SAS multipathing performance.

2010-06-21 Thread James C. McPherson

On 22/06/10 12:55 AM, Josh Simon wrote:

I should also mention that I am running snv_134.

Thanks!

Josh Simon

On 06/21/2010 10:06 AM, Josh Simon wrote:

I'm seeing seriously degraded performance with round-robin SAS
multipathing. I'm hoping you guys can help me achieve full throughput
across both paths...



Hi Josh,
I suggest logging a bug on bugs.opensolaris.org, using
solaris / driver / mptsas. Please include the info you
sent in the earlier message too.


thankyou,
James C. McPherson
--
Senior Software Engineer, Solaris
Oracle
http://www.jmcp.homeunix.com/blog
___
storage-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/storage-discuss


Re: [storage-discuss] Seriously degraded SAS multipathing performance.

2010-06-21 Thread Josh Simon

I should also mention that I am running snv_134.

Thanks!

Josh Simon

On 06/21/2010 10:06 AM, Josh Simon wrote:

I'm seeing seriously degraded performance with round-robin SAS
multipathing. I'm hoping you guys can help me achieve full throughput
across both paths...

___
storage-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/storage-discuss