Richard;

Firstly, let me say that I welcome this discussion.

I was searching for comments and alternative viewpoints in the first place and at the very least, people in the list and myself can learn and understand the Solaris IO parameters better.

I did not make the assumption that maxphys overrides sd/ssd.conf max_xfer_size. From the beginning I believe, you MAY have misunderstood my email.

My objective is simple. Fabric attached high end storages like EMC have stripe width around 8 Mbyte. Can we not match the SCSI transfer size to be in or around the same size?

The original intention is to match the SCSI transfer size to the stripe width of EMC arrays and other similar Fabric attached H/W RAID arrays.

I wanted to set maxphys=8388608 (8 Mbyte).

I fully realized that by setting maxphys=8388608 and NOT changing sd/ssd.conf, the limit will still be determine by sd/ssd.conf maximum transfer size, which is 1 Mbyte.

Thus I also suggested changing sd/ssd.conf to a higher value. Together with relevant vxio and md parameters.

It is my impression that by changing the above values affects only MAXIMUM transfer size and will not affect IO which have a smaller data payload.

Am I wrong? At this point, I am unsure.

Examples of articles which recommends setting maxphys to 8388608 are

http://www.samag.com/documents/s=7667/sam0213b/0213b.htm
http://www.samag.com/documents/s=7610/sam0210j/0210j.htm
http://www.pluzzi.com/pluzzi/os/perf_tuning/solaris_8_performance_tuning.html
http://www.schuba.com/pub/courses/it454-ss2003/handouts/solaris_internals.pdf

Examples of articles which recommends setting maxphys to 2 Mbyte are;

http://people.ee.ethz.ch/~oetiker/tobjour/2003-02-12-09-00-f-1.html

All of the above articles also suggest setting the respective sd/ssd.conf and md and vxio to more appropriate values to take advantage of the larger maxphys.

I would love an "unofficial" non binding, in Good Faith answer from Sun with respect to the above.

Warmest Regards
Steven Sim

Richard Elling wrote:

Richard;

I do believe I mention in my original email that the
default sd and ssd xfer is 1 Mbyte.

The problem, as I said, lies in the fact that maxphys is default 128 Mbyte which does not make sense especially since the sd/ssd driver max_xfer_size is 1 Mbyte. (that includes ufs also!)

The limit of size of transfers is, for anything interesting, 1 Mbyte.
Perhaps you are making the assumption that maxphys overrides
the maximum transfer size dictated by the driver.

In almost all references to maxphys I have come across, almost all have recommended that maxphys be set to 8388608 bytes as this improves large transfers without affecting small IO.

First, write applications which make such large writes.  The largest
write by a commercial application (I'm aware of) is 1 Mbyte.

I have a simple observation.

Solaris is an extremely powerful and scalabel OS but out of the box, the IO settings is not appropriate for today's fibre attached storages.

Even setting the max transfer size to 1 PByte won't make any
difference until the applications do such large writes.  But even
if you set it to 8 Mbytes, the protocol overhead per transfer is
so small that the performance gains will be barely noticeable.
In other words, AFAICT the current default has solved this particular problem.

If properly tuned however, I do believe that it can outperform most other general purpose OS.

We're trying :-)  There is more fertile ground elsewhere.

Solaris 10 attempts to change this by removing many /etc/system parameters. But for Solaris < 10......

Water under the bridge.
-- richard
This message posted from opensolaris.org
_______________________________________________
perf-discuss mailing list
perf-discuss@opensolaris.org






Fujitsu Asia Pte. Ltd.
_____________________________________________________

This e-mail is confidential and may also be privileged. If you are not the intended recipient, please notify us immediately. You should not copy or use it for any purpose, nor disclose its contents to any other person.
Opinions, conclusions and other information in this message that do not relate 
to the official business of my firm shall be understood as neither given nor 
endorsed by it.


_______________________________________________
perf-discuss mailing list
perf-discuss@opensolaris.org

Reply via email to