[ On Friday, October 1, 2004 at 15:54:04 (-0400), Dave McGuire wrote: ]
> Subject: Re: ES40, ES45, etc. support?
>
>    This is clearly a major problem.  To save yourself further difficulty 
> in this matter, I suggest you ship the Xserve RAID to me immediately.  
> I'd hate to see you putting up with all those negative numbers; I'm 
> doing this purely as a friendly service.

Hee hee!

On a clear disk I can seek forever, even if I have to go backwards!  ;-)

I can't wait to see the little blue LEDs all lit up!  Too bad it's 20
miles away through heavy weekend rush-hour traffic or I'd go right now!

Now if only I could figure out why Bonnie crashes in its final (and most
important) printf() on the Alpha....  back into gdb...

at least postmark runs correctly (or so I remember on my AS4000)....

Here are the simple block read/write tests to the pair of RAID-5 LUNs
(each LUN is on a separate controller with its own 512MB cache and a
separate FC on the Xserve RAID, and each Qlogic 2340 HBA in the Alpha is
on a separate PCI bus), just to make you cry harder....  :-)

reading with one HBA & LUN alone:

	# dd if=/dev/rsd3c of=/dev/null bs=4m count=500   
	500+0 records in
	500+0 records out
	2097152000 bytes transferred in 42.945 secs (48833438 bytes/sec)
	# dd if=/dev/rsd3c of=/dev/null bs=2m count=1000
	1000+0 records in
	1000+0 records out
	2097152000 bytes transferred in 43.177 secs (48571044 bytes/sec)
	# dd if=/dev/rsd3c of=/dev/null bs=1m count=2000  
	2000+0 records in
	2000+0 records out
	2097152000 bytes transferred in 44.035 secs (47624662 bytes/sec)
	# dd if=/dev/rsd3c of=/dev/null bs=512k count=4000
	4000+0 records in
	4000+0 records out
	2097152000 bytes transferred in 45.358 secs (46235548 bytes/sec)
	# dd if=/dev/rsd3c of=/dev/null bs=64k count=32000 
	32000+0 records in
	32000+0 records out
	2097152000 bytes transferred in 72.578 secs (28895147 bytes/sec)

(looks like it might be a good idea to try to convince the filesystem to
use at least 128k buffers insead of 64k -- or even 512k!)

and with both HBAs & LUNs reading together:

	# dd if=/dev/rsd3c of=/dev/null bs=2m count=1000 & dd if=/dev/rsd2c of=/dev/null bs=2m count=1000 &

	1000+0 records in
	1000+0 records out
	2097152000 bytes transferred in 44.301 secs (47338705 bytes/sec)

	1000+0 records in
	1000+0 records out
	2097152000 bytes transferred in 44.313 secs (47325886 bytes/sec)

one HBA & LUN writing:

	# dd if=/dev/zero of=/dev/rsd2c bs=4m count=500 
	500+0 records in
	500+0 records out
	2097152000 bytes transferred in 53.699 secs (39053837 bytes/sec)
	# dd if=/dev/zero of=/dev/rsd2c bs=2m count=1000 
	1000+0 records in
	1000+0 records out
	2097152000 bytes transferred in 53.708 secs (39047292 bytes/sec)
	# dd if=/dev/zero of=/dev/rsd2c bs=1m count=2000
	2000+0 records in
	2000+0 records out
	2097152000 bytes transferred in 54.875 secs (38216892 bytes/sec)
	# dd if=/dev/zero of=/dev/rsd2c bs=512k count=4000 
	4000+0 records in
	4000+0 records out
	2097152000 bytes transferred in 55.784 secs (37594148 bytes/sec)
	# dd if=/dev/zero of=/dev/rsd2c bs=64k count=32000 
	32000+0 records in
	32000+0 records out
	2097152000 bytes transferred in 81.059 secs (25871920 bytes/sec)

(yes, it does seem to be a good idea to use larger buffers...)

and with both HBAs & LUNs writing together:

	# dd if=/dev/zero of=/dev/rsd2c bs=2m count=1000 & dd if=/dev/zero of=/dev/rsd2c bs=2m count=1000 &

	1000+0 records in
	1000+0 records out
	2097152000 bytes transferred in 63.145 secs (33211687 bytes/sec)

	1000+0 records in
	1000+0 records out
	2097152000 bytes transferred in 63.138 secs (33215369 bytes/sec)

and from one to the other:

	# dd if=/dev/rsd3c of=/dev/rsd2c bs=2m count=1000
	1000+0 records in
	1000+0 records out
	2097152000 bytes transferred in 84.246 secs (24893193 bytes/sec)

keeping in mind that this is RAID-5, and a clear, full, 1.4TB/LUN each,
it's not bad, not bad at all, even if I do say so myself!  ;-)

(though I wonder why that last read/write test is so slow...)

postmark will the the real test though -- gotta see those transactions
per second!

and I suppose just for fun we need to try a RAID-1+0 config too -- it'll
be a hell of a waste of space but it may just scream too!

-- 
						Greg A. Woods
						Planix, Inc.

<woods@planix.com>     +1 416 489-5852 x122     http://www.planix.com/