Summary: Disk write speed

Hi Mangers,

Thanks all the people who responded

Many of them had Hitachi storage ( i think it's best Storage ever made till
date) had write time of 8-15 seconds ( Max) depending on the server load and
fragamentation.

Of the T3+ and 3510 many of them were faster in writing at 10-15 seconds
some of them were in minutes .

I put this puzzle in fromn to sun , they suggested me to move to switch
fabric mode insted of loop and impliment zones.

Also to group the t3+ and server in to port 1-4 5-8 etc.

We couldn't bring up the MPXIO with switch fabric in SAN due to
incomatibality with VX 3.1 , But just rearrnging the server and T3+
connectivity as suggested by sun did improve the performence by almost 50%.

Now I able create a 1Gb file in 11-15 in newly created file system and
around 30 seconds in higly fragmented file system.

******************************************************************
On my T3+ connected to E420R via GB/s fibre this takes 15.04s

Dave


Raghu,

Those times that you were experiencing are very slow indeed. Here is the 
output
from a V880 with 8 x 1050MHz USIII's. Disk Array is a fabric attached 
3510 with 146 GB Drives.

This system is a Oracle 8,9 and 10 DB Server running Solaris 9 04/04 
with SFK 4.4

I did the test 3 times as it takes the kernel a little bit of time to 
optimise itself for IO.

[root@xxxxxx scott]#> time /usr/sbin/mkfile 1024m file1

real    0m14.173s
user    0m0.100s
sys     0m8.620s
[root@xxxxxx scott]#> time /usr/sbin/mkfile 1024m file2

real    0m9.201s
user    0m0.150s
sys     0m8.450s
[root@xxxxxx scott]#> time /usr/sbin/mkfile 1024m file3

real    0m9.898s
user    0m0.030s
sys     0m9.000s
[root@xxxxxx scott]#>

Hope this helps.

Scott.

I ran the test on 3510s, mirrored with Vxvm.
# time /usr/sbin/mkfile 1024m file1

real       12.8
user        0.0
sys         9.9

12.8 seconds

Tom Jones

hope you summarize, I'd like to see how the times compare

/usr/bin/time /usr/sbin/mkfile 1024m file1

real       20.2
user        0.0
sys         8.3

It was tested on a normally busy system (not highly but also
not resting)

GB


Raghu

On a T3, raid 5, 35 to 39 seconds
On a T3, raid 0+1, 21, 26, 33, 41 seconds

Cheers

Paul


Hi there,
 
Test1 :
 
conf SAN : 2 Gb HBA ( switch macdata ) + Hitachi 9580 ( Lun 50 Go )
 
root@pla133us007 # newfs -r 10000 -f 8192 -b 8192
/dev/dsk/c5t50060E8000C28B54d13s2
root@pla133us007 # mount /dev/dsk/c5t50060E8000C28B54d13s2 /aa
root@pla133us007 # cd /aa
root@pla133us007 # timex /usr/sbin/mkfile 1024m ./test
 
real       10.29
user        0.07
sys         8.59
 
root@pla133us007 #
 
 
Good luck

Hi Raghu,

an interesting experiment - I like this kind of post.  Im inspired to
do a few myself - perhaps we could set up a benchmark site to enter
the results of some standard write test for various setups.

I would be interested to see the results on the same machines for
writing a) 100 x various 10Mb files and b) 100 x identical/partially
identical 10 Mb files.  Log files are often good examples of partially
identical files as they often include time or application stamps,

regards,

Sammy


Two things - firstly use time rather than date. Also you've left off the
file name arguement to mkfile.

That said, I've just tried this with a T3+ which is directly attached to a
V480 and is not running Veritas, but is running Solaris 9:

% time mkfile 1024m foo

real    0m12.774s
user    0m0.190s
sys     0m7.310s

****************************************************************************
*******************************************8
-----Original Message-----
From: L, Raghunath (Raghunath) 
Sent: Monday, November 28, 2005 3:33 PM
To: 'sunmanagers@sunmanagers.org'
Subject: Disk write speed


 Hi Managers,
 
 I forgot to put the file argument
 
 Please try /bin/date;/usr/sbin/mkfie 1024m file1 ;/bin/date or
	 time /usr/sbin/mkfie 1024m file1  

 Thanks to Simon Burr for pointing the mistake.
 -- 
 Sincerely,
 Raghu
 SDE Team
 <>Fix the problem, not the blame. <>
 
 	 -----Original Message-----
 	From: 	L, Raghunath (Raghunath)  
 	Sent:	Monday, November 28, 2005 1:08 PM
 	To:	'sunmanagers@sunmanagers.org'
 	Subject:	Disk write time.
 
 	Hi Managers,
 
 	I have done a small expariment with storage to check the time taken
for creating 1Gb file in 
 
 	Location 			Time taken
 
 	1) /tmp (Memory is the fastest) - 3 seconds
 	2) T3+ ( Veiritas managed Vol)- 40 seconds
 	3) T3+ ( Not Veritas Vol )-20 seconds
 	4) 3510 ( Veritas managed)-40 seconds
 	5) Cheethah 10RPM internal disk of V20Z-20 seconds
 	6) Internal disk of 4 year old E420 R- 30 seconds
 
 	T3+ and 3510 have raid five+ volumeslicing enabled.
 	T3+ has 1GB Cache and 3510 has 2GB cache
 	T3+ FC is at 1GBPS and 2GBPS for 3510 
 	qlogic swithes used here are in serial loop
 
 	If you have SAN or direct attached storage with you can you please
try this
 
 	/bin/date;/usr/sbin/mkfie 1024m;/bin/date # this will create a 1Gb
file sp please make sure you have enough space before trying this.
 
 	I am bit surprised by the time taken by T3+ and 3510 as it looks
like it's on the slower side compaed to even internal drive of the old
server.
 	-- 
 	Sincerely,
 	Raghu
 	SDE Team
 	<>Fix the problem, not the blame. <>
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers