RE: Orion, NFS, & File Size

  • From: "CRISLER, JON A" <JC1706@xxxxxxx>
  • To: Austin Hackett <hacketta_57@xxxxxxx>
  • Date: Wed, 20 Jun 2012 18:16:21 +0000

What is the content of your lun file?  These results seem in line with 1 thread 
on a 1 gig enet link- do you have all of the datafiles in your config file?
-----Original Message-----
From: Austin Hackett [mailto:hacketta_57@xxxxxxx] 
Sent: Wednesday, June 20, 2012 1:50 PM
To: CRISLER, JON A
Cc: Austin Hackett; Oracle-L@xxxxxxxxxxxxx
Subject: Re: Orion, NFS, & File Size

Hi Jon
 
I’ve been playing with Orion today on old test machine and filer that’s 
available – just to get familiar with it.
 
I notice in the Orion manual it says:
 
“You may see poor performance doing asynchronous I/O over NFS on Linux 
(including 2.6 kernels)”
 
I’ve created a single 5GB file
 
dd if=/dev/zero of=/orion_test/test1.txt bs=32k count= 163840
 
Listed the file in orion.lun and then done an ./orion –run oltp
 
Doing this test, I got the following results – here’s the iops.csv file 
transposed:
 
1              224
2              232
3              241
4              232
5              241
6              249
7              255
8              259
9              266
10           269
11           266
12           270
13           282
14           294
15           290
16           294
17           306
18           308
19           319
20           317
 
At first I wondered whether I had enough files to drive a decent load. So I 
created a total of 5 x 5GB files and listed these in orion.lun. By tailing the 
orion.trace file, I could see if wasn’t getting any better results. I then 
created a total of 10 files, and tried again, but still the same. Finally I 
tried running with –num_disks 24, but no real differences…
 
Whilst my Orion tests we’re running, our Storage Admin ran some tests using FIO 
on the same volume and was seeing thousands of random read IOPS as expected, so 
the issue doesn’t seem to be the filer.
 
Am I failing to grasp something about how Orion should be used here, or am I 
hitting the bug? My current thinking is it’s the bug, but it would be great to 
get your thoughts if you have the time, given you’ve used it successfully with 
NetApp. It would be a shame for me to discount Orion if I’m just not using it 
properly…

On 20 Jun 2012, at 17:12, "CRISLER, JON A" <JC1706@xxxxxxx> wrote:

> Most of my installs are medium to huge RAC builds, so a 1gb Ethernet path 
> would be a serious bottleneck, but it all depends on your workload.  With a 
> 10 gb enet The NetApp should be able to deliver 2 ms or less average i/o 
> response time, same with 4gb fiber channel.
> 
--
//www.freelists.org/webpage/oracle-l

Other related posts:

  1. Home
  2. oracle-l
  3. Archive
  4. 06-2012