Unsolved
This post is more than 5 years old
54 Posts
0
763
January 23rd, 2007 09:00
NS702G Performance Issue......
Hello
I have an NS702G with a CX600 back-end. The Celerra is configured with one active data mover (server_2). There are 3 VDMs on server_2, each with a pair of 1Gb ethernet ports assigned (VDM3 uses cge2 & cge5 in a Link Aggregated team).
If a 1.3GB directory with 350 files is copied from an XP workstation to a share on VDM3, this can take 6 minutes. The workstation is connected at 100Mb/s FULL and it is a switched enterprise level network.
Copying the same directory to a Windows Server attached to the CX600 takes approx 2.5 minutes. I'm guessing the NAS is much busier than the Windows Server.
I can see the throughput of the file system in question spiking at 1000 IOPS, but I'd like to know how I can determine what is causing this I/O?
(I can see approx 175 users connected to the share through computer management.)
Can Celerra monitor provide this level of detail?
I know I can take a network trace on the NAS, but what do I use to view it?
In summary, the NAS appears to be running slowly, I can see lots of I/O, but can't attribute it to any one thing!
Any suggestions gratefully received.
Regards
I have an NS702G with a CX600 back-end. The Celerra is configured with one active data mover (server_2). There are 3 VDMs on server_2, each with a pair of 1Gb ethernet ports assigned (VDM3 uses cge2 & cge5 in a Link Aggregated team).
If a 1.3GB directory with 350 files is copied from an XP workstation to a share on VDM3, this can take 6 minutes. The workstation is connected at 100Mb/s FULL and it is a switched enterprise level network.
Copying the same directory to a Windows Server attached to the CX600 takes approx 2.5 minutes. I'm guessing the NAS is much busier than the Windows Server.
I can see the throughput of the file system in question spiking at 1000 IOPS, but I'd like to know how I can determine what is causing this I/O?
(I can see approx 175 users connected to the share through computer management.)
Can Celerra monitor provide this level of detail?
I know I can take a network trace on the NAS, but what do I use to view it?
In summary, the NAS appears to be running slowly, I can see lots of I/O, but can't attribute it to any one thing!
Any suggestions gratefully received.
Regards
No Events found!
Nick-4vLB0
10 Posts
1
January 26th, 2007 08:00
AFAIK Celerra Monitor is not an in-depth analysis tool. It will just tell you how many IOPS are running through the system.
We did some performance profiling prior to launching our Celerras live (we have a pair of NS502Gs) and we got really hung up on the difference in the speed between copying to a windows server and copying to the celerra (this was during our migration planning phase). We found out that the Windows servers were better on a 1 to 1 basis, but if you drew concurrent threads to the celerra i.e. multiple jobs copying to the celerra, the Windows server couldn't match the same performance. For our comparison we downloaded a tool called NetBench, and used all the PCs in the IT dept.
Hope this helps. By the way, do any of the users complain of slow performance?
TF-9l6i4
54 Posts
0
January 29th, 2007 01:00
Thanks for your reply.
Have you worked out how to use Ethereal to monitor the ports on the Celerra rather than your workstation? We are on a switched network with the Celerra on a different subnet. If I run Ethereal on my workstation, I'll only see data sent to/from my PC to the Celerra.
complain of slow performance?
Yes, users WERE complaining of performance. The NAS was restarted 2 weekends ago and since then performance has been back to normal (no complaints). It was at this point that I started comparing performance (file copy) against a single Windows server. The Windows server wins every time. So many thanks for sharing your experiences with performance profiling. This has really helped me.
EMC are investigating why the NAS had to be restarted, but so far (2 weeks into the call) only logs have been gathered.
Regards
TF
Nick-4vLB0
10 Posts
0
January 29th, 2007 05:00
With regards to the ethereal trace, you can get the Celerra to do a tcpdump using a "server_tcpdump" command which will create a logfile which you can then use Ethereal to analyse. Knowledge Base solution emc49693 details how to use the server_tcpdump command to gather a trace.
As I mentioned in my last post, the Celerra maintained it's performance when we scaled the load to multiple users, while the Windows server started better but dropped off quite rapidly. We came to the conclusion that CIFS was optimised on Windows servers (imagine that!!), while the SMB server running on the Celerra was not as efficient in a one to one comparison. I have to add, I was quite surprised that this type of benchmark comparison figure was not provided by EMC when we began to ask the performance questions you're now diagnosing.
Good luck.
Nick