StoreVirtual Storage
cancel
Showing results for 
Search instead for 
Did you mean: 

Typical LeftHand Performance?

UselessUser
Frequent Advisor

Typical LeftHand Performance?

Hi,

As mentioned in a previous query I have a 2 node 14.4TB Virtualization P4500.

I am having issues - in particular with Backup Exec performance (For which I have posted a query with them) so I am just wondering what typical performance I can expect from this SAN...

For example at the moment it is running a job from a VMware guest (On the SAN) to a Backup to disk folder (Also on the SAN) via the production network (Treating it like a physical server), the throughput rate on the job is approximately 50MB/s. This I can kind of get my head around because I am reading a file on the SAN, only to write it again on the same SAN.

The verify rate is approximately the same performance which is odd, because I am no longer driving the SAN from two different locations, it is no longer reading the source, so I am wondering if it is a Windows Server 2008 R2 config issue?

The problem with this is I cannot easily add up all throughput going to this SAN... but I was kind of expecting better? (This is a new environment with a few VMware hosts but not much IO)

The backup server is a Server 2008 R2 64 with 2 dedicated iSCSI 1 GB NIC's running LeftHand MPIO driver (So only one link is actually active...)

Any thoughts?
2 REPLIES
Jim Silvia
Advisor

Re: Typical LeftHand Performance?

Most performance problems are network related.

50 MB/s or 400 Mbps isn't horrible.

I've had network problems that dropped those numbers by a factor of 10.

That being said, running backup exec jobs on our exchange 2007 VM to a separate SAN cluster i can see upwards of 800 Mbps.

One quick thing to check - the log files for each node - check the hist.ifconfig.log. In one of our client environments that showed an increasing # of nic errors that we weren't seeing at the switch level. I moved the cat6 modules to new ports and the issues went away and performance started flying (close top 950 Mbps from a robocopy on a Windows 2003 machine using MS ISCSI).

I run with ALB and flow control without jumbo packets.

Check out the following white paper - they were key to helping me optimize several environments

http://bizsupport1.austin.hp.com/bc/docs/support/SupportManual/c01750150/c01750150.pdf

Also check the patches you are running. The earlier 8.5 builds had patches for the NIC that addressed buffer underruns.

Jim
Marcel-p4500
Advisor

Re: Typical LeftHand Performance?

Hi There,

Per SAS drive(15K) I use to calculate: 150 IOPS, so in a P4500 SAN(2 nodes) u get 24 x 150 = 3600.

3 nodes makes 36 x 150 = 5400
4 nodes makes 48 x 150 = 7200

Did several tests and can confirm that this is the physical limit of the spindles(correct me if i am wrong please).

The only thing that is kinda vague are the cache-only-IOPS, have not found what this exactly means, i think it has to do with snapshots and clones, but not 100% sure.

And remember: every setup has it's own configuration, so the IOPS depends on the configuration that is being used.

Hope this helps a bit...