San HQ 2.0 Reports

Hello,
That's what the "Experimental Analysis" feature is for. It helps you determine based on your IO pattern what performance you are getting against what that group can do. Since all the pieces you mentioned directly relate to how the array performs for you. Smaller IOs yield more IOPs but less throughput. So direct comparisons aren't practical, since your IO pattern might be very different from someone else's.

Latencies must be considered in tandem with the average I/O size. Larger I/O’s intuitively take longer to process. The following rules of thumb apply to 16 kilobyte or less average I/O sizes:

In general, average latencies of less than 20ms are considered to be acceptable.

Average latencies between 20ms and 50ms should be considered cautionary. If such levels are reported constantly over time, the administrator may wish to either reduce the workload or add additional storage resources to handle it.

Any average latency of 80ms or more should be considered very problematic, especially if sustained over time. Most enterprise applications will begin to show problems if latencies exceed 100ms. In these scenarios, administrators should either reduce the workload or add additional storage resources.

Average latency Implied load
(16 KB I/Os or less)
-------------------------------------------------------------------------------------------
under 20ms -------------------> Normal operations
20ms to 50ms ----------------> Cautionary if sustained
50ms to 80ms ----------------> Warning level, suggesting overload if sustained
over 80ms ----------------------> Very problematic, suggesting overload if sustained

Regards,
-don

Additionally, there is a nice PDF on how to best use SANHQ


http://www.equallogic.com/resourcecenter/assetview.aspx?id=8749

It's only on the Equallogic Website under Tech Reports. However, that requires a login. The link above does not.

-don

Great information Don. The PDF is extremely helpful. Our average I/O KB is around 10-16KB, but certain hours they spike up to 20-25KB. Interesting, I want to track down what that is. My average latency is still below 20, but does hit 20ms when backups start running overnight. We're pretty impressed with the Equalogic, for the past year. We've thrown everything at this 6500E, I think it's the E, 48 x 1TB drives. Well over 100 vmware virtual servers. With these numbers it'll help to validate getting a faster unit like a 6000v or vx to offload the oracle databases we run on it.

--eapelin
HPU

You are very welcome. Don't be afraid to open a support case too. Between SANHQ and the array diags much about how the array has been performing can be examined. Backups are all about reads, if the blocksize is also larger during that time the latency isn't a problem. You should also see that the MB/sec rates goes way up. Much depends on the RAID level you are running as well. RAID-5/6 is going to perform less in such a random, small write environment like ESX.

-don

We're running raid 50, 48 1TB drives we get ~32GB total storage from the pool. After a year we're at 10GB free on it. I came across the EqualLogic Tech Reports page this morning, looks like lots of great info on best practices. I don't know how diverse a group reads this forum, but I'm curious if any one else has gone the route we have and used a PS as their storage for their virtualized environment? We have over 150 virtualized (vsphere4) servers on a single PS6500E. The physical servers running vsphere are r710's with lots of memory (48GB each server) with 4x Gb ethernet ports (NO HBAs). It's really a simple setup that's given us the greatest bang for the buck. We've thrown everything at the PS and it's still going strong. Exchange 2007, 4 oracle database, mssql databases, ms file servers, web servers, all part of those 150 virtualzed servers.

eapelin,
We don't have quite as many VMs as you, but we're doing the same thing. We have a PS100, PS300, PS5000x and a PS6000xv running RAID 50. We run vsphere on 9 x Dell PE1955. Other than oracle, we run the same virtual servers you mentioned. We've been extremely happy with the Equallogic performance. Just waiting for my new M1000e chassis, M910 blades and Equallogic 6510X to arrive.

I am running 50 Hyper-V VMs on two PS400E in a single group with both members running RAID 50. File, print, SQL, Exchange, BlackBerry, Active Directory, etc are all running as VMs. Older PE1950 with 16 GB to R710 with 72 GB. Performance is pretty good unless I am trying to reboot all of my VMs at once (think EqualLogic firmware upgrade). I need to work on my boot everything process, because when I try to boot too many at a time, a boot that should take 3 minutes can drag on for maybe 20+ minutes. I am sure it would be even worse if I truly tried to boot everything at once.

We manualy start our servers and do about 2-4 VM's per host. Tedious but , we found out the hard way years ago the same thing you ran into khsieh. Booting too many at one time brings everything down to a crawl. Luckily we've only had to do that when there were island wide power outages.