MSA1500cs, VMs and performance questions

[Kris Thornton]

I am doing planning for a possible expansion of our MSA1500cs for next year. Currently, the MSA1500 only has one MSA30 attached and is used solely for disk to disk backups. Next year I would like to add two MSA20s, a 2nd MSA1000 controller and a second FC switch and start using the SAN to store the boot and data drives for virtual machines that boot from the SAN. There will be about 10 VMs that all have pretty low IO usage except for one SQL Server and an Exchange Server. Because one of the goals of this project is to speed up the performance of the Exchange Server, the MSA20s will have 6 * 15K 300GB HDs in them and I will create a RAID1+0 array for the logical drives to be held on. There will be an additional 2 * 15K 300GB HDs in each MSA20 that will have a RAID6 array on it and will be used for the boot LUN for each of the VMs.

So, my questions are:

1. Is there a formula I can use to approximate what the IO usage is going to be with all of this new traffic going to and from the SAN? What are the theoretical limits of the MSA1500?

2. Best practice for SQL and Exchange is that transaction logs are kept on separate physical drives from the data. Is it enough to create a separate logical drive on the RAID1+0 array for the transaction logs to go on or should I have a 2nd RAID1+0 array that is just used for the log file logical drives?

Thanks,
Kris

[raadek · Accepted solution]

Hi Kris,

Re 1. - you may find this thread useful:
http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=1066572

Re 2. - at least in the MSA world you really should have a separate array (i.e. a bunch of disks) for your logs; this is becuse logs generate mostly sequential I/O writes requests, while data tables are accessed with a random pattern.

Rgds.

[Martin Smoral]

Kris,
for perfromance and reliability, I would reccomend using the MSA30's(SCSI disks) for your SQL/Exchange/VM's and use the MSA20's (SATA) for your disk to disk backups

from the quick specs:
Performance - The MSA1500 cs and its components have been designed using the latest in standard 2 Gb interconnects and Fibre Channel technology. Performance is rated at a transmission rate of up to 30,000 IOPs from cache and bandwidth of up to 200 MB per second. Performance can be substantially impacted by number of users and type of application. With dual controller active/active configurations, performance increase to >43,000 IOP/s from cache and 367MB/s throughput.

Regards,
Marty

[calgbid]

We configured our MSA1500cs in Active/Active and tested between one to five blades running Windows 2003 Server, including running MPIO software. We test using IOMeter with 128kb (or larger), 100% sequential, 100% read.

After some tweaks, e.g. locking HBA ports to 2Gb to match 2Gb FC to controller, setting F port type, running MPIO software in the client, updating all firmware, increasing our MSA cache to 512MB per controller, we get a maximum of 240MB/sec throughput with Active/Active on the MSA1500cs.

We are interested in the exact configuration of anyone with more than 240MB/sec, or close to the 367MB/sec quoted, throughput.

Our blades are BL460c and our disks are 28x U320 300GB SCSI 10K. We have dual StorageWorks 4/8 SAN switches, each with one 2Gbit FC to the SCSI controller.

Our three LUN are configured as (1) 9x disk R5 64kb slit across two shelves, (2) 9x disk R5 64kb slit across two shelves, (3) 8x disk R0 128kb split across two shelves. We found this LUN configuration optimal for our hardware for testing read throughput.

We request suggestions to improve our total throughput. Our controller CPU during read testing is #1=48% and #2=25%. Our switch performance says it is under-utilized, the 2Gb FC to disks is at 36% on the busy switch.

[calgbid]

Forget answering my last posted question.
We figured out why our limit was 240MB/sec throughput with the MSA1500cs.

Our disks are 10k limiting the maximum MSA1000/MSA1500cs throughput, e.g. 10k/15k = 66%, and 240MBsec/367MBsec = 66%.

The lesson learned is full suggested HP performance from the MSA1000/MSA1500cs requires use of 15k SCSI disks in each of our MSA30 chassis.

I wish this was mentioned in the Quick Specs or simply more obviously stated elsewhere. Maybe the source of the problem is caused by an outdated controller, such that 15k SCSI disks are required to achieve 367MB/sec maximum throughput. It seemed to me, but was shown to be false, that 28x 10k U320 disks might achieve 367MB/sec. Now we know.