VSA Performance with SSD tier & SmartCache/CacheCade?

[Paul Hutchings]

Does anyone have any real-world performance they'd care to share on using StoreVirtual VSA with a good sized SSD tier and perhaps adding an SSD backed RAID cache (SmartCache/CacheCade) please?

We use StoreVirtual right now (4300/4500 G2) and whilst the feature set is a bit lacking it's been flawless in terms of uptime so when it's replaced one option is two or four boxes running VSA doing combined storage/compute.

I'm still not entirely clear whether I can "force" a LUN to SSD (i.e. SQL databases) or if all I can do is enable tiering and hope/assume it will mostly be tiered?

Looking at something like a 16 or 20 bay dual E5-26xx box with 4x SSDs in RAID5 and 7.2K SAS for the rest in hardware RAID10.

Also I'd be very interested in thoughts on 2 nodes vs. 4 in a two-site metro-cluster with a dedicated third site running the FOM.

[a_o]

YMMV! 

These are real world numbers that I took from my ever evolving setup. As you can see the numbers are all over the place.
My take is this: I've so far found that VSA 2014 with AO does not offer the same performance as using SSDs only as the underlying storage for a VSA. It might have to do with the amount of data that I currently have on the AO enabled nodes.
Maybe the AO cache hit ratios are too low?

My storage is mostly used for VM boots, SQL and Exch.  Using most HP Enterprise SSD and BL IO (fusion IO) cards, and HP 6GBs SAS.
VSAs are  on Proliant BLC blades with 10GbE interconnects w. 4Gbs Max dedicated to the SAN Vnet in VC. I've thrown in a direct attached SSD metric for reference.
Note that the formatted block size for the LUNs will naturally affect the performance of the LUNs. I've included 4K and 8K metrics.

QD - 4                               4K Write    4K Read    8K Write    8K Read
Direct attach MLC SSD R1 (No VSA)    230MB/sec   360MB/sec  240MB/sec   354MB/sec
iscsi VSA 10.5     SSD MLC  R5       62MB/sec    138MB/sec  74MB/sec    160MB/sec
iscsi VSA 10.5 SSD FusionIO  R0     226MB/sec    200MB/sec  242MB/sec   250MB/sec
iscsi VSA 10.5     SATA  R5          90MB/sec    182MB/sec  72MB/sec   149MB/sec
iscsi  P4300G2 V11 (1GbNIC) R5       62MB/sec    117MB/sec   62/MB/sec  117MB/sec
iscsi VSA2014 w/AO SSD.MLC/SAS.R5   102MB/sec    219MB/sec  145/MB/sec  249MB/sec
iscsi VSA2014 w/AO Fusion.IO/SAS R5 125MB/sec    190MB/sec  132MB/sec   257MB/sec

--Edited to correct transposed row.

[Paul Hutchings]

Thanks, be interested in knowing a little more on the disk subsystem being used?

Also do you have any IOPS numbers please?

[a_o]

WRT to the disk subsystem, it's mostly HP D2700 and D2600 SAS shelves connected to 2 x 2 SAS Switches. Zoning is done as groups of disks on the shelves. For example on one D2600 there are 3 zones :
1, 5 x 800GB MLC SSD
2, 10 x 600GB 10KSAS 6Gbps
3, 10 x 1TB 7.2K SAS 6Gbps.

Additionally, there are 4 SB2200 storage blades with a variety of SSDs and SAS drives.

On the compute blades that are used for storage, each one has HP P711m BL raid controller and one BL 640GB IO Accelerator.
The SB2200 blades have a P410i Raid controller.

I try to keep everything HP because of the SmartSSD Wear Gauge support that was recently introduced in ACU.
IMO, this is arequirement for using SSDs in a production environment.

[HPstorageTom]

You cannot force a LUN to SSD. The only option you do have is to enable (that's the default) or disable AO for StoreVirtual volumes.

[Paul Hutchings]

Thought that was the case sadly.

It's stil up there.  My biggest concern right now is restripe times - the 50TB VSA licenses are a great idea because we could get the capacity in 2 boxes (with hardware RAID10 for performance) but the restripe time should I lose a box concerns me a lot.