Adding Flash cache to an existing deployment

[Josh26]

Hey guys,

I'm considering adding Flash Cache to an existing 7200, on 3.1.2 MU3.

I'm expecting the first step is to upgrade to current firmware, followed by adding in SSD (we were looking at 8 x 450GB).

The guide suggests it's a matter of a one line command to create the flash cache after this - I just wanted to check if I'm missing anything.

[Schmoog]

It pretty much is just the one line command.

I would simulate it first to see if it will help before you buy though

[Architect]

AFC is not supported on the C-MLC 480GB SSDs, so be carefull that you order the right type of SSDs.

[gsb]

This is what I am doing on my 7200:

The 7000 series will need a minimum of 4 SSD drives and will support up 768GB per node pair. So I have on order 4 of these drives: HP M6710 480GB. The drives are mirrored so the space is 480GB x 2 =980GB. So I am wasting some space. 960GB - 768GB = 192GB.

But if you have the money, you can order eight and from what I understand, the benefit would be to use the rest for SSD volume space which I can't do with only 4 SSD's.

[afidel]

You might not be wasting any space, at least under 3.1.2 the 3par uses a ton of space on a small ssd tier for spare chunklets. AFAIR it was around 1 of 8 drives worth of chunklets.

[Patrick]

This is what I am doing on my 7200:

The 7000 series will need a minimum of 4 SSD drives and will support up 768GB per node pair. So I have on order 4 of these drives: HP M6710 480GB. The drives are mirrored so the space is 480GB x 2 =980GB. So I am wasting some space. 960GB - 768GB = 192GB.

But if you have the money, you can order eight and from what I understand, the benefit would be to use the rest for SSD volume space which I can't do with only 4 SSD's.

Where are you seeing 4 drives min?

[JohnMH]

Four is the minimum per node pair for AFC on 7000 series, however if you do this then any space left over will be wasted, hence why the 480GB MLC's are the sweet spot for AFC only configurations and yes spare space is still reserved.

If you go with 8 disks or more per node pair then the remainder can be used for AO and or a Thin Dedupe tier. The 480GB cMLC's aren't supported as they don't have the endurance required for AFC's longer term use case.

As well as the 480GB MLC the 920GB MLC and 1.92TB cMLC are also supported but probably not advisable for a AFC only configurations due to the stranded space.

[gsb]

Statcache command after two hours with the new Adaptive Flash cache

Works good.

[Architect]

Interesting to see. Might you have any before and after graphs on change in host response time as well?