FAST Cache with new writes?

Question

Hello,

We recently implemented FAST and FAST Cache with some new SATA drives. Write performance of new data (for example storage vMotion of VM from old Raid Group to new disk pool) to SATA drives is very poor, and it seems like FAST Cache is doing nothing to improve write performance.

Can someone confirm how FAST Cache actually works for writes, I think I may have understood it wrong.

Are all writes committed from DRAM cache to FAST Cache and from there slowly to drives?

Or are writes committed to FAST Cache only when destination block has been previously promoted to FAST Cache based on read activity of that block?

Thanks!

Storagesavvy · Accepted Answer

Writes only hit FASTCache when the blocks have previously been promoted to FASTCache due to read activity.

It may also take some time for FASTCache to warm up (ie: a couple days) after which performance may improve. FASTCache will alleviate backend IO to the disk, freeing those disks to handle IO that is not hitting FASTCache.

If you're primary performance problem is writes, and the data you are writing is not being promoted because you are not reading those blocks, then you may be better off using the EFDs with FASTVP, then setting the LUN tiering to "Highest Allocation". New writes will hit highest tier as long as there is free space in that tier. FASTVP keeps 10% of the tier space free for new writes after each reallocation has occurred.

thakala · Answer

Richard, thank you very much making this clear for me. Is there any way to set 'Highest Available Tier' option being default for new pool LUNs?

Storagesavvy · Answer

I don't think you can set it as a default, you'd have to set it manually during or after creation of the LUN.

You may want to see how FASTCache performs over time. For example, if Storage vMotion is slow but actual VM performance is great, then FASTCache may still be the best use of the EFD disks. Once you put the EFDs into the FASTVP pool, you can't remove them to my knowledge. And the write performance using Highest Allocation will be dependent on the amount of new writes between FASTVP reallocation versus the amount of free space in the EFD tier (10% of the EFD space). If you write more data than can fit in the 10%, writes will start hitting SATA until the next reallocation process completes.

Are you using VAAI (FLARE30 and vSphere 4.1?) VAAI significantly changes the storage vmotion process since the array copies the blocks on the back end rather than the host writing the new blocks.

thakala · Answer

Yes we are using vSphere 4.1 and VAAI is enabled. I think FAST Cache is still best use for those EFDs, we have FC drives also so I just have to make sure all new LUNs are configured with 'Highest Available Tier' option set so all new writes hit fast FC and not slow SATA.

Kumar_A · Answer

Correction in the previous couple of posts: FAST Cache promotion happens in response to reads AND writes to the same block in the backend. If you do ANY kind of I/O (read and/or write) to the same 64KB chunk in the backend, it will get promoted into FAST Cache and from that point on, you will be accessing the data from flash drives, rather than the mechanical drives.

Another point to consider is that that FAST Cache takes some time to "warm up". What this means is that you will need to wait for some time during which enough blocks get promoted into FAST Cache, after which you will see appreciable performance difference. The time it takes to warm up depends on the size of your LUN on which FAST Cache is working.

Do you have an idea of your read/write ratio? The higher the read component, the more benefit you can expect to see from FAST Cache.

thakala · Answer

Yep, I understand that consequent writes to same block will promote data to FAST Cache just as reads do. But my point was that FAST Cache is not buffering ALL writes like DRAM write cache does, this was a little disappointment to me as I thought that FAST Cache would absorb bursts of any type writes to slow SATA drives.

I am currently doing a LOT data movement as I am migrating from Raid Groups to pools so workload is very bad for FAST Cache, once things are settled I believe it will start working much better.

Kumar_A · Answer

Yes, as you have noticed FAST Cache will be effective only for the blocks which has already been promoted into FAST Cache because of their IO activity. As a general rule, it is intended for data that is being used very actively on your storage system - which is not really true in your migration from RAID Groups to Pools.

CLARiiON

FAST Cache with new writes?

Was this post helpful?