Striping Solid State Disks

[JoJoGabor]

SSDs are becoming common place is servers, particularly Blades, and are also in place in various SAN arrays, such as Compellent, EMC and 3Par. These are acting as a tier-1 storage tier, which fit inbetween the memory cache and your fast FC or SAS disks, so I dont think reliability is an issue. If striping the data you will get an increase in performance as each disk can perform read and write simultaneously.

[PRC1964]

We've recently put in a couple of Intel SSDs (not RAID yet, but we will try that in the next six months) and they have tripled the speed of the SQL read/writes we are doing.

From what I've read they might wear out over a period of 5-10 years. For the cost/performance I'd still have bought them if they wore out every six months.

The day is not far away when we will look back at mechanical HDDs as bizarre old technology.

[doodab]

No need for correction. This seems to be a doing something for something's sake. Typical techy nonsense. If you really need a raid array, you need to be going old school on the disks.

It depends on what you're trying to achieve.

If you need reliability and large capacity e.g. for a file server, then an array of conventional disks is the way to go, and if you need good sequential throughput e.g. for editing HD video then an array of conventional disks is still the way to go, but for some workloads SSDs are going to be a better choice.

[doodab]

There seems to be some debate on the issue but write endurance still seems to be limited to 100,000 cycles. That is fine for a laptop but not for a server. Correct me if I am wrong.

For the individual flash cells that is correct but the drives themselves overcome the problem by packing spare capacity and using "wear levelling" to balance the writes evenly across the drive i.e. they try to ensure that each block of flash is written to a roughly equal number of times.

[Jeebo72]

There seems to be some debate on the issue but write endurance still seems to be limited to 100,000 cycles. That is fine for a laptop but not for a server. Correct me if I am wrong.

No need for correction. This seems to be a doing something for something's sake. Typical techy nonsense. If you really need a raid array, you need to be going old school on the disks.

[Paddy]

There seems to be some debate on the issue but write endurance still seems to be limited to 100,000 cycles. That is fine for a laptop but not for a server. Correct me if I am wrong.

[doodab]

Lots of people have tried it, there are a few articles on sites like anandtech e.g:

http://it.anandtech.com/IT/showdoc.aspx?i=3532

The answer by and large seems to be what you would expect i.e. it gives you more IOPS and more throughput.

The latest fast and light Sony laptops use 4 x 64GB SSDs in RAID 0.

The current generation drives have addressed the early reliability problems and the intel, samsung, seagate etc ones are now as good or better than conventional drives in terms of MTBF, error rates and so on.

[VectraMan]

I thought it was meant to be awesome:

http://www.youtube.com/watch?v=26enkCzkJHQ

[Paddy]

Has anyone striped/RAIDed a collection of SSDs? A client is wondering if they'd achieve a performance boost. I've feeling it probably would. Especially on SEEK. So I'm asking for your input.

Incidentally the definition of 'RAID' and the cost of 'SSD' in combination made me chuckle.

I would not recommend using SSDs because of the life limiting R/Ws. SSDs have a very short life if you are doing a lot of R/Ws such as what a server OS would do.

[stingman123]

for once I agree with DP, I can't really see a perfomance boost coming from these, good luck though

[DimPrawn]

No.

They wouldn't get a performance boost, only a cost boost.

[threaded]

Has anyone striped/RAIDed a collection of SSDs? A client is wondering if they'd achieve a performance boost. I've feeling it probably would. Especially on SEEK. So I'm asking for your input.

Incidentally the definition of 'RAID' and the cost of 'SSD' in combination made me chuckle.