EMC will be the first major storage vendor to add solid-state drives for its high-end gear. The company said it will offer flash-based memory from STEC in its Symmetrix DMX-4 systems by the end of March. Solid-state drives (SSDs) prove speedier and less energy consumptive than traditional spinning disk storage. But the …
So how much faster than SAS?
How much faster is the flash drive when compared to a 15K scsi drive?
this depends on the IO load. compared response time is 10% for low IO rates and decreases with increasing load (much flatter curve than a standard drive)
the comparisons are afaik between the 73G flash and a 73G 15k FC/SAS drive
SSD response times
One of my favorite demonstrations of the response times of solid state drives is on the Storage Performance Council SPC benchmarks page. Results were submitted for a 72 GB device from Texas Memory systems. Even at 100% load (over 120,000 IOPS), the response times were still below 1 ms. Compared to an array (the IBM DS8300, which turns in a comparable IOPS result) comprised of 256 72 GB 15k drives (along with 256 146 GB drives and 256 GB of cache), the response time starts out around 1 ms and goes up to between 14 - 17 ms (depending on the operation).
But as the article points out, you have to balance the response time with dollars, as these drives are going to be pricey.
Re. "EMC will be the *first* major storage vendor to add solid-state drives for its high-end gear"
Although they probably no longer qualify as a "major storage vendor", StorageTek (now part of Sun) beat them to it by 30 years as they offered high-end SSDs in 1978. They were prohibitively expensive then too.
If you need to read data fast then your DB's cache is probably doing a very good job for you and this technology won't help.
Generally the problem is with witting being the bottleneck. I haven't checked the link given by another poster but I read somewhere else that these devices aren't very good at concurrent random writes.
I'd be more interested in how many times you can write to these devices. I know that it's a different tech to your USB pen, but in a datacenter, if every block can only take about 300,000 writes (to pull a figure completely out of the air) before it goes all "Quantum-physics" (maybe a 1, maybe a 0, maybe both) on you then that's not a lot of good. EMCs announcement and FAQ on their powerlink site don't cover this at all..
A quote from Wikipedia on the "endurance" section of the article on Flash Memory: The endurance of NAND flash is much greater than that of NOR flash (typically 1,000,000 cycles vs. 100,000 cycles). So can we assume, best case, that you can only overwrite a block 1M times? And how does the array handle this? I guess it could keep a write count for blocks and slander a drive as 'failed' if a block gets written > n times.. The info on powerlink confirms that these drives do hotsparing just as traditional drives do, but how/when does the array decide a flash drive is dead?
Just an element that I thought might need discussing - anyone who knows more about high end flash, or who works for EMC and can clarify the overwrite position without breaking any NDAs please feel free to tell me if I'm communicating from the wrong orrifice.. :)
30x IO at 30x price...
...and FC drives are about, what, £200 for a good quality high-capacity one?
£6 000! Jesus H Christ!
Re: SSD First
I don't have the links but SSD Drives have been around for a Long time. I think EMC had their first SSD System out in late 1980's.
The exploding need for more storage effectively killed the SSD drives as they became too hard to manufacture and too expensive for the needs at the time.
I imagine that those drives are made only for applications that have a critical need for good response times and they are not overwritten too often. Not sure what those might be but if some one is paying a chunk of change needed for the SSD drive, they would have an idea what to use them for.
Alan Parsons wrote:
> I'd be more interested in how many times you can write to these devices
The actual internal capacities are multiples of the exposed capacity allowing a whole-lot-more sector remapping than you'd get on, say, a SCSI/FCAL drive.
I/O speed is pleasant, but in my role it is not my primary concern.
My interest in this technology is power consumption and heat reduction.
DMX's ( along with superdomes and c-class chassis ) draw a hell of a lot of power per square meter compared to all other devices in a datacentre.
Electricity is a very large percentage of my operating budget, and the pricing for it is only going in one direction in the near future.
Especially if Iemma has his way selling of more state assets! ( sorry to all outside NSW )
Need a power Icon.