So I got a price quote on Netapp's Flash drives that you stick into a Flash Pool and having my eyebrows rise off my forehead. Of course Netapp's prices are/were no worse than EMC's prices for single SSD drives, etc. Basically crazy expensive in both cases.
Now I'm entertaining the silly idea of just using a mix of SAS and SATA drives in a flash pool instead. Get *some* of the advantages, but lots cheaper pricing. Esp using those big fat slow 4Tb drives in the dual drive sleds. Front them with a bunch of SAS disks and it should do quite nicely.
God knows there shouldn't be any technical reason this can't work, it's just a matter of moving blocks from one raidgroup in an aggregate to another. But the Netapp technical sales guy was surprised by the suggestion.
I'm loving the transition of SSDs into the enterprise space, because it's forcing all kinds of changes on the vendors that are long overdue. Just putting a pair of mirrored SSDs into a server is a big boost. Using it to cache is good too.
What do others think?
John
John Stoffel wrote:
Now I'm entertaining the silly idea of just using a mix of SAS and SATA drives in a flash pool instead. Get *some* of the advantages, but lots cheaper pricing. Esp using those big fat slow 4Tb drives in the dual drive sleds. Front them with a bunch of SAS disks and it should do quite nicely.
Not so nicely I think...
God knows there shouldn't be any technical reason this can't work, it's just a matter of moving blocks from one raidgroup in an aggregate to another.
Well, the problem is basically that if you put spinning disks in that/those RGs which are FlashPool, they will get thrashed so badly it will just make them melt and latency will go through the roof anyway There's no point, just run the system with the FAS box and som FlashCahce in it, as much as you can afford -- it will work better
/M
Hi John,
to add a technical reason: SAS and SATA have a different sector format.
With SAS drives the sectors are formatted with 520 Bytes, leaving 8 sectors with 64 'Extra-Bytes' for stuff like checksums used e.g. for media scrubbing (data verification). SATA drives are formatted with 512 Bytes/sector (or 4K/s) and use a different protection scheme. Can't combine the two in a RG or even in an aggregate, therefore no SAS/SATA-Pool aggregate possible.
Anyway, you'll get more IOPS/$ with flash, so I don't see a SAS/SATA combination in the future. Also the SSDs take random load off of the disks, so putting them on SAS instead of SSDs doesn't eliminate the rotational/head latency and would get you only a relatively minor improvement. If you'll ask me, in a few years it'll probably be SSD/SATA disks (best capacity/$ & best IOPS/$) with SAS disks being squashed out in between them.
Sebastian
On 3/11/2014 4:28 PM, Michael Bergman wrote:
John Stoffel wrote:
Now I'm entertaining the silly idea of just using a mix of SAS and SATA drives in a flash pool instead. Get *some* of the advantages, but lots cheaper pricing. Esp using those big fat slow 4Tb drives in the dual drive sleds. Front them with a bunch of SAS disks and it should do quite nicely.
Not so nicely I think...
God knows there shouldn't be any technical reason this can't work, it's just a matter of moving blocks from one raidgroup in an aggregate to another.
Well, the problem is basically that if you put spinning disks in that/those RGs which are FlashPool, they will get thrashed so badly it will just make them melt and latency will go through the roof anyway There's no point, just run the system with the FAS box and som FlashCahce in it, as much as you can afford -- it will work better
/M _______________________________________________ Toasters mailing list Toasters@teaparty.net http://www.teaparty.net/mailman/listinfo/toasters
Sebastian Goetze wrote:
to add a technical reason: SAS and SATA have a different sector format.
Could always use the NL-SAS 4TB (and up) drives then ;-)
If you'll ask me, in a few years it'll probably be SSD/SATA disks (best capacity/$ & best IOPS/$) with SAS disks being squashed out in between them.
Of course, SAS 10K rpm has no future that I can see at least. Unless the eMLC and going fwd techniques fail for some reason We already have 1.6 TB eMLC SSD out in the field AFAIK. To dear still sure, but it surely will drop
/M
-
John Stoffel -
Michael Bergman -
Sebastian Goetze