I'm trying to figure out if there is a true benefit in moving to SAS. What do you guys say? You obviously get more per gigabyte with SATA, but on the other hand you do benefit from greater performance with SAS.

SAS has the benefit of better I/O performance. I can get a RAID-5 for 3x300GB SAS drives at 15000rpm, or I can get a RAID-5 for 3x1024GB SATA drives at 7200rpm.

From what I've read, the difference in speed will not be large for SATA versus SAS when RAID-5 is implemented.

Same answer as ATA vs SCSI for anything.

SCSI/FC drives are the batches that test out for higher reliability. SATA drives are the ones that don't.

SATA has finally gotten to the point that it's not totally fucking braindead IDE crap, so on low end systems it's effectively just as good. For any actual performance intensive system I'd be going with a FC attached EVA4400 or equivalent and the question is largley moot. Once you've got some sort of RAID system in between you and the drives, the cache size of the RAID becomes the dominant performance factor.

I think it honestly depends on what you need and the application utilizing the storage.

I attended a really good TPC class recently. If there is interest I will scan some of the metrics and upload them (of course it is IBM centric though). We've been researching performance products and finally went with TPC, gah what a beast but anyway.

Obviously cache helps but so does structure of your array (striping at the OS etc.) Those speeds and sizes do start catching up with you during peak times. As does paths/thru-put, OS capability. I used to micro-manage every lun on my arrays, until we got our SVCs, now? Not so much and no more OS striping but that's completely different.

Oddly enough in the last 6 months I've had about 8 failed 143G/15k drives where as my slow SATA drives are rock solid and haven't failed in a long while (jinx). Of course they are strictly used for backup to disk/staging area where as my truly expensive disks are serving up vital apps. I have not had time to really read much about SAS but given what I'm experiencing now I see no need to go out and get something new. It is however from what I know of my limited knowledge that the benefit is not as significant as SCSI/FC over SATA.

...
Oddly enough in the last 6 months I've had about 8 failed 143G/15k drives where as my slow SATA drives are rock solid and haven't failed in a long while (jinx)...


Not really that odd when you consider the mechanics. Faster spinning disks = more wear and more chances of failure. Yes, I know they don't just take a standard hard drive and spin it faster, but no matter how well you engineer a hard drive, when you push limits you typically shorten the life of the hardware.

Malse hit the nail on the head... going SAS is exactly the same as going with standard SCSI. You gain performance but at an added cost. I deployed a $35k SQL server (price included licensing, 2 quad procs) for a client a few weeks ago with 16 SAS drives in it and 32GB of ram. Thing was a screamer, as would be expected. You get what you pay for (usually).

That said, I haven't done much side-by-side comparison myself. Personally, I'm all for getting as much hard drive performance as possible since disk IO has been the one area of system performance that has not had any major upgrades (affordable ones anyway) in the last 10 years. Consumer hard drives are still spinning 7200RPMs... they've SLOWLY gained IO speeds... but where Processing power has increased 10 fold in the last 10 years (gogo Pentium II... actually more than 10X if you consider multi-core processors), disk IO speeds have only doubled (50ish MB/s to 100ish MB/s for standard consumer drives).

I'm drunk and an academic nerd. I opened this thread thinking it said Stata vs SAS. WTF...

Clearly a SAS commando can take a SATA array, no matter how fast it's spinning.