Overclock.net banner

4 10k drives raid 10 vs SSD

6.9K views 18 replies 11 participants last post by  u3b3rg33k  
#1 ·
Price being no concern what would be better the raid array or the single SSD?
 
#4 ·
I just replaced a RAID 5 with 3 500GB caviar blacks with an OCZ vertex 3 120GB and the SSD wins hands down. Its the best upgrade that I've ever done by far (even beats going from a 5770 to a 580). It's so fast that starting up windows (including typing in my password and login) is at least four times as fast as my POST.
 
#5 ·
I have a budget SSD and it makes things fly. Linux boots in 5 seconds and has a usable browser in 10. My bios take twice as long. I could only image how much faster a high performance ssd would be.

Also if price is no concern then why not just raid some SSD's?
 
#6 ·
Quote:
Originally Posted by animal0307;13766330
I have a budget SSD and it makes things fly. Linux boots in 5 seconds and has a usable browser in 10. My bios take twice as long. I could only image how much faster a high performance ssd would be.
I have a F60 SATAII with 275 read/write speeds and my PC takes longer to post then it does to get into windows and open firefox lol

I can imagine like an INTEL 510 or Vertex 3
eek.gif
 
#7 ·
I have 4 10k Savios - and can confirm that even a budget SSD wipes them away, even in RAID0 let alone RAID10.

The SAS drives are sat doing OS/VM duty in my server rather than my workstation. Better value/GB than SSDs (at least when I bought them), but not fast enough to wow you.
 
#8 ·
After two hours of continuous writes, a C300 drops to 15ms access time and 2k iops. Now imagine that sucker after a day a two. I'll stick with my spinners, thanks. =)

That being said...4 drives is really not that many, especially if you are concerned about write performance. With hard drives, it's the opposite of SSDs - you want to go with as many small drives as possible. 10 x 36GB / 15k / 2.5" would be a lot more impressive than 4 x 146GB / 10k / 3.5".

The only thing that can really match or exceed a decent sas array in performance is a ram disk.
 
#9 ·
Quote:
Originally Posted by thefreeaccount;13782797
After two hours of continuous writes, a C300 drops to 15ms access time and 2k iops. Now imagine that sucker after a day a two. I'll stick with my spinners, thanks.
That sounds like a firmware problem, not all SSDs behave this way.

See this thread: http://www.xtremesystems.org/forums/showthread.php?t=271063&page=14 After 64 TERABYTES of writes to a 40gb drive, speeds have not dropped at all. Performs just like a brand new drive. (post #335 and 337)
 
#10 ·
Quote:
Originally Posted by thefreeaccount;13782797
After two hours of continuous writes, a C300 drops to 15ms access time and 2k iops. Now imagine that sucker after a day a two. I'll stick with my spinners, thanks. =)

That being said...4 drives is really not that many, especially if you are concerned about write performance. With hard drives, it's the opposite of SSDs - you want to go with as many small drives as possible. 10 x 36GB / 15k / 2.5" would be a lot more impressive than 4 x 146GB / 10k / 3.5".

The only thing that can really match or exceed a decent sas array in performance is a ram disk.
I'm not sure where you got the degradation statistic from, but I don't think it's correct.

And I'm also pretty sure that SSDs are in fact much faster than RAID arrays, and you 'sticking with your spinners' isn't a wise move if you want the best performance...
 
#11 ·
Quote:
Originally Posted by thefreeaccount;13782797
After two hours of continuous writes, a C300 drops to 15ms access time and 2k iops. Now imagine that sucker after a day a two. I'll stick with my spinners, thanks. =)

That being said...4 drives is really not that many, especially if you are concerned about write performance. With hard drives, it's the opposite of SSDs - you want to go with as many small drives as possible. 10 x 36GB / 15k / 2.5" would be a lot more impressive than 4 x 146GB / 10k / 3.5".

The only thing that can really match or exceed a decent sas array in performance is a ram disk.
2k IOPs is still more than you'll get from either of your SAS arrays...
 
#12 ·
Quote:
Originally Posted by Scorpii;13782869
And I'm also pretty sure that SSDs are in fact much faster than RAID arrays, and you 'sticking with your spinners' isn't a wise move if you want the best performance...
SSD and RAID are not exclusive...

As I plan on proving as soon as I can justify $2k on SSDs.
 
#13 ·
Quote:
Originally Posted by u3b3rg33k;13783320
SSD and RAID are not exclusive...

As I plan on proving as soon as I can justify $2k on SSDs.
and what would you do with them? Because for 90+% of uses you'd be wasting upwards of $1700...
 
#14 ·
I plain don't trust storage devices. And striped SSDs is faster than not striped drives, striping yields a bigger drive... I'd only be "wasting" $500, as one of the drives is effectively (not literally) dedicated to parity.

Besides, my raid card is capable of "800 MB/sec RAID 5 reads and exceeds 380 MB/sec RAID 5 writes", so it's not like it'd be slow.
(source http://www.supermicro.cc/Datenblatt/9550SX.pdf)
 
#15 ·
Quote:
Originally Posted by u3b3rg33k;13783571
I plain don't trust storage devices. And striped SSDs is faster than not striped drives, striping yields a bigger drive... I'd only be "wasting" $500, as one of the drives is effectively (not literally) dedicated to parity.

Besides, my raid card is capable of "800 MB/sec RAID 5 reads and exceeds 380 MB/sec RAID 5 writes", so it's not like it'd be slow.
(source http://www.supermicro.cc/Datenblatt/9550SX.pdf)
You'd be wasting the money because you wouldn't be gaining any real-world speed.

And even with a hardware RAID card you wouldn't want to run RAID5 for your OS - the performance would be lower than with a single drive for most operations.
 
#16 ·
I run only raid 5 in my desktop, and it's definitely not slower than a single drive, but that's probably due to NCQ and the 128MB of cache. It is noticeably slower with the cache off (but why would I run like that?).

The controller also gets me a few other benefits:
I can swap out drives without powering down
I can grow the array by putting in bigger drives, without powering down, or re-installing the OS.
 
#18 ·
Quote:
Originally Posted by u3b3rg33k;13791677
I run only raid 5 in my desktop, and it's definitely not slower than a single drive, but that's probably due to NCQ and the 128MB of cache. It is noticeably slower with the cache off (but why would I run like that?).

The controller also gets me a few other benefits:
I can swap out drives without powering down
I can grow the array by putting in bigger drives, without powering down, or re-installing the OS.
What controller are you using?

RAID5 does tend to be slower than a single drive in the most important factors for an application and OS drive..... random read/write.
Quote:
Originally Posted by thefreeaccount;13782797
After two hours of continuous writes, a C300 drops to 15ms access time and 2k iops. Now imagine that sucker after a day a two. I'll stick with my spinners, thanks. =)

The only thing that can really match or exceed a decent sas array in performance is a ram disk.
What are you doing that is do continuous writes/deletes/updates for that period? This is definitely non-typical desktop usage (but I suspect this is some server).

Depends on your performance metric. A SAS array might be good in sequential but a single SSD can destroy SAS arrays in random performance. In fact, an array of SSDs can destroy SAS in sequential as well. The peak sequential read of a SAS HDD is around 200MB/s last I checked. SSDs are hitting 500MB/s reads now.

An SAS array will not be able to keep up to a SSD in database selects.
 
#19 ·
Quote:
Originally Posted by DuckieHo;13809312
An SAS array will not be able to keep up to a SSD in database selects.
http://www.hardwarezone.com/tech-news/view/175705
"Talos series, a line of Serial Attached SCSI (SAS) SSDs"

What about an array of SAS SSDs vs SATA SSDs?
I would be surprised to see an enterprise setup that doesn't use some form of redundancy, especially for a database setup.

The bus/controller/casing color does not determine the storage medium. We should make that distinction.