Areca ARC 1880-ix-24 some quick benchmarks

[Rubycon]

This is a new 6Gbps SAS host and is FAST! :awe:

As tested:

Six Intel X25-M 160GB SSD in RAID0, 4KB stripe, write back, 4GB ECC DDR Cache with 72hr battery backup module.

 

[DominionSeraph]

:/
Sequential is faster than my RAM.

 

[Emulex]

i'd disable the write back cache and let the drives do the caching. well err those drives don't cache. but try that anyways

 

[Rubycon]

The controller has cache and yes the benchmarks reflect this. Even with it disabled completely read speeds exceed 1.6GB/S and write close to 700MB/S. (Intel's writes are lower as everyone knows but IOPS are through the roof and THAT's what matters most - to me anyways. )

Since the cache is 4GB and most stuff runs in that cache programs run like the wind as you can imagine.

I was using a core i7 notebook with a single SSD and it's pretty fast but going to this feels like ludicrous speed. - http://www.youtube.com/watch?v=wI4tevra8Lg

Ok maybe not, but it's fast(er)...

:awe:

 

[Emulex]

4gb!! insane!! how the heck do you flash back that up?? another ssd? that must cost hella lot of protect.

wow. Cool. So if you had a drive that had policy write back cache you could get higher iops by bypassing the cache on board and using the drives?

What if you took the 4gb and gave that to the o/s for buffer. i wonder what the net would be if it was locked in use for lazy writes(sql)?

 

[Rubycon]

4gb!! insane!! how the heck do you flash back that up?? another ssd? that must cost hella lot of protect.

wow. Cool. So if you had a drive that had policy write back cache you could get higher iops by bypassing the cache on board and using the drives?

What if you took the 4gb and gave that to the o/s for buffer. i wonder what the net would be if it was locked in use for lazy writes(sql)?

It's handled internally and the firmware does allow for performance tuning but (IMO) should have more "enthusiast" options available for fine(r) tuning. Comparing the cache to an OS based RAMdisk there is little difference. I suspect either the benchmarks are not designed to properly scale to this level or it's an OS limitation (Win7 64 bit) or a combination of both.

AS SSD total score is over 6K(!) on a compressed volume but as you can imagine there's a huge write penalty. If the write task (again software limits here) was truly multithreaded performance would be higher but at an expense of compute cycles for other things.

 

[n7]

 

[God Mode]

Weaksauce.

 

[ViviTheMage]

Damn, what application/purpose is this for?!

 

[God Mode]

Damn, what application/purpose is this for?!

Charting the galaxy on the enterprise.

 

[Burner27]

Damn, what application/purpose is this for?!

Solitaire.

 

[Emulex]

you should move to vmware/sql type benchmarks to be fair when doing such large writeback cache. heck for 512meg writeback cache i think you need to do a more exhausting test. Check out the vmware forums for SAN benchmarks. i did a post a while back but nobody seemed interested. you could probably find all you need by searching this forum on my nick 4-6 months back.

 

[Rubycon]

I've seen those and while they present an interesting and increasing workload don't reflect what this array was intended for.

The data sets are highly random bursts (counting) and then get streamed and dumped out.

 

[DominionSeraph]

Must... see... 4k aligned SandForce.

 

[Gillbot]

now you need to get aigo to upgrade to real i/o eh? ;-)

 

[Rubycon]

Must... see... 4k aligned SandForce.

Sandfarce too slow in i/o department for this application. It took 8 of them to match six Intels for sequential transfer but iops trailed by roughly one fourth!

 

[DominionSeraph]

Sandfarce too slow in i/o department for this application. It took 8 of them to match six Intels for sequential transfer but iops trailed by roughly one fourth!

Did you use the right ones?

 

[Rubycon]

Did you use the right ones?

I was not doing the actual testing but they were indeed so called "enterprise" drives. The intel drives are faster and as a matter of fact the only way to significantly speed up things is to step up to true enterprise (SAS) SSDs that cost several thousands of dollars EACH.

 

[mooseracing]

Wow, nice. I am using a 6GB SAS carb but limited by using SAS drives, not SSD. I was happy to get 1.4Gb/s. Puts 10Gb network to work.

 

[Cr0nJ0b]

well now I know what i'm going to get with my lottery winnings....that is...when I win the lottery. Until then I'll have to only dream of buying that kind of performance.