OCZ Core SSD + RR3520LF PCI-E Raid controller = 300MB/s writes!

Printable View

10-18-2008, 11:38 AM
eva2000

OCZ Core SSD + RR3520LF PCI-E Raid controller = 300MB/s writes!

I just setup my Highpoint RocketRaid 3520LF PCI-E raid 6 controller which has 256MB DDR2 cache with 4x32GB OCZ SSD Core original first batch that came out in Australia way back - ATTO 500+MB read/300+ MB writes, pcmark05 hdd test = 38,000+ and Everest Random Write ~179MB/s average :)

http://i4memory.com/reviewimages/har...0lf_oczssd.png

Highpoint RocketRaid 3520LF + 4x32GB OCZ SSD Core V1 Raid 0

A preview for ATTO Bench 256MB length and HDTach 32MB with 4x32GB OCZ SSD Core V1 raid 0 in various controller block size configurations.

ATTO Bench 256MB
256K Block size saw ATTO pull out well over 500+ MB/s read and 300+ MB/s write

16K Block size vs 64K Block size
http://fileshosts.com/intel/DFI/DFI_...atto_256mb.png http://fileshosts.com/intel/DFI/DFI_...atto_256mb.png

128K Block size vs 256K Block size
http://fileshosts.com/intel/DFI/DFI_...atto_256mb.png http://fileshosts.com/intel/DFI/DFI_...atto_256mb.png

HDTach 32MB
Again 256K Block size led the way with 482.8MB/s average read speed and 1002MB/s burst speeds

16K Block size vs 64K Block size
http://fileshosts.com/intel/DFI/DFI_...dtach_32mb.png http://fileshosts.com/intel/DFI/DFI_...dtach_32mb.png

128K Block size vs 256K Block size
http://fileshosts.com/intel/DFI/DFI_...dtach_32mb.png http://fileshosts.com/intel/DFI/DFI_...dtach_32mb.png

Compiling more app benches right now :)
10-18-2008, 03:24 PM
coo-coo-clocker

Sweet results!
Not surprising on the improvements from the mobo-based controller. Wondering about access times and random read/writes...
10-19-2008, 12:29 AM
eva2000

Yeah still compiling results, FYI everest random write speeds are ~179-192MB/s :)
10-19-2008, 02:11 AM
gillll

wooooooooooot!!!!!!! sexy !!!!! :D
10-19-2008, 04:15 AM
saint-francis

How does just one of the Cores preform when using a hardware RAID controller? And when are motherboard manufacturers going to start adding decent RAID controllers to their boards?
10-20-2008, 01:48 AM
eva2000

Quote:

Originally Posted by gillll

wooooooooooot!!!!!!! sexy !!!!! :D

thanks

Quote:

Originally Posted by saint-francis

How does just one of the Cores preform when using a hardware RAID controller? And when are motherboard manufacturers going to start adding decent RAID controllers to their boards?

haven't tested single non-raid OCZ SSD Core on a hardware raid controller yet.

I've finished compiling all the block size comparison results now and it's interesting to see how much difference in performance some block sizes have http://i4memory.com/showthread.php?t=10744 !
10-20-2008, 05:57 AM
Levish

Quote:

Originally Posted by saint-francis

How does just one of the Cores preform when using a hardware RAID controller? And when are motherboard manufacturers going to start adding decent RAID controllers to their boards?

Motherboards are more than bloated already, are you willing to pay a 400$ on top of a 250-350$ mobo?
10-20-2008, 07:20 AM
One_Hertz

nice! My two V2 60gb SSDs get 290mb/s read 220mb/s write 0.2ms on onboard ICH9. Entirely linear scaling too, single drives were 145/110. Pretty nice for sub 600 bucks...

There are some annoying freezing issues at times though. It hasn't pissed me off yet, but it might eventually.
10-20-2008, 12:31 PM
alfaunits

Any chance to IOmeter random write tests?

Plus, minimum random write is 12MB/s?
10-20-2008, 12:56 PM
Levish

Quote:

Originally Posted by alfaunits

Any chance to IOmeter random write tests?

Plus, minimum random write is 12MB/s?

random writes will look like the OP's Everest results which can be found in the link to their site

summary: nearly 350MB/s while there is cache space after which the typical spikey random writes showing 20ish MB/sec to 100MB/sec
10-20-2008, 07:01 PM
alfaunits

I couldn't give a rat's a** for random writes while there is cache space... that's worthless in the long run.
The average and minimum could be good or bad, depending on the buffer size used in the random write. No way it's 4K chunks. If it's 1MB chunk - that's measly 12 random write IOPS. So, we need a longer run with IOmeter.
10-20-2008, 11:01 PM
LinusTech

I'll be doing a quad SSD setup soon and I'm glad to see your results. I'll probably end up having to do my own testing anyway though because the card I have on order is an Adaptec 5805 instead of a Highpoint card.

I'm also not sure what SSDs to use. There's so little information out there on that weird hitching that Anandtech was reporting about... I don't want to drop 3-4k on a storage solution to freezes up periodically every time I get an IM....
10-21-2008, 02:13 AM
eva2000

i'm testing iometer 4k random writes 100% right now across different block sizes and getting ~417-470 IOPS and average i/o time is ~2.2ms

FYI, i've never had freezes, stuttering or corruption on ICH10R or RR3520LF as yet for my OCZ Core SSD since I got them 2 months ago :)

Update: Iometer 64bit results for 4k random writes http://i4memory.com/98305-post7.html

http://i4memory.com/reviewimages/har.../iometer64.png
10-21-2008, 05:05 AM
Oliver

awesome work as always George
10-21-2008, 06:05 AM
eva2000

Quote:

Originally Posted by Oliver

awesome work as always George

Thanks.. got more tests I want to run later on :D
10-21-2008, 06:19 AM
alfaunits

Can you do a longer run on the IOmeter, i.e. 10 mins?
1-1.7second longest I/O response time isn't promising :(
10-21-2008, 06:24 AM
Afterburner

Is this with or without the optimisations recommended in the OCZ Forum for OS eva2000.... :confused: :p:
10-21-2008, 06:30 AM
NeedMoMegaHurtZ

Thanks for sharing your testing results eva2000! Nice setup

As with alfaunits, I'm a bit concerned about the max I/O reponse times

Just a suggestion if you have time: can you include standardized IOMeter access pattern profiles i.e. file server, workstation, web server, database ? This way you can compare your results with many other published results
10-22-2008, 12:31 PM
Levish

Quote:

Originally Posted by alfaunits

I couldn't give a rat's a** for random writes while there is cache space... that's worthless in the long run.
The average and minimum could be good or bad, depending on the buffer size used in the random write. No way it's 4K chunks. If it's 1MB chunk - that's measly 12 random write IOPS. So, we need a longer run with IOmeter.

Maybe a iram or the Acard drive if it offers you enough storage space otherwise you are looking for a drive which does not exist (yet).
11-20-2008, 10:04 AM
NapalmV5

^ nice write performance @ atto

256k block
http://img171.imageshack.us/img171/6003/tach3510el6.png

as you can see.. read/burst is all good but write performance is that of a 2x

actually 2x write is higher on the arc1210.. 4x 6400aaks were performing the same way

looks more and more to me.. my 3510s write performance aint normal
11-22-2008, 01:07 AM
eva2000

been a bit too busy to do more iometer runs.. but will get around to them eventually :)

Quote:

Originally Posted by NapalmV5

^ nice write performance @ atto

256k block
http://img171.imageshack.us/img171/6003/tach3510el6.png

as you can see.. read/burst is all good but write performance is that of a 2x

actually 2x write is higher on the arc1210.. 4x 6400aaks were performing the same way

looks more and more to me.. my 3510s write performance aint normal

Yeah i'm loving with 4x OCZ Core SSD + 3520LF combo :)

which ssds are those results from ?
11-23-2008, 06:57 PM
NapalmV5

the ones here: http://www.xtremesystems.org/forums/...d.php?t=207557

when you got time.. post tach read/write..

if its not the same as my 3510.. ill get a 3520

edit: whats your 3520s pcb version ?