Sign in with
Sign up | Sign in
Your question

Software RAID vs. Hardware RAID and dual core

Last response: in Storage
Share
September 8, 2006 4:09:34 AM

Will the processor overhead associated with software implemented RAID operation be handled well enough in a dual core conroe system to out-perform hardware raid for single-user use?

What are the considerations in examining this?
September 8, 2006 4:33:04 AM

Quote:
Will the processor overhead associated with software implemented RAID operation be handled well enough in a dual core conroe system to out-perform hardware raid for single-user use?

What are the considerations in examining this?


Don't use software RAID. Unless you are doing RAID 0 with no hope of data recovery... there are many other issues to consider besides the overhead, software RAID is just worse in many other ways ( rebuilding arrays for example).
September 8, 2006 12:43:07 PM

Your question is two part, so first: Yes duel core can handle the overhead associated with software raid. Second, you can look at this question in many ways. You need to look at the class of cards your using, chipset, on board memory, on board processor as well as BUS interface. A majority answer would be no because more people are using higher end raid controllers on a PCI-E bus with independent processing on the controller. This takes huge percentage of over hard off the system CPU. As technology progresses this may be harder to illustrate, but a proof of concept would be to look at the way native DMA raid 0 (on board) waked CPU usage.
Related resources
September 8, 2006 1:34:25 PM

Out of curiousity, is the raid on an Nforce chip a "proper" raid?
It'll max out at 140Mb/s at the best spots, so I figure it's speedy enuff.
Two 36gb raptors.
September 8, 2006 2:24:25 PM

Quote:
Will the processor overhead associated with software implemented RAID operation be handled well enough in a dual core conroe system to out-perform hardware raid for single-user use?

What are the considerations in examining this?

Theres also a third question. Are you using a large database or doing large video editing that needs a RAID? If not you may be better off with a SSD or a RAM disk. The worst SSD, being Iram, you can find will beat the fastest RAID, 2X raptors, on any none database and video editing program. Anandtech done a review on the Iram v/s RAID and the Iram made the RAID look bad. Iram has a faster boot for XP and with room for XP to have a super fast virtual memory. If you have any old DDR memory this would be a killer way of reusing and maybe getting a longer life of a maxed out RAM system.
http://www.anandtech.com/printarticle.aspx?i=2480
The artical does state that a 8GB Iram would be better for games but gigabyte has a second version on the way which is 8GB and SATA2.
September 8, 2006 2:55:43 PM

This technology has not progressed since I've touched it back in 2000. I used an enterprise solid state disk by Platypus. The throughput, rendering, fast rw access than r were limitations stood out the most. Maybe they will get better, But the overall cost to speed to functionality ratio really hasn't been archived yet. Perhaps this why the server end application of SSD has been pass down to consumers.
September 8, 2006 3:10:40 PM

I'm running matrixed arrays of RAID 0 (4 disks) and RAID 5 (same 4 disks) through a D955XBK. The RAID 0 works just fine - no noticeable CPU overhead and pretty darn fast. The RAID 5 does not incur a lot of CPU overhead, but does have very slow write speed. This is because for every write, it actually needs to do a read-merge-write operation. Intel calls their onboard RAID 5 implementation "intelligent RAID". I call it slow - it's slower than 20 Mbps write speed and is inadequate for recording live TV shows. It's great however for anything that's write-once, read many, such as your picture collection that you don't want to lose. Read speed is also decent (because it's basically a 3-disk RAID 0) which makes it a good idea for the O/S as well. My HTPC boots a lot faster than any single-disk systems I work with these days. You just don't want to put your pagefile, temp space, or anything like that on RAID 5 so careful configuration is important.

I had a HDD fail recently I was able to rebuild the RAID 5 array very easily. Even doing a RAID 5 rebuild, the CPU usage stays pretty low. If I use a different disk for temp storage (I have 6 installed), I can even watch and record TV on MCE during the rebuild with no lost frames.

I used to have a RAID 10 (01?) array based on the same Mobo and disks, but the performance of that wasn't great, and the storage penalty was too high.

If you have only two disks, RAID 5 isn't even an option. In that situation, I will say that neither RAID 0 nor RAID 1 should incur a big CPU hit because neither require the costly XOR operation that RAID 5 uses.

Update: After fixing a problem with a driver, I have observed up to 5 MB / s (40 mbps, twice the previous stated number) write performance with the RAID 5 array.
September 8, 2006 3:31:13 PM

There's also the rebuild question and system availabilty. With many hardware
ie. controller based RAID solutions, the rebuild must be done using the
onboard firmware before the OS can boot or the volume is made available.
I have a Redhat 9 server with mirrored 60gb drives running under software,
MD. If a drive fails or the array needs to be rebuilt, this can be done while
it's being used. And the system performance is still adequate. We have a
small departmental server with 4 300gb drives mirrored for 600gb online
running W2k server 2000. When one drive failed last year, we had to endure
3 hours of downtime before the OS would boot. So, software RAID has its
advantages and with the fast CPUs today, the performance matches or in
some cases exceeds hardware only RAID solutions.
September 8, 2006 9:07:07 PM

Quote:
This technology has not progressed since I've touched it back in 2000. I used an enterprise solid state disk by Platypus. The throughput, rendering, fast rw access than r were limitations stood out the most. Maybe they will get better, But the overall cost to speed to functionality ratio really hasn't been archived yet. Perhaps this why the server end application of SSD has been pass down to consumers.

I see. I guess you've never used a SSD for the XP virtual drive in a system with maxed out memory. Iram in my old maxed out 1GB system cut my Maya3D renders in half. There is 1 added benefit, to SSD's, as my HD is used quit a bit less and should last much longer.

The big problem with your statement is in 2000 PATA could come close to being maxed out by a high performance RAID. SATA offers about 15 times the thoughput of old PATA and RAID may use 10% of its thoughput. SSD on the other hand can use almost all of a SATA's thoughput. The only thing RAID has on SSD is size which any new single HD can fill.
September 9, 2006 11:47:32 PM

Quote:
SATA offers about 15 times the thoughput of old PATA

UTTER WRONG

http://www.sata-io.org/3g.asp
Ditto!

My experince has only come from the server side of the industry. Could you site some of your data, I would be interested in seeing some white papers.

Until the IEEE standards sink into everyones head on SATAI|II, there will always be miss-information and slanted opinions of a technology that has not reached it 'therological' specs. This is why SSD technology has taken so long to be pushed out and why it's on teh consumer level.
September 10, 2006 12:59:57 AM

Quote:
SATA offers about 15 times the thoughput of old PATA

UTTER WRONG

http://www.sata-io.org/3g.asp
I left the dot out so sue me. 1.5 times faster than PATA-100. Sorry for the confusion. I wish SATA was 15 times faster than PATA. We could have 20 RAID drives on that and have extra bandwidth for a bunch of optical drives.
September 10, 2006 2:18:52 AM

Quote:
That's the difference between pass and failure in any exam. :lol: 

LOL! Thats true but I didnt get the study the night before the test. My dog was sick. I had a flat tire and ended up having to sleep in the back seat till the tow truck arrived.

what great storys we would miss out on if we didnt make mistakes.
!