Tough descision - RAID configuration (RAID5 or RAID0&RAID5)

[kokoon]

i meant RAID10 writes are much faster compared to RAID10 reads. if the benchmarks above are correct. isn't that weird?

 

[hafa]

no i'm really glad you posted these, i needed to know how fast RAID10 is. it's writes are really fasts but reads aren't THAT fast. i wonder why writes are faster.

still, what would happen if i put different disks in a RAID10 array? will it even work? how slower will it be and why? you guys have any links?

I found the realatively slow read times interesting as well; I'd have expected them to be more even...

This link has qite a bit of good information about matched drives. Although it deals with Linux RAID, the hardware principles remain the same.

 

[Patched]

I'm guessing the writes are being cached into the ram (disks or system) since the file size is so small and the reads actually have to read from the disks? Could you try with files a lot larger?

 

[Randyman...]

Exactly - 4MB files are likely small enough to get handled by the Write Back Cache. You are benching the speed of the Cache, not the speed of the Writes to the Array...

See http://www.ocforums.com/showthread.php?p=4830341#post4830341 , and look at Casix's first posted results with a small file transfer (post #17), and then look at his second post using a larger file (post #20). As you can see, the small files show insanely high transfer rates because they are writing to the Cache (and then the array transfers the Cache to the disk at its leisure). The large file bench is actually measuring the Arrays Write Speeds since the "test file" is larger than the availabe Cache.

I still don't know how to select 256MB file sizes in ATTO (like Casix did in his second post) - I can only select 32MB IIRC. I guess that is a feature of the "Pro" version?

 

[attack]

RAID5 writes are always slow when you use a non XOR controller, it has to send those calculations to the CPU, and then from CPU back to the chipset for the final write. It does put a load on the CPU but that's the least of your worries, the latency for the calculations is what screws the non XOR controllers. Also with an add on card you can put memory...256mb+ easily which can cache all the data to RAM and store it so it can write it after you've stopped sending.

I think your just much better off with the RAID 10 all around redundancy and performance. It does have more overhead, but I feel the performance gains are worth it, especialy with storage being very inexpensive.

As for putting pagefile on a different partition, you'll actualy get WORSE performance and heres why:
Most likely you'll partition it last, making it put closest to the center of the disk where linear speeds of the disk are much slower, and even if it is at the beginning (outer most part of the disk) the arm is going to need to swing from the middle where most likely it's been reading, to the outside passing over empty space left for your large pagefile which takes more time. If you have a seperate drive(s) on a different array of disks or solo then it's better to put the pagefile on it because it's completely independant of your other array and your drive can focus on only pagefile and the other arrays/drives on whatever your normaly working on.

 

[Casix]

I still don't know how to select 256MB file sizes in ATTO (like Casix did in his second post) - I can only select 32MB IIRC. I guess that is a feature of the "Pro" version?

I guess there are more versions, i also downloaded a ATTO once and it only let me go to 32mb, this version lets me go to 256mb: http://www.hugesystems.com/supportspace/bench32.exe

 

[Randyman...]

Thanks for that link, Casix! Here is the results from my 256MB Bench:

http://www.ocforums.com/showthread.php?t=487961&page=2

 

[hyperasus]

I hate to say this, but anymore I highly recommend against Raid5. It's just so dang slow, Raid 5 is great for NAS boxes and things that don't require much performance. But on a desktop, its too slow. Raid 10 still isnt crazy fast but its alot faster than Raid5. Fast enough to actually use anyways. I did the matrix raid thing on my computer with a small Raid0 for windows partition and then a Raid5 for my documents and storage. Even just using it for my documents it was so laggy I couldnt take it. Changed over to Raid10 and everythings been peachy since. I have some good benchies for Raid5 vs Raid10 but I'm at the wrong computer atm so I cant show you. I believe I posted them in bings thread somewhere.


Edit: found it
http://www.ocforums.com/showpost.php?p=4705935&postcount=227

 

[hyperasus]

RAID5 writes are always slow when you use a non XOR controller

I see this alot when talking about Raid5.......but is this really true now days? I kind of doubt it. Can someone show us some benchmarks of Raid5 XOR beating Intels ICH8 Raid5 with an overclocked conroe? pxi-e buss is slow compared to the southbridge. Even if this is true, it is going to take one heck of an expensive Raid5 card to outperform the southbridge matrix raid.

 

[MadMan007]

Hey hyperasus, I see you have the P5B-D. Which SATA controller ports were you using for your 4 disks in those benches you linked? I remember seeing somewhere on these forums that using different ports helped a lot, but it was buried in some other thread and I can't remember where.

 

[hyperasus]

I was using ports 1,2,3,4. After seeing others say its better to use 1,2,5,6 I changed mine. I personally couldnt see any difference after re running the benchmarks. I'd just go ahead and use ports 1,2,5,6. Can't hurt anything.

 

[MadMan007]

Did you run ATTO on the RAID5 once you were on 1,2,5,6?

Also is the setting on page 4-14 to [RAID] - I'd guess you already had that though.

 

[hyperasus]

Yes I ran ATTO after changing the ports and didn't see an increase in performance.

I'm not sure what you are referring to page 4-14[RAID]?