Opinions from experience - RAID 0 vs separate disks

Jimbo Mahoney

Jimbo Mahoney

Jimbo Mahoney

Soldato
Joined
18 Oct 2002
Posts
6,669
I'll soon have 2 x 150GB Raptors.

Now, traditionally, I've always used two drives in my PC - one for the OS and one for the Swap and programs (swap at the start of the drive in its own small partition to get maximum transfer speeds and prevent fragmentation / make defrag easier).

I currently use a 36GB Raptor as my OS drive and a 200GB 7.2k disk as my games / apps drive. These are both getting long in the tooth and have fairly low transfer speeds (~55MB/sec) and high seek times.

My primary use is gaming, overclocking and benching, there will be no irreplaceable data on this machine - it is almost entirely expendable as it's purely a gaming / clocking rig.

I'm almost sure to do a test with both configurations, but I'd be intrigued to hear from people who have tried both options - two drives as one RAID 0 volume, giving higher transfer rates vs. two separate drives giving independent disk usage to the OS and the games / swap.

OS will be Vista 64, 4GB RAM. RAID would be motherboard RAID, not separate card.

If I do go RAID 0, given my planned usage, would my best bet to optimise performance be to analyse the files on the disk using JDiskReport and choose a Stripe size accordingly? e.g. say my most common file size is ~ 64kB, then my best Stripe size should be 32kB when using two disks?

EDIT - [LittleBritainCarol]Anandtech says no.[/LittleBritainCarol] Anandtech reckons largest stripe size available, so 128kB typically.

My gut tells me that RAID 0 would create a faster system, given that I am really only doing one thing at a time (e.g. booting OS, then benching, then playing a game)...

EDIT 2 - I've decided not to bother with RAID 0 as the overwhelming real-world evidence of reducing level loads, as well as analysis of my drives' typical files size, all point to around 1% speed increases.

64kB Stripe:

http://www.anandtech.com/storage/showdoc.aspx?i=2969&p=8

128kB Stripe, Raptors:

http://www.anandtech.com/storage/showdoc.aspx?i=2101&p=10

Analyse your drives' file sizes to help determine stripe size:

http://www.jgoodies.com/freeware/jdiskreport/index.html
 
Last edited:
My opinion is that Raid 0 will be much faster, the supposedly higher seek time has never been factor that has made a notieable difference, where the transfer speed is immeadiately noticable as being much faster. I can't see any benefit to be gained from a descrete raid card for raid 0, and so you are correct to use the onboard solution.

Regarding stripe size, I don't have the answer, opinion seems to reflect that bigger stripe sizes are better for large files, small for small files. i am not sure that I understand this, as I would have thought that this relates more to the formated block size over the hardware stripe size. Most onboard controlers don't give you the option of anything larger than a 128kb stripe, and when you think of the file sizes that you are dealing with, there are no going to be many that are smaller than this.

Hope that this helps
 
Pandobear said:
Regarding stripe size, I don't have the answer, opinion seems to reflect that bigger stripe sizes are better for large files, small for small files. i am not sure that I understand this, as I would have thought that this relates more to the formated block size over the hardware stripe size.
Click to expand...

I would assume it's because the file gets split into the stripe size before getting written to disk using the cluster size.
 
Putting benchmarks aside (I never base my buying decisions on them) it just feels better, installs are faster, games load faster and general usage just feels smoother.

The question is not if it makes a difference (because it does) but if it is worth the hassle? I think it is, it's not difficult to do.
 
I've just gone from a single 640gb wd to 2 of them in raid 0 and have to say I've not noticed that much of a difference , I was certainly expecting more of a performance boost, no doubt it is noticeable at certain times but it just doesn't generally feel any quicker :confused:

You do have to be more on the ball regarding backups if your going raid 0 all the way aswell as it seems rather fragile when windows crashes and you need to hit the reset button and i believe power failures are not good at all
 
When I first got my 640gb black drives, I tried just one to see how quick they were as single drives, it was snappier than a 500gb aaks and blew away the 80gb hitachi which I had used in 2003 in a raid array.

Then I used all 4 640gb drives in a raid 0 array with the last half split as a raid 5 array (1tb-ish for each partition) and did some real world testing.
Everything I did was much, much quicker than a single drive; file copying, installing apps, loading games and os booting. I went from a 128k stripe to a 64k and then to a 32k. I noticed that the 128k was the slowest of the lot (had to try to notice) but the 32k lost its speed very quickly (obvious) which can be recovered with a deep defrag of the drive which has to be done often.

I had the same experience on my laptop, where 2, 200gb 7200 drives were much quicker for general use over one drive. I stuck to the 64k stripe size after a bit of testing.

To summarise, RAID 0 is a simple way to boost transfer speeds and results in noticeable gains over a number of different tasks. The stripe sizes you should generally go for are either 128, 64 or 32. 128 and 64 will get away with occasional defragging and are good performers but 32k and lower while have slightly higher transfer rates in benchmarks and a few apps are not, imo, worth the extra defragging you have to do. I stick to 64k.

A quick pointer, moving from the ich9 to a hardware raid card, in my case an areca 24 port with a 1.2ghz cpu and some cache (512mb to 2gb) also resulted in a very good boost in general performance.

*edit* My bad, I was a stripe size out, it's 32k I stick with, not 64k. 8-16k was the one that I noticed a drop in drive performance after around a week, cured with deep defragging ofc.

I have since moved from the twin 200gb's that were included in my laptop and onto a pair of WD 320gb blacks, will try 16k and see how it plays out as they are very different drives. (need to do a clean install on it)
 
Last edited:
I have used a 2 HDD RAID0 setups in my systems for the last 4 years, started off with 2 x 200GB Seagate 7200.7s, then 2 x 74GB Raptors (8MB cache), followed by 2 x 250GB Seagate 7200.10s, am now on 2 x 36GB (16MB cache) Raptors for running the OS on (have used other HDD on RAID card setups), these setups have mostly been on the Intel ICHxR controllers, from ICH5R to the present ICH10R, have also used a couple of (true) hardware RAID cards from Areca.

I also did some experiementation with stripe size, 16KB, 32KB, 64KB and 128KB (also cluster size), of these I found that 32KB was best for overall "general" usage, and this is the stripe size I have used for almost 4 years in every RAID0 array I have setup, however, I don't find that this stripe size is especially prone to fragmentation, which I only find I need to do every 3 weeks generally.

As regards the performance comparison between RAID0 and a single HDD (for OS), the only real increase I have found is in, copying/moving large (video) files around HDDs, and unzipping large (4GB+) WinRAR/WinZip/ISO files. I did not find any real improvement in boot times, and only slightly faster when loading games.

I did/do not find any increase in disc access times between a single HDD and RAID0, for me they basically remain around the same time.

A dedicated (true) hard ware RAID card will give you better performance in write values in RAID5, as the onboard processor will carry out the parity calculations, whereas with an onboard (software) implementaion of RAID5, the actual CPU has to do these which results in quite poor write values. For RAID0 implementation I found not a great difference (in a 2 HDD setup) between a software implementation and a hardware.
 
Recently bought 3x250gb WD re3's as my first raid 0 setup and have been very pleased with the result. Overall everything is snappier especially with startup, opening folders with masses of files (photos music etc), load times are improved but not dramatically, installing anything takes no time at all now. I would suggest it to anyone with lots of non irreplaceable date, otherwise just have another single drive for backup. I chose 32kb stripe size heres my result after a few weeks use http://forums.overclockers.co.uk/showpost.php?p=12862380&postcount=880
 
Thanks for the feedback. There's some useful tips there.

I've done some analysis of my C Drive (OS) and E drive (games - mainly Steam and WoW) and it came up with surprisingly small files:

C Drive - 80% of files are <64kB, 62% are <16kB

E Drive - 85% of files are <64kB, 72% are <16kB

This means that choosing a stripe size of 64kB or larger will only benefit ~15 - 20% of files!

C Drive - most common file size is 1 - 4kB

E Drive - most common file size is 4 - 16kB

From this data, I would be inclined to go with 8kB, 16kB or 32kB stripe size.

EDIT - What's the smallest stripe size ICH10 offers?
 
Last edited:
Can you create various volumes with different stripe sizes?

That way, I could have the OS with a 8kB stripe and the games with a 16kB stripe...

EDIT - Man, I'm really pouring my brain straight onto here whilst reading about RAID at the same time. Too small stripes = too much overhead, making seek time a limiting factor.

Given my drives' content, I'm thinking that RAID 0 is not going to offer much performance at all. The files are pretty small and in order to get any benefit from RAID 0, the stripe size is going to be 1/2 the typical file size, meaning even smaller, which means I'm going to be seek-limited anyway. If my files were consistently 512kB+, I can see the benefit, but with so many files around 16kB or less...

After all this, I think I'm going to stick to two separate drives.

More proof here:

http://www.anandtech.com/storage/showdoc.aspx?i=2969&p=8

(64kB Stripe size)
 
Last edited:
yes on the ich10 you can create more than 1 raid volume with different strips

i've got a 100GB for windows at 32KB strip and the other 1100odd GB at 128KB strip for everything else

I am somewhat curious that your games drive reports so many small files as mine only has 3.1% under 256KB

and my C: drive only has 2.7% under 64KB , are you sure read it right?
 
RAID0 will give you more bandwidth, but will do nothing to improve your seek times (in fact, it'll probably make it negligibly worse). If you're seek-time limited, RAID0 won't help, unless you can do something to reduce seek times. Yes, the lamest performance answer ever comes to mind - defragment your RAID0 stripe.

Also, on new south-bridge fake RAID setups, smaller stripes up to 16KB seem to yield best results. Since most modern disks support transferring up to 16 sectors in a single operation, and since one sector is 512 bytes, 8KB is a reasonable block size to go for. And ICH9R isn't at all bad, especially if you already have it. :)
 
da_ging said:
yes on the ich10 you can create more than 1 raid volume with different strips

i've got a 100GB for windows at 32KB strip and the other 1100odd GB at 128KB strip for everything else

I am somewhat curious that your games drive reports so many small files as mine only has 3.1% under 256KB

and my C: drive only has 2.7% under 64KB , are you sure read it right?
Click to expand...

Are you sure you've read yours right?

Make sure you look at the % of files, not % of total. (Although I suppose we could argue which is a better measure of the correct stripe size...)

Also, remember to sum, e.g. On my laptop drive, 15% are 0-1kB, 12% are 1kB-4kB, 18% are 4kB-16kB, 20% are 16kB-64kB, thus it can be said that 65% of files are under 64kB in size.
 
Basically, the people who say RAID (in an average desktop machine) is "much faster" or they "feel an increase in loading times" are, in most cases, experiencing a placebo effect. Benchmarks tell the real story.
 
Subterranean said:
Basically, the people who say RAID (in an average desktop machine) is "much faster" or they "feel an increase in loading times" are, in most cases, experiencing a placebo effect. Benchmarks tell the real story.
Click to expand...
Benchmarks tell the story of benchmarks.

Real world testing is what counts and it varies for each person. I found with raid that rar'ing and unrar'ing the same 9.5gb test file 2-3 times quicker and maps loading in bf2142 twice as fast not to be a placebo effect.
 
You must log in or register to reply here.
Back
Top Bottom