Sign in with
Sign up | Sign in
Your question

No discernable performance improvement in RAID0 over nonRAID

Last response: in Storage
Share
May 27, 2006 2:06:37 PM

Hi,

Here's the thing:

I got my 2 250GB Hitachi Deskstars, and I thought, 'Groovy, RAID0 here we come'.

I created a RAID 0 array in Windows 2000 with default cluster size, and did a half-life benchmark (start HL2 and start timing... choose new game.... wait for "Half-Life 2" logo to appear and stop timing)... and compared to non-raid it was 30 SECONDS SLOWER!

I rebooted between each timing run to make sure nothing was being cached in memory by the OS.

I then did the same in a RAID0 array with cluster size = 32k .... and I got results 1 second quicker. Just 1 measly second - i.e. nothing.

The same with Oblivion (again rebooting between runs), a game load took 31 seconds non-raid, and 30 seconds RAID0 32k cluster size.

What's going on? I got RAID 0 for the purpose of increasing loading times, what am I doing wrong?

Thanks
May 27, 2006 2:34:50 PM

RAID-0 write times are considerably faster than single drive implementations but read times do not necessarily improve - if anything RAID-1 implementations give better read times because the data can be read from 2 drives at the same time, in RAID-0 half the data is being read from each drive so it isn't necessarily any quicker.

How have you done the RAID array - software through Windows drive management? or have you created it using a raid bios setup (intel, nforce4 or whatever)? Windows 'soft raid' is not as optimised as the 'proper' raid implementations done through the onboard controllers and correct raid drivers.

Matt
May 27, 2006 3:24:51 PM

Quote:
RAID-0 write times are considerably faster than single drive implementations but read times do not necessarily improve - if anything RAID-1 implementations give better read times because the data can be read from 2 drives at the same time, in RAID-0 half the data is being read from each drive so it isn't necessarily any quicker.


You are very confused about RAID and are getting many of the basic facts not just wrong but backwards. I suggest you read some FAQ's, check out some hard drive reviews any maybe checkout HD Tachs's performance library .... and set yourself straight.
---

RAID 0 usually results in impressive sustained read rate increases, for example a sinlge 74 GB Raptor gets 65 MiBps, but two in RAID 0 have a sustained average read of up to 129.5 MiBps.

However
1) Windows Software RAID performs poorely compared to even a cheap integrated raid controller.

2) Contrary to popular belief the hard drive is most often NOT the chief bottleneck when loading games or levels withing a game. I would estimate that for the typical game mabye 25% of the load time is transfering the data off the hard drive and the other 75% is spent decompressing and processing that data.

So hard drive read rate matters, but not nearly as much as people assume.

Applications that don't use highly compressed data or require a lot of processing on startup are the most likely to see proportionate decreases in load time with increase in hard drive read rates.

---

Anyway I suggest you benchmark your system to see what if any performance gains you are getting from software RAID 0 vs a single drive.

Also try look for some benchmarks showing game/level loading times for single drives vs RAID 0 to see what you should expect with a fast RAID 0 setup.

I don't have any bookmarked but people invariably overestimate the impact that RAID 0 has on their gaming performance when they look at the actual numbers.
Related resources
a b G Storage
May 27, 2006 4:08:23 PM

What are you doing wrong? Nothing. Its working just as I'd expect. Perhaps you understand why I don't suggest AID0.
May 27, 2006 5:51:47 PM

The reason I want to use Windows software RAID is because I dual-boot to Linux, and I'd like share part of each hard disk so I can use Linux's own software RAID, not some binary drivers for the nForce chipset. Apparently, Linux software RAID is better than other 'fakeraid' solutions: http://linuxmafia.com/faq/Hardware/sata.html#fakeraid

Wish there was some way to use Linux software RAID through Windows!

Oh well...
May 27, 2006 6:46:53 PM

That link you posted is very interesting.

In Windows Land; integrated controllers vastly outpuerform Windows Software RAID.

My own tests on Windows XP Pro Softwaare RAID 0, 1 and 5 (requires hack) showed horrible performance.

I always wondered if it the integrated controllers were faster because they offered at least minimal hardware support or whether Windows XP Software RAID just plain sucked.

Now I know Windows Software RAID sucks.

---
http://www.anandtech.com/storage/showdoc.aspx?i=2760&p=10

PS

If you just look at the 1x750 vs the 2x750 RAID 0 numbers you can see what impact RAID 0 when functioning at full speeds has on various games

If you read the whole article you will find several uses under which RAID 0 offeres dramatically enhanced performance, gaming just isn't one of them.
May 27, 2006 10:38:43 PM

It's always nice to speak to a knowledgable person about these things. Should have done it before I bought two of the bloody drives!

Nah, seriously though - it may seem a waste that I bought two T7K250s, but but in fact one reason I bought them was because it was a pain in the neck always making DVD backups of my files. Now I can just zip up my stuff and bung it onto the other drive, this'll allow me to make backups a bit less frequently. Yes, RAID 1 would help against drive failures, but it don't help against human error!

The other reason was to speed up game loads in RAID 0 - hell with that now; although the opening screen of Oblivion comes up noticeably quicker in RAID 0, and I have noticed a difference in a few other places, the levels don't load up any faster, making it a bit pointless.

Also, 2 250GB drives are cheaper than 1 500GB drive - 2 250GB drives are almost half the price of 1 500GB drive. So, I guess I'm still happy.
May 27, 2006 11:19:23 PM

Quote:
You are very confused about RAID and are getting many of the basic facts not just wrong but backwards. I suggest you read some FAQ's, check out some hard drive reviews any maybe checkout HD Tachs's performance library .... and set yourself straight.


Yeah, sorry codesmith, you're right, splitting I/O operations over 2 disks does indeed increase performance all round but I'm hungover :p 

I run a RAID-0 array for performance but what I meant is in some testing there aren't consistent access or read improvements, it depends on benchmark/program and test. Obviously 80% of the time you do benefit from enhanced performance on most RAID-0 configurations, especially when copying large files but loading a game is indeed not all about disk access - as you have said!

Quote:
If the system is heavily loaded with lots of I/O, statistically, some of it will go to one disk, and some to the others. Thus, performance will improve over a single large disk. The actual improvement depends a lot on the actual data, stripe sizes, and other factors. In a system with low I/O usage, the performance is equal to that of a single disk. - linas.org


This was more the information I was trying to impart!

Matt
May 27, 2006 11:32:09 PM

The best way I have found out to get get good performance from any raid 0 array is to know the optimum stripe size for the important data.
In my situation dealing with thousands of mp3 files its 128k.

There is a little utility called stripe size calculator,,,I think,, floating around somewhere on the net,,, I have a copy of it somewhere. (google it)
Basically if you already have your data on a single drive, this utility can be run on the drive. It will check all files on the drive & give you a estimated average file size for all files on the drive.
From there you should choose the next lower size & use that to configure the raid 0 array ,,, assuming you are using 2 drives. If 3 drives divide by 3 etc... I have seen siginifant performance in my raid arrays.

I even have a DAW with a 2 drive raid 0 for my OS & programs only & a 4 drive raid 0 for my digital audio files. both have different stripe sizes.
Basically your array has to be optimized/tuned for your specific application & one particular stripe size will not fit all your needs.
May 30, 2006 10:41:08 PM

If you have really important data consider getting EMC Retrospect.

Everytime you backup your selected folders it saves a snapshot, plus it doesn't store identical files more than once, even if you move/rename them.

So it not only protects me against accidental loss of my financial documents, but I can also chose to retrieve past versions of a document or a folder.

Plus I can move transfer the backup sets to DVDs when I feel like it.

Simply using a compression program for backup is ok for some uses. But you won't have a single searchable catalog that will let you select from multiple revisions of the same file.
!