RAID benchmark

[rallynavvie]

I'm going to conduct my own experiment on the performance benefits of running RAID 0 on a computer. I think there are a lot of misconceptions out there regarding the real benefit of running such a volatile disk array. My only foray into the matter was once with two Seagate 80GB IDE drives in which the stripe on one corrupted during a brown-out and lost a lot of data. After that I ran RAID 1 just because I could but then wanted the extra space of the drives seperately.

I just need some input as to how I should set this up and which "real world" benchmarks I should use.

The control hardware: (my secondary video workstation)
Tyan 2460 running two Athlon 1900 MPs
1GB Mushkin PC2100 ECC Registered memory (ECC enabled)
Antec NeoPower 480W behind a CyberPower 1500AVR UPS

The variable hardware:
Two Seagate 18GB U160 15krpm Cheetah hard drives
LSI Logic 21320-R dual-channel HBA

Benchmarks:
Windows install times (from initial format to where it first asks for user input)
Windows boot times (from BIOS beep to login screen)
BF1942 map load times
BF1942 framerates
TMPGEnc encode to same drive
TMPGEnc encode to third IDE drive

I plan on attempting each of the marks (less the Windows install) three times, between each rebooting the PC (of course for the Windows boot time). Would anyone else like to see any other benchmarks run on this system? It isn't a high-end system so most other games probably won't fly, but maybe some other apps? Also should I do one set in RAID 1, another in RAID 0, and the third to a single drive?

I may be doing this next week during the (American) holiday since I've got to go in a re-arrange some hardware in that machine anyway and I've been kinda interested to show "real world" benchmarks on what sort of performance RAID really offers.

[redwudz]

Not really 'real world' and you are probably aware of it, but SiSandra might give you some synthetic benchmarks that are fairly easy to relate to other systems, along with your other more practical use tests.

I've ran RAID0 in the past, but I never captured with that system, so really did nothing to take advantage of any possible speed gain. My array crashed several times. I stopped using it for boot because of that and just used it for editing for about a year. 2X80Gb Maxtors with a on-board Silicon Image controller.

Looking forward to your results.

[shelbyGT]

I have a RAID 0 because that's how it came from the manufacturer. Hasn't crashed yet. I know the risks of it which is why anything of importance is backed up. Getting a laptop soon anyway, so I can copy important files to that.

As for speed... I really haven't noticed anything I would consider huge gains.

[rallynavvie]

I never saw any big gains when I had my setup either yet all these 1337-kiddies keep talking about how "uber-f4star" their computer is because they have RAID 0 with their 7200rpm drives and yearn for two 10krpm Raptors to do it with. I figure running the dual 15krpm with higher bandwidth capabilities than the SATA drives to begin with will keep them from coming back with something like "well yeah, old 7200rpm IDE drives won't see any difference but two 10k SATA Raptors will be t3h fastar!". Synthetic benches are probably the only place you'll see performance differences, thus the "real world" benches I plan to run just with a stopwatch (and my camcorder watching the monitor in case they cry foul).

[arcorob]

Wow ! You want real world data ? Heres some.

I do 70 weddinsgs per year. Spead in rendering , etc is a factor. Spead in general and reliablity of capture.

I am running two RAID 0 arrays on my ASUS P4P800 Delux (VIA RAID)
2 WESTERN DIGITAL 120 GIGS on one side
2 MAXTOR 120's on the other

So 240 by 240 or 480 gig of storage.

I use the first to capture. I then edit and render to second set.

I then create the dvd with second as source and 1st as target for VIDEO TS AUDIO TS and burn from there

Results
Flawless for 2 years +
Speed improvement, tremendous

When I copy disk to disk, render disk to disk or create disk to disk, there isnt a drive , sata or ide that can possbly come close

Now I went even one more step

C: Normal FAT32 MAXTOR 120 Primary IDE
E: DVDRW +- 16X by NEC

D: Mammoth Drive - 2 X 300 GIG SATA 150 in Intel Raid 0 array - Now used for raw capturen (600 gig total)

G: RAid 0 - 2 Western Dig 120's (240 gig total)
H: Raid 0 - 2 Maxtor 120's (240 gig total)


Total of 1.2 Terra Bytes of storage in one sweet Tower

No problems..so dont be afraid of raid zero. Just use it properly and necessary safeguards

[rallynavvie]

And when I set mine up the speed increase wasn't even worth having the striping set up, two seperate volumes ended up being a better option as it was easier to find files.

In your case you needed two striping arrays because in the video world we need to be reading from one volume and writing to another. Most folks aren't looking to invest in four drives, usually just two and leave it at one volume between the two. Reading and writing to that single array is slower than doing so between two seperate volumes because you're losing some performance in the overhead of the controller. And with the rig you're running I doubt you're exceeding a single IDE drive's bandwidth in capturing or encoding. Now if you're doing several HD caps/encodes at once then perhaps you'd be stressing your drives without striping across multiple drives but then your bottleneck isn't you hard drives anymore, it's your processing power.

My experiment is gains from a single array since that's what most folks can afford. The only time I recommend striping is when you're already running high-end hardware (such as an SMP workstation) and need to squeeze every ounce of performance you can from your equipment no matter the cost. In my opinion if you are to go to that extent you may as well spend some more on nested hard drive arrays so that your striped volumes are redundant, not to mention striped across more than just two physical drives.

We'll see some physical results soon enough. I plan on doing this tomorrow. Make sure to get your test requests in before then if you want them incorporated into my experiment.

[dl090465]

all i have to say is, i have 3 computers, two desktops and one notebook. i had to reload the notebook for various reaons during the past 12 months. both of the others are using RAID 0, and i am a believer. one time one of them got corrupted, and the bios fixed it without losing any DATA!!!!!

[arcorob]

Rally...I hear ya but I have 7 drives total, it can be doen with 4.

I) C Drive which you leave inviolate
2) A standard IDE drive for captures (and subsequent work)
3) Two drives striped array - for renders

WHen you write from drive 2, I see a marked decrease in render time as processor tends to be okay in the P4 world but the bottleneck always seems to be IO, so with the raid drive its like a huge catchers mit taking all I can throw at it.

Sisoft sandra also comes up with some pretty impressive times too...

I guess it really depends on need.

Good luck on the testing !!! Try Tmpgenc as a control test for a render

[rallynavvie]

Well I couldn't get it done today. The HBA was hanging after loading its ROM so I couldn't even get into its configuration to set anything up. The 29160 HBA that's normally in that rig doesn't support RAID and there's no way I'm ponying up the big bucks for another RAID HBA so I'll have to swap things around this weekend and see if I can't get it to work.

It's this damn finicky AMD 762 chipset that's always causing problems with that machine. However it should support all of the branded HBAs on the market because that's what it was made to handle.

[SaSi]

The LSI controllers are a messy thing to work with under Windows. Too many compatibility issues. One extreme case is that a particular model would not boot with an ATI VGA and wanted an nVidia to work !!!

Stripping with SCSI drives is a totaly different issue compared to the IDE examples offered by other members. The SCSI drives allow the drive to disconnect from the SCSI bus while seeking, allow the controller to issue multiple requests to drives and queue the responses. All-in-all, SCSI drives can offer performance no IDE setup can.

HOWEVER (!!).
Bullding a high performance disk bucket of, say, 400Gb with SCSI can be really costly. You may achieve double performance compared to a well balanced IDE system, but at 5 times the cost. It's your choice.

My recommendations:
If you stripe, you need a UPS (you seem to know this )

Stripping two IDE disks on a motherboards IDE controller will give you marginal benefits in terms of speed. If you populate all controller ports, the benefits will disappear. (have been there and came back).

Having separate, standalone partitions is a sound choice.

The best performance - save for RAID - is to use Ultra 320 drives @15000rpm with a good, cacheing controller.

You could RAID-0 these SCSI disks for even higher performance. Typically, 2 U320 drives in RAID-0 will give you 1.5 performance increase. 3xdrives will give you 2x increase or better.

Replace old, low capacity 5400 disks with brand new 7200 or even 10000rpm disks.
(I used to make do with an old 40Gb drive as a system disk. Had quite a few of those and used to swap the system disks to switch OS. At some stage, I used a brand new 160Gb drive as a system disk and the system became much faster. Even with 2 Gb RAM - which should minimize the impact of the swap file - swap file access was a bottleneck).

Arrange your work so that video processing and encoding will read from one disk and write to another. Have enough disks so that you can configure applications that make use of temporary files to store them on disks without any data used (best example is CoolEdit PRO).

The benefits from such arrangements are real and measurable and require no other compromises or exotic configurations.

And if you want to experiment with stripping, by all means enjoy the ride!!!