Software RAID0

[LC1104]

Check out the RAID0 Software striping in Win2k over at www.tomshardware.com

Sounds cool. I wish that he compared the results to a two and four channel fasttrak though. Also, I'd like to see some CPU utilization charts. hopefully he'll update. Has anyone tried this out. Any opinions? If it doesn't kill cpu usage, then maybe i won't have to buy a tx4 afterall...maybe...

jason

[Saladin]

Well, it's [software raid, that is] cool in that you don't have to shell out for a raid card.

However......The true benefits of raid arrays really happen (for IDE) when each drive is on a seperate channel. (you know, only one drive per ide cable plugging into the ide connector on either the m/b or a pci card)

And here you can see why it isn't so good:

If you have a m/b with extra ide connectors, most of those have a particular controller. (ie, such as promise 66, or highpoint...shudder) I haven't really had great results with those earlier onboard connectors and win2000 in raid 0 for capture, but that's more due to the limitations of the controller, I believe, and the later boards now have controllers that are capable of raid anyway....

And if your board DOESN'T have extra connectors, you'd be plugging two drives into both connectors on a single cable. As both drives can't use the cable at the same instant, it's nowhere near as beneficial.

I really have to agree with the line that Dr. Mordrid makes on this point: The Fasttrak 100 (NOT the 33 or 66) is a seriously good idea for capture. He uses 4 drives on the TX4 model, which is....kinda cool, but man, that's expensive. I'm trying to get a TX2, which supports 'only' 2 drives.
On the other hand, a single good drive is capable of supporting very high datarates these days, so I'm more of the opinion that SOME of the benefits of a Raid0 array would be
1. Being able to capture more (well, duh, its a bigger drive....that comment doesn't make me a genius, eh)
2. Being able to go past the 80% mark on the capacity with lesser impact....that's a good point...not sure at what point on a raid setup the performance goes downhill at....or if it does.....any comment regarding your 240gig setup and something like HDtach, doc?

Whew...Sorry to rabbit on so much...

End summary: The win2000 raid is....interesting, but due to hardware limitations you're going to find hooking up the physical drives which will result in you buying a pci ide card anyway, I'd get the fasttrak. (100 model only tho....<grin>)

Saladin

'Always remember: Amateurs built the Ark. Professionals built the Titanic'

[Sciascia]

This is what I did. It works like a champ, haven't had but a very small problem and that was my fault.

http://www.tweakhardware.com/guide/raid100/

A little cheaper this way and it is more fun to "make" your own.

[LC1104]

yeah, but I'm interested in getting a TX4 and that's a lot. If I already have two onboard controllers and buy an ultra100 card, for four channel raid, that's way cheaper than getting a tx4 card. The performance looked enticing in the review.

jason

[Saladin]

I'd be careful with that tho.....

I have had a problem with different IDE controllers and mixing drives over them. (okay, one of them was an onboard highpoint and the bx440 on a BP6, so this comment is a leeeetle bit 'not-quite-but-maybe-applicable' kind of thing)

I couldn't get win2000 to stably write to a 4 drive setup in raid0 (or raid 5 for that matter) using 4 identical drives with 1 drive on each controller. (system was booting from a scsi)

I believe this was due to timing issues between the two different chipsets (they are now running perfectly as 2 seperate raid 0 setups, ie two (2x20gig) setups, giving 80gig total

Given that Vidcap requires very reliable writing, any type of minor error in the writing across the drives might result in dropped frames, or better still, the logical drive becoming corrupted and unusable. (that's why I switched to the 2 seperate arrays)

So...while I can't blame you for wanting to save money, just be aware of the fact that problems may arise. Best solution in your case may be an unmodified ultra100 with 2 drives using win2000 for the stripe?

I don't have any real information on how well extended vidcaps at serious bitrates 10Meg+ would go on that type of setup....anyone tried ultra100 software raid? (admittedly, that's pretty much what you asked in the first place....)
Pretty Benchmarking aside, you never know how it's going to work with your combination of hardware/chipset/capture card/drives/software etc etc yawn, until you actually get the #$$@ thing set up. (and then let loose the dogs of swearing)

[Dr Mordrid]

I can speak directly about the TX4 since I have a preview board.

It has 4 master headers, no slaves, so the diminishing returns you see with the FT100 or other RAID cards when adding a 3rd or 4th drive to an array are not a major factor anymore.

It can work in either a conventional PCI/33 133 mb/s slot or in a PCI/66 slot as found on server boards. It'll also be at home in the upcoming consumer boards with PCI/66 slots.

What may scare some folks (unjustifiably) is that it uses 2 IRQ's. This is because it has two ASIC's (controller chips) in order to get four IDE channels. These are connected by a PCI bridge chip that also provides the PCI/66 and PCI/33 interfacing.

Since the TX4 shares IRQ's very nicely this is not really a factor. On my RT-2000 system it shares with both the RT's IEEE-1394 device and the G400 Flex display card with NO problems at all.

Now...as to performance.....

first a basis for comparison;

With a 2 channel Fasttrak100 you would expect this kind of speed multipliers with 2, 3 and 4 drive RAID0 arrays.

2 drive: 2x single drive speed - 2x single drive capacity.
3 drive: 2.5x single drive speed - 3x single drive capacity.
4 drive: no speed increase - 4x single drive capacity.

With the TX4 this is how it tests out on my RT-2000 system;

2 drive: 2x single drive speed - 2x single drive capacity.
3 drive: 2.9x single drive speed - 3x single drive capacity.
4 drive: 3.8x single drive speed - 4x single drive capacity.

Now....a moderately fast drive these days can do ~25-30 mb/s as a single drive. 3.8x that comes to 95-114 mb/s throughput. A really fast single drive can do 35 mb/s or more. Do the math

So, in theory and with fast drives, the current PCI/33 bus can be pretty much saturated by a 4 drive TX4 RAID0 array. What does this mean in terms of real speed? Sit down.....

My array now tests between 95 and 102 mb/s sequential writes & is consistantly over 100 mb/s in sequential reads. These results are being reported by both SANDRA and the RT-2000's drive test.

On a PCI/66 system (ex: Asus CUR-DLS) and using some future 50 mb/s capable drive (not far off IMHO) the array should be able to achieve close to 200 megs/sec.

Obviously a full-shot TX4 RAID0 belongs in a pretty system.

How does this affect real world use? I mean, NO video editing setup needs those speeds for editing, right? Today, no...next spring this could very well be WRONG.

As realtime becomes more capable, particularly software realtime using the new fast CPU's, the need to play multiple video streams simultaneously off the video drive becomes a factor.

When dualstreaming (as current 2 stream realtime cards like the RT-2000 do) you are pushing two video streams, and often an overlay track, off the same drive at the same time. Even DV's paltry 3.5 mb/s all of a sudden needs 10-12 mb/s to do this reliably, and that's doing realtime with just a 3 track stack + a filter.

Coming editing software capable of doing realtime using dualstream plus multiple overlay tracks, filters, bluescreens, PIP's etc., limited only by the CPU used, will send this throughput requirement through the roof.

These products are going to need VERY fast drive subsystems, and they are closer than you think.

Dr. Mordrid

[Flying dutchman]

@Saladin: I LOVE your quote!

[wkulecz]

Doc, 100MB/s+ and mulltiple realtime streams sounds great but you are forgetting that if it takes 8+ milliseconds to move the heads (settling time is the real performance killer) only a small fraction of that performance is actually usable. Seek times have hardly changed at all going from 3600 RPM PIO4 to 7200 RPM UDMA5 drives. IF I recall correctly event the fastest 10000 RPM SCSI drives have seek times of ~6 mS.

For the multiple realtime streams you envision, random access read/write performance will be a better performance indicator than sequential speeds.

--wally.

[Dr Mordrid]

That's what drive and Windows caches are for Wally. They buffer the data while the heads move.

This is a test of that array using the FT100 (haven't saved a TX4 pic yet);



The figures shown for dualstreaming are per stream, so single vs. dual shows about a 1:2.2 ratio.

I would suspect that the ratio for multiple streams (dual A/B + 1 or 2 realtime PIP's with their own clips) using a high quality codec (11 mb/s HuffYUV, 8 mb/s MJPeg etc.) would make someone desirous of a very fast RAID.

Dr. Mordrid

[wkulecz]

Is this Matrox disk benchmark available? It'd be interesting to see what it says about my w2k "stripeset" software raid setups.

I've four 40 GB maxtors as a 160GB stripe and am setting up two of the 60GB IBM drives that may be the same model as what you have.

--wally.

[Dr Mordrid]

Nope, it's not available since it's part of the RT-2000 driver installation.

What were you getting with SANDRA with the 4 Maxtor rig?

Dr. Mordrid

[wkulecz]

Doc,
I've never bothered with SANDRA as I'm not a big believer in benchmarks -- outputting glitch free video from the timelime is the only benchmark I care about!

If there is a link where I can download it, and its not too onerous to install, I'll run it as I think it'd be a useful comparison. I'm curious as to how W2K stripeset software RAID0 stacks up against a hardware solution. I was quite impressed with software raid on four old SCSI drives stripped across two controllers on NT4 but that was back when IDE was clearly inferior to SCSI and everything was too slow for video.

--wally.

PS
Your private mailbox is full

[Indiana]

Doc, could you please post a Sandra Screenshot as well (with the window big enough to see all the single results, not just the "average")?

[dZeus]

Isn't sisoft sandra about as usefull as hdtach for measuring hd performance? i.e. completely useless?

[Indiana]

In my experience Sandra gives a. consistant results and b. results that are in order with the other HD benchmarking programs. So it's not that bad at all. With all benchmarking programs (no matter if gfx, CPU, HD, RAM, whatever) you ALWAYS have to take the results with a grain of salt, 'cause there will always be some special cases where the results are turned into the opposite.
And I defintiely won't call the results of HDTach useless: if one drive gets 20MB/s and the other 40MB/s the second WILL BE FASTER.
Maybe you've always looked only at the average results Sandra gives, this is, of course, somewhat questionable, it's the detail results that matter (sequential vs. random writes/reads, access time).

And it's not even the actual absolute numbers that are important here, it's the relative performance against another system measured with the same program. Since nearly noone - how many RT2000/2500 have actually been sold? - has this Matrox benchmarking program (that is as useless as all the other HD-benchmark tools if you insist on that point of view) I'd simply like to see the results with a program that is available for everyone.

HDTach would be fine with me, as would the ATTO benchmark or ANY other freely available program that gives repeatable results.