RAID 0 - a call for a metareview

[Adcadet]

hey all -
we've discussed at different times the possibility of SF doing metareviews of sorts in which we gather all available reviews of a product/technology, and critically appraise them to come to a sort of final judgement. This is done all the time in clinical research when there are a good number of studies that often seem conflict. I think this type of article is something we SF people are uniquely suited to doing.

I propose we look at RAID 0 setups, especially for OS's. It seems many people/groups think that RAID 0 is the best thing since the 486 for gaming, workstations, etc. Other sites have shown nearly no benefit. I for one am wondering about the performance effects for my personal setup.

Anybody else interested in this?

Thanks,
Adcadet

 

[ddrueding]

Well, if you guys are interested, I can contribute my 2xWD360GDs in RAID0 on 3112R to the mix... I have a couple of them.

As a note, I don't believe this config has given me any more than a 5% boost in anything I do; I'm just too lazy to switch back...

 

[e_dawg]

I recommend the following post as a great jumping point to a bunch of good threads on RAID 0 and benchmarking. IMO, without understanding the limitations inherent in benchmarking and without understanding the theory behind localization, concurrent accesses, and the impact of stripe sizes and additional heads with RAID 0, one cannot do a meta-analysis/meta-review, because there are no criteria against which to evaluate said reviews.

"Required reading": http://forums.storagereview.net/index.php?showtopic=13514&view=findpost&p=143025

But as it stands, I personally feel that benchmarking, especially RAID setups, are fundamentally flawed, as they only test how well a particular drive setup performs for that specific benchmark -- not necessarily real world usage, and not necessarily for the access patterns and HD setup that are specific to each user. For example, if one finds an "excellent review" that shows the Promise FasTrack TX2000 and two WD600BB drives performing splendidly in RAID 0 with 64 K stripes in the IOMeter Workstation test at medium queue depths, what does that really mean to you and your computing needs?

BTW, I apologize for not providing useful solutions to the issues that I raised, but I find it easier to do just that and let more intelligent and more motivated folks do the problem solving. :mrgrn:

 

[Handruin]

If real world tests are in order, what can we use as real world tests?

Valuable data to me would be file serving. What's the performance impact/benefit of introducing RAID 0?

Other possible real world uses would be databases. Analyzing complex queries and inserts into a SQL server/Oracle database setup. (before and after RAID implementation)

Another test could be video post production. What are the pros/cons of capture, edit and render.... Maybe even audio production testing.


Wouldn't that information be useful? I'm sure we can find other items to tag onto the list. We could still run the benchmark's and then associate their number against the real world test.

Setup X produced an I/O meter score of Y
Setup X also produced a real world score in Oracle of Y
Setup X also produced a real world score in File sharing of Y

(Oracle & file sharing testing would obviously have to be defined prior to the results.)

 

[CougTek]

To prove or reject the impact of RAID 0 on games, we would need modern games with heavy textures. I've heard of at least one game for which stripping is supposed to have a noticeable effect. But for most others, it doesn't make a difference.

 

[Bozo]

I am still running a 3Ware 6500 with 2-IBM 75GXPs in my system. These are configured in RAID 0.

I could run some test on this setup if that would help. Maybe ATTO?

I've had this setup since the 75GXPs first came out and I've never really noticed any big speed improvement over a single 75GXP. I copy application and game CDs to the RAID settup and use it to install to the operating system. The main operating system is on a removable hard drive allowing me to test different operating systems. It is definitely faster than CD

Bozo :mrgrn:

 

[LunarMist]

I am still running a 3Ware 6500 with 2-IBM 75GXPs in my system. These are configured in RAID 0.

I could run some test on this setup if that would help. Maybe ATTO?

I've had this setup since the 75GXPs first came out and I've never really noticed any big speed improvement over a single 75GXP. I copy application and game CDs to the RAID settup and use it to install to the operating system. The main operating system is on a removable hard drive allowing me to test different operating systems. It is definitely faster than CD

Bozo :mrgrn:

I can't believe you still have those striped 30GB 75GXP drives chugging along! When was that, Q3 '00?

 

[blakerwry]

I bought two 75gxps in july 2000.. had them striped for over 2 years on promise controllers... have since broken up the array and sold one of the disks... the other disk is in my family's computer and was until recently used for the OS, now it just sits there as a backup if they need anything off of it.

 

[sechs]

To prove or reject the impact of RAID 0 on games, we would need modern games with heavy textures. I've heard of at least one game for which stripping is supposed to have a noticeable effect. But for most others, it doesn't make a difference.

For most games, everything is loaded up when the map starts. For multiplayer, any good game would give no advantage to those who can load the map faster.

So, really, you just change what you're doing when you wait, not that you wait.

 

[blakerwry]

some games, simulators for instance, require that things (parts of a map, textures, models, sounds) get loaded mid-game.

these can cause slight pauses or skips, but are really hard to measure because they effect the overall fps so little.

Like alot of things, it's complex and hard to measure the performance difference going from a single drive to RAID 0 with 2 drives. Some things speed up, some don't. Some speed up in different ways. The things that do speed up aren't alwaways the things we're looking to improve either.

It's easy to see how the low level measurements are affected by RAID 0, but not as easy to see how that translates into an overal experience for the person sitting infront of the monitor.

 

[sechs]

True. But it's not Microsoft Flight Simulator players who are going to RAID 0. It's FPS and RTS players.

 

[Bozo]

My two 75GXPs are only 15Gb each. Bought them February '00? They are still running and haven't started making weird noises yet. But, they do not run 24/7.

Bozo :mrgrn:

 

[blakerwry]

the loading happens in never winter nights and halo... call of duty has it too I think. It doesnt apear to happen in quake, unreal, halflife games... I dont remember it with battlefield 1942(but the game does require a good bit of RAM -512MB to play smoothly).

 

[Adcadet]

metareview (from Adcadet's Instant Dictionary) - a study type that reviews previously published studies but conducts no new data analysis.

 

[Adcadet]

metareview (from Adcadet's Instant Dictionary) - a study type that reviews previously published studies but conducts no new data analysis.

uhh..make that "but conducts no new data collection"

 

[sechs]

the loading happens in never winter nights and halo... call of duty has it too I think.

That's just poor game design. Unless you're optimising your computer for Halo, there's not much reason on that front.

 

[Tea]

What it comes down to, is actual real-world impact on tasks that actually make you wait.

Now this doesn't mean tasks that take a long time, or not always, it means tasks that you actually wait for, which is not the same thing at all.

I'll give you an example. Let's take ... er .... system boot. Sure, it's nicer when my system boots faster, but te reality is, even with a solid-state mega-drive, it's still going to take long enough that someone of my limited attention span is always going to go off and do something else while I'm waiting - look for a spare mango, perhaps. So, in reality, and within reasonable limits, it doesn't much matter how much faster or slower my system boots, because I'm never going to sit there and wait fo it anyway. A similar point could be made about time taken to burn a CDR, or several other tasks.

Here is another example: time taken to load Opera. This one doesn't matter either, as it's so close to instant that I don't notice it, and any improvement really isn't noticable.

(Er ... Tea? You are doing this on an XP2500 and an X15-36LP, you know. If you had, say, a 5400 RPM IDE drive, you might well find that the time taken to load Opera was quite noticable.)

(Oh. Good point, tall grey-furred one. Let's just scratch example #2.)

OK, now an example where performance really does matter. When I'm going through our photo collection, I often open a folder with a lot of sub-folders in it, each one of them containing perhaps 300 files. And I have to wait, typically for perhaps two seconds. That's on an 8MB cache 7200. On the X-15, it's close enough to instant as to make no difference.

Two seconds is slow enough that it bugs me (drives me up the wall in fact), but not long enough to make me walk away looking for chocolate - even I can't find a block of chocolate in 2 or 3 seconds - so I just have to sit and fume.

The point I'm getting at is, for each of us, there is usually a task that really matters, performance wise, and that, in reality, it is that task and that task only that matters to us.

This is why benchmarks are so controversial: we all have different "key performance isses" and thus we all favour different benchmarks.

 

[LunarMist]

What it comes down to, is actual real-world impact on tasks that actually make you wait.

Now this doesn't mean tasks that take a long time, or not always, it means tasks that you actually wait for, which is not the same thing at all.

The point I'm getting at is, for each of us, there is usually a task that really matters, performance wise, and that, in reality, it is that task and that task only that matters to us.

This is why benchmarks are so controversial: we all have different "key performance isses" and thus we all favour different benchmarks.

That is too practical. Unfortunately some persons at some other sites are single-mindedly interested in benchmarks as the task.

 

[sechs]

And they can't seem to admit it....

 

[Santilli]

I don't like waiting, at all.

I'm looking forward to 4 X 15's, in raid 0, running on a Ultra 320 card, and motherboard with no chipset limitations.

I can live with 180 mb/sec, or slightly faster. :mrgrn:


Likewise, I like the idea of dual processors, so I have no waiting for processing tasks.

I've always liked raid 0 boot disks, but, I use scsi Seagate Cheetahs, Granite Digital cables, and terminators, and high quality raid controllers.

But, that's just me...

If it says "Promise" I know that's all it's going to do, preformance wise :mrgrn:

s

 

[Fushigi]

I don't like waiting, at all.

...

Likewise, I like the idea of dual processors, so I have no waiting for processing tasks.

How about one of these, then.

 

[Mercutio]

Greg, what application do you have that requires 180MB/sec STR?

 

[Santilli]

Internet Explorer, Windows 2000, Outlook, Quake III, some golf games, and Photoshop and because I can :wink:

Actually, I have these X 15's setup from my Athlon Box, and they originally ran in a Mac 333 mhz, G3. The difference, with that level of processor, was pretty dramatic, even though the chipset only allowed
75 mb/sec.

I started raiding the first generation cheetahs with 233 mhz G 3 beiges, and noticed a huge difference in preformance.

Since the drives continue to run, I see no reason not to use them as described. It's going to be strange actually having them run at their full capacity, since they have been motherboard chipset, or PCI slot limited
so far, in the Asus A7m266.

I have yet to loose a Seagate Cheetah. It's been 7 years, and the first generation drives still work.

s

 

[sechs]

One must realise that there is a difference between those who do RAID 0 because they can, and those who think that there will be some monumental change in performance.

 

[The JoJo]

[quote="FushigiHow about one of these, then.[/quote]

Oh, goodie, again something to drool over...

 

[The JoJo]

AARGhhh...my bad, sorry...

*mumbling* I think I saw a preview button once....wonder where that was...gotta try it someday.......

 

[Computer Generated Baby]

Would they accept a meta-review of 3-each Colorado 40 MB tape drives running together as a RAID-0 stripe?

 

[sechs]

You can do that?

 

[Platform]

You can do that?

Sure. I was thinking of creating a RAID-0 stripe out of the Colorado 40 MB tape drives with an arc welder.

 

[Onomatopoeic]

Joking aside, yes, you can most *definitely* do Tape RAID with the right software.

I've setup the occasional Tape RAID backup -- usually RAID-0, but also RAID 1 -- using 2 drives (AIT, AIT2, LTO, and I believe a 40GB DLT once a while back).

 

[Santilli]

One must realise that there is a difference between those who do RAID 0 because they can, and those who think that there will be some monumental change in performance.

I have noticed as the processors get faster, and we use more ram, and the other components improve, the impact of the hard drives is a bit less.

However, neglected in this is the increase in hard drive cache. I have 16 mb of cache, and this, along with even a lowly 110 mb/sec is pretty helpful with any kind of hard drive stressing program.

Photoshop comes to mind.

I think I'm going to configure Raid 0 4 first generation X 15's, storage XL 18 Cheetah, and use my third generation 15.3 Cheetah as a scratch disk for photoshop, and, in part, spread the pagefile over the raid, and the 15.3.

I'm pulling the scsi stuff out of my Mac, and putting the drives in my pc for storage, and replacing it with a Quaxtor with 8 mb cache, 160 gig, and something like 9.3 ms access time.

I had setup the machine to do scanning at work, but, the way it's worked out, all the stuff scanned in mac Photoshop 4.0 in bitmap, requires about 3 steps, and doesn't read in Photoshop 6.01 on my pc, even when copied to CD.

I've given up scanning at work mostly, and plan on doing it at home, on the new machine, and archiving here.

Exhausted. time for bed.

s

 

[Tea]

Would they accept a meta-review of 3-each Colorado 40 MB tape drives running together as a RAID-0 stripe?

Go and wash your mouth out, CG Baby! Pleae, no more of the C-word.

 

[Fushigi]

If I had the right software on our AS/400, I could test striping 2 Ultrium2 tape drives. What's included in the base OS will not stripe but will go from one drive to the next after a tape fills up.

Individually, each tape drive has been tested on a non-dedicated system at 203GB/hr (56MB/s). That includes the OS overhead of gathering info about the objects, doing the save, and updating the object's description to note the saved date field.

My guess is that striped I would see around 350-375GB/hr as I'm sure there's some extra overhead involved.

 

[Corvair]

This is the most popular software-based Tape RAID around:

• BrightStor ARCserve Backup Tape RAID Option increases backup performance by striping data across multiple drives. When three or more drives are used, RAID provides fault tolerance, enabling data restoration even if a tape is lost or damaged. The NT Tape RAID option supports 0,1 & 5, mirroring and striping with and without parity. The Option is also capable of supporting individual tape drives (RAIT) or RAID across multiple libraries (RAIL).

http://www3.ca.com/Solutions/ProductOption.asp?ID=2745

 

[Corvair]

Go and wash your mouth out, CG Baby! Pleae, no more of the C-word.

For a number of years, on a semi-regular basis, I used to remind each every person that I knew that went out and pissed their money away buying a Colorado tape drive that this was probaly the worst computer hardware investment they probably ever made.

Most of them eventually agreed with me, especially when they found out just how "reliable" their backups were a few weeks or months later when they really needed them!

And, dare I mention the raw speed at which these contraptions could backup files via the floppy port? Bwoooohaaahaaahaaahahahaha!!!!!!

 

[Mercutio]

My least favorite combination was those stupid low-end Colorado drives combined with Arcserve on a Netware 3 server. If you could get the stupid NLM to not abend your machine, the ^%$^$ing tape drive was almost certain to screw you over.

 

[Santilli]

Well, I've been using one drive, a third generation Cheetah, for booting, in the same machine I've used the 4 cheetah raid 0, on the same ATTO UL3D. There is a HUGE speed difference, in loading applications, program speed, and OS snapiness.

s

 

[Tea]

Wot? You mean it's faster than your tape drive, Greg?

 

[Santilli]

Tea: Yes, just a little bit :wink:

Funny I expected it to be considerably faster, since it should pull 75 ms sec but, on my machine only does 65 mb/s. Those are many atto 250 benchmarks. My raid does a bit more, about 110 mb/sec, and, I guess that is nearly twice as fast. Still, it seems faster then double, if you know what I mean.

Part of it was a problem with some sort of VIA Ultra driver, that slowed the machine to a crawl, but with that gone, it's nice and snappy, but not as fast as I expected. I thought, considering the specs, this would be a one drive raid 0. It isn't.

Benchmarks aren't everything. None the less, this drive does get up and go, for a single drive.

s