x
knightfly 19 March 2002

OK, here is some "real world" (not mine yet though - damn...) info on Raid vs NoRaid -
I subscribe to Maximum PC magazine, which I HIGHLY recommend to anyone who's interested enough to be reading this post - If you can still find it, go out NOW and at least buy the March 2002 issue. The whole issue is aimed at building your own PC, and even though this is a "gamer oriented" magazine, 99% of what they advocate applies equally well to a music PC. Get the March issue - you'll thank me.

Now, the raid stuff - I posted on this about 3 weeks ago, and f**ked it up as usual. After re-reading the text of the article about 5 times, I reversed which results were for write and which for read performance. The title of the article is "Raid revisited", and it's on page 60 of the March issue. Here are the basic details of the test: They used 4 WD1200 drives, which are 7200 rpm drives with 2 MB cache, with an Adaptec 2400A raid controller. This controller also allows raid 5 configuration, which let me know why I don't care if onboard controllers ever use raid 5. (2-3 times as long to transfer 540 MB from Ramdisk to drive, and back.) With the above hardware, the test was this: 768 MB ram, 540 MB of which was set up as a RamDisk. Transfer 540 MB of small, medium and large (their designation, not mine) files from ramdisk to drive(s) and back, reconfiguring the raid and rebooting the machine each time. Record the time taken, make a table. The ramdisk was used to eliminate other devices as the bottleneck to speed, simulating realworld external sources for recording as close as I could imagine short of actually recording x # of tracks simultaneously. For most of us, the read data will be more important than write data, since everybody's machine sooner or later will have to read back all tracks simultaneously to mix down, while most of us will be limited to no more than 16 or less tracks recorded at once. (My assumption, sue Opus if I'm wrong, he's the first one to suggest that I become a moderator here)

Anyway, to the meat (as Monica would say) Raid 0 x 4 (4 drives striped to emulate a single drive) result- best was 35.00 sec. (read) with 64k stripe size. Write under same conditions was 20.33 sec. A single drive came in at 42 sec. read, 28.67 sec. write. (raid5 x 4 took around 70-90 sec. for the same test - phooey!) 42 seconds read (single drive) translates to 12.85 MB/sec sustained, which is good for a theoretical 35 tracks at 32 bit 96k. This is the worst case of the test, ignoring raid 5 tests. The raid config that was best came in at 42 tracks. Write tests were better, by a large margin. However, anyone who's checked out RME's site and read their comments about raid will probably come to the same conclusion I did - by the time you slow the raid down enough to get rid of the PCI bus glitches that impact audio performance, you may as well just use those two connectors as extra IDE channels for drive isolation. Also, raid 1 tests (using 2 mirrored drives) came in at 57.33 sec. for read, 35 sec. for write, which translates to 25 tracks read at 32/96. If you really want to accept 16/44 as a working resolution, the track count should never be a problem - just multiply the above #'s by about 4.3 - 'course, those 150 tracks will add up to crap to the 16th power if you do anything but play 'em back, but wow!

Salesman talk - up to 100 MB/second ! Real World - 12 MB/Second.(IDE)
Salesman talk - up to 80 MB/second ! Real World - 12-13 MB/Second (U2SCSI)
Salesman talk - up to 50 MB/Second ! RealWorld - ? (FireWire)

Granted, that test was with WD drives and an expensive Adaptec pro raid controller, but the single vs 4 stripe numbers should hold true for Maxtors/IBM's and either a plugin or built-in raid controller. Point is, theory doesn't seem to be ANYWHERE NEAR reality here. Jon's comments about REAL thruput are definitely confirmed here, as are my previous experiments with U2SCSI drives. All said, I will still do my own tests once I get Abit and Soyo to give me a manual so I can get my new system beyond drool stage.

Oh, yeah - GET MAXIMUM PC ! It's really refreshing to read a mag that calls a piece of crap "a piece of crap" - I just ignore the parts about video cards and custom wierdo cases, factor in a few of the known no-no's of music computing, and I'm set! Later... Steve

Comments

Tommy P. Tue, 03/19/2002 - 13:54

Maximum PC a great mag, I've been reading it since it was called BOOT. The funny thing that happened over the years is, because they are in print, their info is dated, compared to online sites such as Toms Hardware or DAWworld. They are still on the money though, with a no-holes barred approach to reviews.

I gotta go get my issue.... :D

teddancin Wed, 03/20/2002 - 23:46

Hi all,

All I have to say about IDE raid is: I tried about 6 months ago to get a 2 hard drive (IBM 75gxp's) stripe running properly using EITHER win98 or 2000 (both on board, and 2 different types of controller cards were used), and about 3 months ago (after I had already consulted with some one that knows quite a bit more about computers than I do) I finally gave up.

It was one of THE most frustrating things I've ever experienced in my life, and after all that crap, I still never got it working *properly*. I used two different Motherboards, one tyan, one abit (I'll never buy another tyan motherboard again btw), and I got the same results.

As for people that HAVE got IDE raid working, I've never met one, but I know good ol' Tom got it working from wwww.tomshardware.com , but the results with more than 2 drives (striped) were not that astounding.

I aslo want to add that RAID (striped) is totally unnecessary as long as you have 1 solid HDD that has an average seek time of under 10ms and that spins at 7200 rpm that's at least ata66 with a 2mb buffer.

Secondly, I just want to say that I have read maximum PC from time to time, and I find that there's a lot of bogus information in there. Not to say that there isn't good info in there too, it's just that I find myself NOT reading that magazine more than I find myself reading it because I can never tell if the stuff they're talking about is true or not.

-teddancin

anonymous Thu, 03/21/2002 - 05:41

If raid is for bugs, then....

"For all you do, disk bugs for you!"

(apologies to Budweiser)

Seriously though, I'm running raid on my current PC and will be using it again on the new one I'm building. I use the Promise FastTrak cards and run Raid-1 mirroring, not striping, because I once lost a whole disk full of data and swore I'd never let it happen again. My existing machine has a FastTrak-66 card and a pair of 20GB IBM drives mirrored, my new machine will have a FastTrak100-TX2 with a pair of IBM 80GB drives mirrored. I've had no problems since I always start with a blank empty pair of drives and do a fresh install of the operating system.

anonymous Thu, 03/21/2002 - 06:30

Originally posted by teddancin:
Yup, IDE mirroring works great. But when you're striping..... good luck.

As for the "For all you do, disk bugs for you!", that type of behaviour will not be tollerated in this message board! hehe.
-teddancin

Has anybody here ever tried Raid0+1 with a set of 4 matching drives? (mirrored pair of striped pairs)

knightfly Thu, 03/21/2002 - 07:31

Hey Nick - I haven't tried raid 0+1 yet, but intend to once I get a manual for the two MoBo's I've narrowed things down to - I did a search on Pricewatch.com, for "i845 DDR raid" and it came up with 4 Mobo's - (5 if you count 2 versions of one, with and without USB 2.0) MSI 845 ultra ARU, Abit BD-7 raid, Gigabyte GA8 IRXP, and Soyo Fire Dragon. Of the four, the MSI and Gigabyte use the Promise embedded raid chip, while the Abit and Soyo use the HiPoint raid chip. I have discovered thru reading several manuals on raid Mobo's, that the embedded version of Promise raid does NOT do 0+1 but the HiPoint chip does. This narrowed things down (for me) to just the Abit and Soyo boards, which (naturally) are the two I cannot yet get manuals for (Abit first said they would mail me a hard copy (one month ago they said 3 days, I called them yesterday and the story changed to "sorry, out of stock, call again in 2 weeks" - Soyo told me yesterday that they would mail a hard copy. Hope they are better about it than Abit...

Anyway, I have seen a couple of reviews on both the Abit and Soyo, and both had the crap beat out of them and never flinched. Even though the Soyo is more expensive ($204 vs. $135) it has built-in Firewire, 10/100 Lan and Smart card reader, 2 of which functions I will use, so why take up slots with things you can get on the board? Also, according to the review on motherboard.org the Soyo has a few more overclocking goodies than the Abit, which needs a BIOS update to get rid of its prejudice against Northwood chips. (When you put in a Northwood, the board disables CPU voltage adjustment ??!?)

So, when I finally get time to order, wait for, play with, and document the new machine, raid 0+1 is one of the things I will know more about. I run Samplitude in 32 bit float, and will (once I get the audio hardware) be running at 24/96 in, 32/96 internal and storage, then whatever the project calls for on the output. One of the tests I want to do is take a stereo track, copy it as many times as necessary, and see how many tracks of 32/96 it takes to choke the machine at different disk options (single, 2 striped, 4 striped, 2x2) That way I'll know exactly where the ceiling is and whether raid is worth it, or to just consider the raid option as two extra IDE connectors.

I guess the short answer to your question is "not yet"... Steve

teddancin Thu, 03/21/2002 - 07:32

That would be an awesome setup. The safety netting, plus the performance! I don't see why anyone with that kinda bucks would go for IDE 0+1 when they could buy SCSI.

Not that SCSI is much better at all than IDE, but IF you did have that kinda money in the first place, SCSI is much simpler to setup RAID with, plus you could put all of the drives on ONE SCSI channel even just running SCSI160 as opposed to ultra wide 3. Plus you get quicker seek times with SCSI and better sustained transfer rates.
-teddancin

SonOfSmawg Thu, 03/21/2002 - 20:15

"I don't see why anyone with that kinda bucks would go for IDE 0+1 when they could buy SCSI."
Because usually a RAID mobo is only about $10 more than the regular version. You can get four 40Gb Maxtor 7200rpm 133uata HDDs for about $400. It sounds great if you're on a budget, but if you can get it to work, how WELL will it really work?
It's not the cost, it's how well and reliably it's going to work. This is definately one of those situations where you get what you pay for. I'm ordering an iWill XP333 mobo in the morning, but I'm not going to waste the $10 for the RAID. The puter will be plenty fast, and the only time I ever lost a HDD was because of NETSCAPE, and that will never be an issue again, because I'll never put a Netscape product in a puter again. If you do decent backups, you're never really in a position where you're going to be totally fucked anyway. If you REALLY need the speed and security that a RAID system gives you, then you really NEED SCSI.

teddancin Fri, 03/22/2002 - 00:03

Actually SonOfSmawg, I would totally go for the onboard RAID option whenever I'm buying a motherboard if it's optional. I have a CD burner, and a CD-ROM, and one hard drive and one 250 zip drive right now.
Mathematically that's only 4 IDE devices, but as you don't want anything (exept other equally fast Hard drives) to be on the same channel as your main hard-drive, then that means that I need at least 3 IDE ports.
I always use the RAID ones for my HDD's, then I just put everything else on the others even though I'm not using it for it's RAID functionality.
As for the IDE RAID 0+1 issue. The only people that truly need that are people that CAN afford SCSI. If you can't afford SCSI, it kinda sucks, cause performance is sub-par (to what it should be), and it's a total pain in the ass to get running properly and efficiently.
-teddancin

Doublehelix Fri, 03/22/2002 - 02:54

Salesman talk - up to 100 MB/second ! Real World - 12 MB/Second.(IDE)
Salesman talk - up to 80 MB/second ! Real World - 12-13 MB/Second (U2SCSI)
Salesman talk - up to 50 MB/Second ! RealWorld - ??? (FireWire)

And what the hell is wrong with "Salesman talk"??? (I am a Salesman...and I *never* lie (much) :D . ("Read my lips..."!!!)

I have a RAID controller in one of my servers, and I use it like Teddancin does...as a way to get more IDE drives without sharing channels.

SonOfSmawg Fri, 03/22/2002 - 09:36

Yup, true, you don't want to ssllloooowwwww down your HDD IDE channel with a storage device! Lol ... you definately have the market cornered on storage solutions! Why do you use a zip drive? ... You have a CDRW ...
So, you're both using RAID boards for the extra IDE channels, but not using a RAID configuration. Makes sense ...

knightfly Fri, 03/22/2002 - 09:40

Guise - I'm not rich, I'm just sick of fucking computers and want to make music for a long time before I have to configure one again, so I figure if I set one up as badass as possible NOW, I won't have to fuck with it for 2-3 years, and may even be able(eventually) to remember which end of the guitar goes in your mouth, possibly even how to do hammer-ons and pull-offs on a Hammond... :=)

I have been burned by SCSI twice already, and the only way to get even close to the advertised speeds is a 5 disk raid. Yeah, it's nice to get 7 or 15 devices for the price of one IRQ, tough) For the same capacity as a 4-disk X 80 gb IDE (raid or otherwise), you would need 5 SCSI drives, each of which costs at least twice what the IDE counterpart does. I have a 18 gb U2SCSI drive in my old daw, which is lucky to benchmark at 15 MB/Sec, whereas a buddy bought a newer generic machine with an ATA66 controller and a 7200 rpm ide, and it benches at almost twice what my SCSI does. The capacity I will have, will cost me $1000 for 8 80 gb drives, and I'll be able to try it in raid 0, 1, 0+1, or single. From my research so far, it will probably end up as single drives (since there only seems to be about a 4-5 track difference between the best raid performance and a single drive), which I will use 2 of for online backups and the other two for personal projects. The two primary IDE channels will have the boot/apps drive tied to a "scratch pad" drive, and two removable bays for customer drives. All optical drives will be in the control room on a firewire hub.

The closest SCSI drive (73.4 gb)I can find is a 10k rpm Maxtor for $453 each. (Yeah, I know 10krpm is faster, but)8 of those would cost $3624, plus a more expensive controller, and that difference would buy me (for example) a second DM-24 with cascade kit AND meterbridge, and most of the extra RME stuff to set it up at 24/96. Since the only Video I intend to do is enough for scoring picture, I can't justify the extra expense for SCSI. Plus, at $125 for 80 gb, it almost isn't worth it to use DVD's for backup. Just buy another drive and stop waiting overnight for backups.

That's my story and I'm stickin' to it (at least for now)

On a different note ( B#) Power supplies - just noticed last nite that the regulation specs on a PC Power & Cooling 450 watter are 10 TIMES as good as the Enermax, for me that's worth the extra $. Damn cheap keyboards, post my blather before I be all blathered out... Steve

LittleJames Fri, 03/22/2002 - 09:44

If yer wondering about RAID 0+1 configs check out the nuendo board. Alot of folks that use RAID arrays have posted alot of valuable info on set up, controllers, etc...

Ultra160 is great but isn't very cost effective in terms of GB/dollar ratio. But hell if you got the cash you might as well buy a few ultra160 drives and get a nice RAID controller and look out :D

teddancin Fri, 03/22/2002 - 11:22

Well, sounds like you've got your mind made up knightfly. I hope it all works out for you and you don't waste a good 3 months screwing with it like I did. My best advice to you would be: Try to get it working properly for about 2 weeks tops, if you can't by then, cut your losses.

As for SCSI, it's damned sweet, but in no way is its price justified. I would almost never buy a SCSI setup unless I had so much money that it didn't matter one way or another, or if something was SOOOOO critical (cure for cancer/aids) and I had to store it on a hard drive.

As for my Zip drive. My CD-R (at 24x burn) takes less time than it does to move the contents of a full zip drive, this is true. But, with a zip drive. I never have to buy new stuff for it, ever. And I like for people to give stuff to me (vice versa too), but if they don't have a CD-R, then I'm up pooh river. Unless they have a zip drive, then I just loan them my zip disk and I'm good. In other words, just more accessibility options for me.

-teddancin

Tommy P. Fri, 03/22/2002 - 12:46

On a different note ( B#) Power supplies - just noticed last nite that the regulation specs on a PC Power & Cooling 450 watter are 10 TIMES as good as the Enermax

Now there's the place to put hard earned money. Years ago, I bought a high quality 400watt P/S when the one in my Celeron300 rig died. It's been powering everything I've thrown at it since. ;)

anonymous Fri, 03/22/2002 - 13:51

Originally posted by LittleJames:

Ultra160 is great but isn't very cost effective in terms of GB/dollar ratio. But hell if you got the cash you might as well buy a few ultra160 drives and get a nice RAID controller and look out :D

If money wasn't a factor, here's the ultimate SCSI storage device for a DAW, [[url=http://[/URL]="http://www.soliddat…"]solid state disks!!![/]="http://www.soliddat…"]solid state disks!!![/]. It has an internal normal hard drive for keeping your data persistant and with an access time of 14 MICROSECONDS and sustained data xfer rate of 40MB/s, could you imagine how fast this sucker would be? The problem is that it only costs around $75,000.00 :eek:

knightfly Fri, 03/22/2002 - 14:16

Yeah, TD, 2 weeks is at least a week too long. From what I've seen about raid vs. single disk, I'm figuring on 8 individual drives, then just dragging one to the other for backup after the sessions - I will try the raid thing just to know for myself, but the track count difference probably won't justify the hassle.

LittleJames, thanks for the headsup on the Nuendo board, I usually don't go there but I will now.

Tommy - I was gonna go for the PCPower 450 anyway, just hadn't noticed the GLARING difference between SUPPOSEDLY high end units. BLEHCCCHHH!!! Now I'm DEFINITELY payin' the extra...

Nick, if money wasn't a factor, it'd be because I'd had a sex-change operation and talked Bill Gates into dumpin' his wife (like she married him for love or looks anyway...) Still, an 80 gig ramdisk WOULD be cool, til the asshole across the street got a 160 gig one... Steve

teddancin Fri, 03/22/2002 - 21:49

That's awesome Knightfly. I really hope it works out for you.

That stuff's tough, but if you DO get it working properly, you should have a pretty RAID (rad.. bad joke) setup.

Just so you know (when you DO get all your gear and you're testing it out). Raid 0 should be about twice as fast 1 hard drive. Also, you should have a relatively low CPU usage.

On my dual Proc. system, I usually get 32,000 KB/S with 0 % cpu usage with one 7200 RPM IBM 75 GXP hard drive in win2k (win98 should be a little faster, but don't be fooled, 98 still sux, hehe). I know a lot of people use Sysoft Sandra for benchmarking, but I've found that they're not all that accurate sometimes, so I use this Bench program..... it's really solid.

;)

Remember to install service pack 2 (after you install all your other drivers) because SP2 enables ata100. Have phun and good luck!

-teddancin

knightfly Sat, 03/23/2002 - 09:20

Backatcha, TD - Although, if the MaxPC guys didn't mess up their "raid revisited" piece, I may only get about 20% faster with striped drives than a single. That would be borderline worth it, as long as I can get 32 tracks of 32/96 off disk. If any SW can do it, it would be Samplitude, especially with a gig of DDR at 552 mHz - Samp. uses what they call "track speed", which is a glorified ramdisk built into the pgm. Not real good for fast response to commands during playback, but solid for getting best glitch-free playback on high track counts. They pre-fetch audio in chunks and dump it into ram so it can play from there, so any sags in disk thruput get smoothed out. Works great so far. Plus, you can turn it off to see what your disks can really do in a bind. Gotta run, getting the final Pricewatch printouts ready to lock down prices/vendors, etc... Steve

teddancin Sat, 03/23/2002 - 09:40

Well, I'll tell ya something. If you can get RAID 0 working properly (which I couldn't), it should have no problem doing 60,000 KB/S. When I was trying to get it to work, one of the ways I set it up was in win2k's internal SoftWare RAID settings, and I got it to go at around 80,000 KB/S Max Write speed. Really fricken fast, but being that it was Software Raid as opposed to my HardWare card running it, the CPU usage was like 15 percent. I'd rather have a HDD that's twice as slow and an over all computer that's much snappier than run the SoftWare RAID. So it's definately possible to get RAID 0 to go twice as fast as just one hard drive.

Tommy P. Mon, 03/25/2002 - 17:36

I never got into this RAID debate 'cause its a can O worms. Don't forget, when SCSI drives were maxing out at 5mb/sec, and IDE was anemic at 3, RAID striping(0) was necessary. Not so anymore.
And unless you have a real hardware card(the Promise Fasttrack is not, the low end Highpoint, and Megatrends are not, they are all software based controllers), you are wasting CPU cycles fer nuttin.
The default setup on all these cards is optimized for slow drives, so with current speedy 7200rpm drives, you have to hit the right combination of stripe size/cluster size to see any improvement.
I spent a friggin weekend a few years back formatting and reformatting and re-installing the OS to see exactly WTF was going on(with a Promise card and later with a AMI). First off, the CPU usage is inversely proportional to the stripe size :D
My .02

Tommy P.

knightfly Mon, 03/25/2002 - 21:40

Tommy, if you've read my posts you'll see we pretty much agree - When I finally make the plunge, I expect to end up with 8 separate IDE drives, 4 ATA 100 and 4 ATA 133, with 2 of the ATA 100's on removable Dataport stuff. I do want to at least experiment for myself, but not for any great length of time. The one test I read in Maximum PC claimed that mirroring was really slow, like 57 seconds to read 512 MB of mixed files instead of 35 seconds. They found 64k stripes fastest for read, and 128k stripes fastest for write. A single drive came in at 42 seconds, vs. 35 for their best 0x4 raid time. For the hassle involved, I doubt I would bother with raid for that little bit. 42 seconds is 12.19 MB/sec, which translates to 33 tracks at 32/96, and bunches more at lower rates. Not too bad for a single drive. I will be testing for performance differences with just one or both devices connected on a single IDE cable also - I think that since this machine will have no MIDI except for sync, I will be fine using single drives. 'course, some people think they will like brussel sprouts too, til they actually TASTE them... Steve

anonymous Tue, 03/26/2002 - 18:39

Originally posted by Tommy P.:
I'd like to run a RAID server in another computer and stream files from my audio app across my home network to the server. Does anyone know how to do this? OPUS! :D

With a 100Mbps full duplex switched network link between your DAW and the fileserver, you'll be lucky to get the equivalent of just under 10MB/s sustained thruput between your audio app and the "network drive" shared off your typical Windows NT or 2000 server.

You'd need to have switched full duplex gigabit ethernet between your workstation and the server to get data xfer thruput that would rival that of a local ATA/100 disc drive. If you wanted to pour a lot of money at such a system, just any decent fileserver (like a Compaq Proliant or equivalent powerful other brand) with a big raid array in it (and using a hardware caching raid controller) and a switched gigabit ethernet link would do the trick. In my professional employment, I'm a network administrator who runs a network of about 30-some-odd various servers and about 500 workstations for a living and I get the kind of data thruput that would be suitable for DAW purposes from our Compaq Proliant ML530 dual Xeon servers with 15K rpm Ultra160 LVD-SCSI Seagate drives in raid arrays using the Compaq controllers and IBM p660 RS6000 machines with SSA disk arrays and run a Cisco Catalyst switched network with gigabit backbone. Of course we run huge databases and client-server apps on this platform, not DAWs :D

anonymous Wed, 03/27/2002 - 13:28

Originally posted by Tommy P.:
Thanks Nick, I really appreciate the info. I may try the 10/100 setup, just to get my feet wet. Then when gigabit equipment comes down in price, I'll give it a shot. :D
Any quick tips on set-up? I'm running Win2K Pro and WinXP Pro on my computers at home.

Tommy P.

Get good 10/100 network cards... the el-cheapo ones tend to make the CPU do a lot more work in software in the drivers, rather than doing the work in hardware in a chip on the card (a lot like some soundcards, eh?). My favorite 10/100 card of all time is the good old Intel 10/100 Pro E100B card with the 82557 chip. Anything with a TI "Thunderlan" chip on it is very good too. If you possibly can, then give the network card a dedicated IRQ, just as you would give to an audio card or a SCSI card. Don't use a hub or a "crossover" cable to connect your workstations together, use a real ethernet switch and make sure it'll support full duplex operation... a hub or crossover cable can only do half-duplex and a crossover cable cannot resolve packet collisions, causing your network card and software driver to do more wasted work. A cheap switch should be less than $100 for a 4 to 8 port unit. After you install the nic drivers (ha!! LOL, now you know where I got my internet "handle" of Nick Driver!!!! --not my real name :p ) you'll need to manually set the properties of the driver to use speed=100 and duplex=full, because when they are set to auto-sensing, you usually get half duplex and 100Mbps. Use good ethernet patch cords, at least CAT5... the inexpensive "Belkin" brand commonly seen at computer stores usually work pretty good.

Peer-to-peer sharing of drives or folders within Win 2000 Pro or XP Pro workstation works just as good as the NT4 or 2000 Server products, you just can't have as many shared connections between machines (5 to 10) and you use a "workgroup" security model. There are a bazillion How-To sites on the web with basic instructions on how to configure Windows 2000 and XP machines for a small network. Here's one: [[url=http://[/URL]="http://www.homeneth…"]Home Networking Tutorial[/]="http://www.homeneth…"]Home Networking Tutorial[/] A Google search will reveal dozens more similar sites.

x

User login