Welcome to the MacNN Forums.

Tyler McAdams · Jul 23, 2004, 03:56 AM

Anybody know a good 64bit/33mhz or better *hardware* raid card for OS X? It needs to be bootable in raid 0... thx

Detrius · Jul 25, 2004, 10:46 PM

Originally posted by Tyler McAdams:
Anybody know a good 64bit/33mhz or better *hardware* raid card for OS X? It needs to be bootable in raid 0... thx

ACARD makes some RAID serial ATA cards. Check into them.

dtriska · Jul 25, 2004, 10:53 PM

http://www.sonnettech.com/product/tempo_raid133.html

Tyler McAdams · Jul 25, 2004, 11:07 PM

I'm actually looking for a ultra160/ultra320 scsi card... thanks for the help though... I should have put that in the post, sorry...

l008com · Jul 26, 2004, 03:54 AM

I've had shitty luck using SCSI under OS X. I finally gave up and use IDE & SATA RAIDS

Tyler McAdams · Jul 26, 2004, 04:46 PM

SATA is just not fast enough for my needs right now... I need 1500rpm drives in raid0 array.

milhous · Jul 26, 2004, 11:49 PM

here you go:

http://attotech.com/epciview.html

problem solved.

28 posts to go...

Tyler McAdams · Jul 27, 2004, 12:32 AM

Those are software raid driven.... I've got a couple of those cards actually. Thanks for the work though!

Spliffdaddy · Jul 27, 2004, 12:43 AM

Originally posted by Tyler McAdams:
SATA is just not fast enough for my needs right now... I need 1500rpm drives in raid0 array.

Current 10,000rpm SATA drives are just as fast - if not faster - than high-end 15K rpm SCSI drives. Western Digital's Raptor series (SATA) boasts 2.9ms latency and 55MB+/sec. Most 15K SCSI drives are *higher* than 3.5ms with a similar data throughput rate.

Other than the ability to address more than 1 device per channel, SCSI offers zero benefit over SATA. Keep in mind that while a typical SCSI interface can address 7 devices per channel - it can only communicate with *one* device at a time. You'll still need to use a 2 channel SCSI RAID card with a hard drive on each channel in order to have an effective 2-drive RAID 0 (stripe).

So you might as well save $650 by using a SATA RAID card and a pair of 10K SATA drives, instead of going with SCSI. Because that's the price difference - and there's no performance difference. None. If you want to go *faster*, the price difference is more than that.

SCSI is archaic, and fiber channel is only used when poor planning leaves no other alternatives for harddisk expansion.

Is there such a thing as a bootable SATA RAID card for Macs? I dunno.

Tyler McAdams · Jul 27, 2004, 01:20 AM

Originally posted by Spliffdaddy:

Other than the ability to address more than 1 device per channel, SCSI offers zero benefit over SATA. Keep in mind that while a typical SCSI interface can address 7 devices per channel - it can only communicate with *one* device at a time. You'll still need to use a 2 channel SCSI RAID card with a hard drive on each channel in order to have an effective 2-drive RAID 0 (stripe).

This is not correct... SCSI can address up to 15 devices per channel and can communicate with all of them at the same time. This is the difference bt *regular* ATA (not serial) and SCSI... ATA can only talk to one drive at a time. This is why you see raid scsi card with only one physical channel.... it only needs one since it's a serial connection.

Spliffdaddy · 12:24 PM

no, I'm still correct.

actually, it's 8 devices addressable per channel, but the controller counts itself as a device - so that leaves 7 devices that can be added.

You can connect 2 SCSI channels into a loop and get the ability to address 15 devices (16, minus the controller).

Communication on SCSI is one device at a time per channel.

just spend the extra $600 and go with SCSI.

That way, people who don't anything about it will be impressed.

Bottom line: you just want to move 100MB/sec+ through a hard disk (or array) that's bootable.

Why is it important that the RAID 0 array be bootable? Yeah, it would be nice - but it would be a lot less expensive to use a 'software' stripe array that wasn't bootable.

I'd get a 2 channel non-RAID SATA controller card and run a 3-disk 'software' stripe (135-160MB/sec from what I've seen in a Windows environment). It would totally destroy any 2 disk 15k rpm SCSI 'hardware' stripe. And all I'd need is 3 inexpensive SATA drives and a cheapo controller card. I'd boot from the 'good enuff' original SATA hard drive. Perhaps you could make OSX use the striped disk for caching and virtual memory. Well, everything possible, anyways.

Devin Lane · Jul 27, 2004, 11:12 AM

Here is a link to a bootable, up to 4 drive hardware SATA raid for Mac:

Serial ATA RAID

However: there are two problems with this:

1. G5 Only (case design)
2. Mac OS X 10.3 Panther + required for booting.
3. Price: $2,199.95 for the RAID card and 4 250 GB drives.

Still, it does get 230MB/s.

Be pretty sweet if you got it.

Hope that helps,

Tyler McAdams · Jul 27, 2004, 09:41 PM

Originally posted by Spliffdaddy:

Communication on SCSI is one device at a time per channel.

Please prove this to me... where did you get this information... I'm not saying you're wrong... I'm saying I need proof then.

l008com · Jul 27, 2004, 09:51 PM

The way I understand it, IDE talkes to when drive, then when its done, it talks to another drive. SCSI Can talk to one drive, then go and talk to another drive, without waiting for the operation on the first drive to complete. So there is an advantage over IDE, though you can't really read or write to all of the drives on a SCSI bus at the same time, but it would still be faster than IDE if you striped the drives all on one channel. That being said, I reccomend a hardware non RAID SATA card, and a software raid. I run my G5 off of a fully bootable software raid using the two SATA drives it came with, and it works great. I just wish I had some more hard drive bays. The bottom line is, no matter which is faster, the price difference is so huge, for the the same price you could always build a faster SATA array, even it that means getting two PCI cards and running like 6 drives for the price of 4 or 3 SCSI drives. But software RAID is really the smart way to go on OS X if your looking for just RAID 0 (thats stripe right?). Thats what I've used on my last two macs, this one SATA and the last IDE (two different buses, one drive per bus) and its a very nice little performance boost.

Tyler McAdams · Jul 27, 2004, 10:13 PM

Well right now I've got a DP533 Digital Audio... I've got a 64bit/33mhz attotech ultra160 card that has 2 atlas ultra160 dirves in software raid0 as my boot drive... I just bought 2 more atlas ultra160s that I'm going to add to the card for a total of 4 ultra160's in raid0... Basically I'm looking for a little more performance via hardware raid.

l008com · Jul 27, 2004, 10:18 PM

Thats a pretty damn slow machine do be aiming for such super fast hard drive space. Why not sell all the SCSI and the G4 and get a load end G5 with a pair of SATA drives raided.

Tyler McAdams · 01:31 AM

Originally posted by l008com:
Thats a pretty damn slow machine do be aiming for such super fast hard drive space. Why not sell all the SCSI and the G4 and get a load end G5 with a pair of SATA drives raided.

I would like to but I don't have that kind of money right now... so what I'm doing is upgrading this system... I just bought a dual 1.33mhz G4 upgrade for it as well. I bought it via apple load and have not finished paying for it yet.

l008com · Jul 27, 2004, 10:31 PM

Wow thats a rough situation your in. I had a dual 533, but i upgraded to a dual 1.25, and recently again to a dual 1.8
I think your best bet is to max your ram if you haven't, and save the rest of your money (honestly i think you shouldn't have spent anything on it, ebayed it all, you can get a dual 1.8 for $1799 with an edu discount)

Tyler McAdams · Jul 27, 2004, 10:46 PM

Originally posted by l008com:
Wow thats a rough situation your in. I had a dual 533, but i upgraded to a dual 1.25, and recently again to a dual 1.8
I think your best bet is to max your ram if you haven't, and save the rest of your money (honestly i think you shouldn't have spent anything on it, ebayed it all, you can get a dual 1.8 for $1799 with an edu discount)

Yeah the ram is maxed out... the base system was 4k... I'll have it paid off in about 2 years. Sucks but I do most of my work on commodity x86 hardware. I love OS X though for web design and DV work.

l008com · Jul 27, 2004, 10:51 PM

base system was 4K? My dual 533 was only $2500 i think. You never want to get into a comptuer for that long, its going to be so old by the time your done paying for it, it won't be worth anything.
If you upgrade every 9 months or so, and sell your old machien on ebay, you'll end up with $500 net new systems. Thats what I do and it works out great. Once you fall too far behind, you get stuck.

Tyler McAdams · Jul 27, 2004, 11:36 PM

It was a custom configuration. Money's tight right now.... once I get my car paid off I'll be back in line.

l008com · Jul 27, 2004, 11:39 PM

Wow you're just Mr Debt now aren't ya!

Spliffdaddy · 11:57 PM

Originally posted by Tyler McAdams:
Well right now I've got a DP533 Digital Audio... I've got a 64bit/33mhz attotech ultra160 card that has 2 atlas ultra160 dirves in software raid0 as my boot drive... I just bought 2 more atlas ultra160s that I'm going to add to the card for a total of 4 ultra160's in raid0... Basically I'm looking for a little more performance via hardware raid.

Um...I've done that before and it doesn't work. Well, it works, but there's zero performance gain if you have more than one disk per controller (aka, channel). You need a new channel for each disk you add to the stripe. Since the controller can only read/write to one disk at a time, adding 2 disks just doubles the amount of time it takes for the controller to move the data. And that negates the point (speed) of adding another disk.

Try it and see. Make a striped array from 2 disks on the same channel (IDE or SCSI). Not any faster than a single unstriped disk is it? Because it can't read/write on both disks simultaneously.

Move the disks to 2 separate channels and the throughput will double.

Tyler McAdams · Jul 28, 2004, 12:29 AM

You're wrong about scsi... ATA true but not scsi... there is no performance penalty for more than one device per channel.

Here's a single channel raid card:
http://www.adaptec.com/worldwide/pro...AID%2fUltra320

Why would they make a single channel raid card if it can only read one disk at a time? ATA yes, scsi no.

l008com · Jul 28, 2004, 12:39 AM

Becuase what I said was right. Which is exactly inbetween what you guys said. Also a single channel raid card would be perfect for someone who needs a more affordable mirrored raid and doesn't need side splitting speed.

Tyler McAdams · 12:56 AM

Originally posted by l008com:
Becuase what I said was right. Which is exactly inbetween what you guys said. Also a single channel raid card would be perfect for someone who needs a more affordable mirrored raid and doesn't need side splitting speed.

I agree that what you are saying makes more sense... I've asked this question time and time again on many forums and the answer has always been that there is no penalty with raid for multiple drives on a single channel/chain... but there is for ATA (regular ATA). Since so many people have said this I'm biased and need more proof... not to start a flame... I'm after the truth.

I should add that I do backups all the time on windows systems that read from one scsi drive on a chain and write to the other... the speed is always extremely high compaired to the same back up from scsi to the ATA on the same machine... average speeds of 1300 MB/min

l008com · Jul 28, 2004, 12:55 AM

Well since YOU are the one that needs the proof, i suggest YOU get to searching...

Tyler McAdams · 01:08 AM

Originally posted by l008com:
Well since YOU are the one that needs the proof, i suggest YOU get to searching...

don't worry I am already.

Here's my email to attotech support... hopefully they can clear this up:

Support:

I need to ask a question to you concerning using your product and software raid 0 configuration. There is a debate right now concerning SCSI technology, raid 0 software vs ATA raid 0 that I have asked at macnn forums:

http://forums.macnn.com/showthread.p...68#post2101368

The question is simple: Is there a performance penalty for a raid 0 configuration with 2 drives stripped in raid 0 on the same channel? It has been said that ATA raid 0 on the same channel suffers due to only being able to address one drive at a time... does this exist with SCSI raid? I'm using your 64bit/33mhz card on OS X with OS x's built in software raid 0 with 2 atlas utlra160 drives currently.

Thanks,

Tyler McAdams

Spliffdaddy · 01:51 AM

PS, there can be only one device per SATA channel - so this isn't an issue. On a PATA interface (aka,IDE & ATA) there can be 2 devices per channel (master/slave) - but (again) you will see zero benefit to striping 2 drives on the same channel.

I'm not lying about SCSI communicating with one device at a time on each channel. There is no simultaneous access to multiple drives on a single channel (controller).

Just try it and you'll see. It didn't make sense to me until I tried it myself.

Tyler McAdams · Jul 28, 2004, 01:50 AM

Originally posted by Spliffdaddy:

I'm not lying about SCSI communicating with one device at a time on each channel. There is no simultaneous access to multiple drives on a single channel (controller).

I would never call you a liar... why would you lie to me about raid? I think you might be a bit mistaken... perhaps due to a poorly performing raid controller. Then again you might be completely right and I've been completely mislead.

Here's something I just found on ars technica:

http://arstechnica.com/paedia/r/raid-1.html

" As a result, system I/O performance improves greatly because the data is spread out over X spindles, on possibly more than one channel. Better performance will be observed as the physical disk-to-controller ratio approaches one. But then again, so will the cost."

That's down beside the picture... It's sounds more an more like l008com has the correct idea.

Spliffdaddy · Jul 28, 2004, 02:02 AM

Sorry, didn't mean to suggest anybody called me a liar.

I've done all sorts of experiments with all sorts of SCSI controllers and disks (up to the point of saturating the PCI bus). My observation has been that there is not only zero benefit to having more than one disk per controller - there is even a slight penalty. So, no, I do not agree with the substance of the Ars article in that regard. The author never actually tried it - he merely assumed what he believed would happen.

Again, try it for yourself. That was the only way I was ever able to get an answer to the question.

Tyler McAdams · Jul 28, 2004, 02:18 AM

I'll be running tests... probably with xBench once I get my 2 extra drives in. Hopefully they will arrive this afternoon or tomorrow. I'll be sure to post the data of my findings. I had planned on photographing to entire upgrade process (cpus and drives) and posting them to my website for kicks anyway.

This whole thread has been very educational... I'm glad I asked this question, otherwise I would be mislead, it seems.

Spliffdaddy · 02:40 AM

I asked this same question a while back.

Most folks said it would work.

A few said it wouldn't work.

One guy told me to just try it and see what happens.

Over the years I've tried and re-tried. Seems so simple now.

I never learned as much by reading as I did by doing.

Disk striping and SMP (multiprocessor) were difficult things for me to grasp, at first. So much misinformation out there. I was a bit let down once I finally understood the limits, benefits, and operation of those things. There wasn't any magic, sadly enough. I learned by doing and observing the result. Life experience is the only thing you can put real faith into.

Tyler McAdams · Jul 28, 2004, 02:38 AM

Very true... I've found that nothing can replace the experience of real world data... I've found many instances where I have been mislead by simply relying on benchmark applications (sandra pro is a pos, fyi). Bottom line I'm trying to to squeeze as much as I can out of my dying system... with out breaking my bank account.

someone_else · Jul 28, 2004, 12:31 PM

SCSI has command queuing, where it can issue multiple requests w/o waiting for a response from each one. Therefore, there can be multiple outstanding requests on the SCSI bus. With ATA there can be only one outstanding request, and another can not be issued until the reply is recieved.

Therefore, it is possible for SCSI to "address" more than one device at one time, but overall throughput on a single channel will probably be a little slower (a lot if one of the devices is running an older protocol version) since all devices still compete for the same bus bandwidth. You may see a little improvement if the HD's are slow and take a while to respond, but with 15K drives that probably won't be too often.

I think this is what l008com and spiffdaddy were trying to say.

HTH.

badtz · Jul 28, 2004, 01:02 PM

Doesn't SATA support Command Queuing? [though I don't think the current G5's support this feature]

????

l008com · Jul 28, 2004, 02:27 PM

Originally posted by badtz:
Doesn't SATA support Command Queuing? [though I don't think the current G5's support this feature]

????

Doesn't SATA ONLY support one drive per bus, so it doesn't really even matter?

Tyler McAdams · Jul 29, 2004, 02:52 AM

Originally posted by badtz:
Doesn't SATA support Command Queuing? [though I don't think the current G5's support this feature]

????

Yeah it does... and a few older ATA drives did also. That and hot swappability... from scsi. But new is one drive per channel.. no more master/slave.... which is obviously a good thing. Who the moron was who named it master/slave I don't know.

Tyler McAdams · Jul 29, 2004, 07:39 AM

Originally posted by l008com:
Wow you're just Mr Debt now aren't ya!

I can't tell if you're genuine or if you're making a joke here... either way please do me a favor and keep your comments on subject please. There are people here that can validate the fact that I do take well to abuse. Thanks.

Tyler McAdams · Aug 1, 2004, 11:07 PM

Well, I finally got my 2 new drives in and I finished installing them today... the array is surprisingly fast and xBench scores are about 40% higher than the original raid 0 array. Thanks for all the help!

badtz · Aug 2, 2004, 03:02 AM

Originally posted by Tyler McAdams:
Well, I finally got my 2 new drives in and I finished installing them today... the array is surprisingly fast and xBench scores are about 40% higher than the original raid 0 array. Thanks for all the help!

how do you have the array setup?

which drives?

details

Tyler McAdams · Aug 2, 2004, 04:14 AM

Most certainly!

The drive array consists of 4 Atlas IV Ultra160 68pin drives... all connected via one 4 device chain.

The card is a BTO Attotech 2 channel card that is oem and can not be bought but through Apple (the card I had configured with the workstation when I originally bought it) The card has 2 channels... one internal and one external... the drives are on the internal channel.

The drives are configured with OS X's built in software raid.

Some issues I had making the arrary were losing jumpers while trying to slide the drives in to the computer and then getting the device chain in the workstation so that the G4's door would close all the way.

Right now it's pretty cramped so I'm going to take the current build apart again this morning and try to fit the drives in better.

Spliffdaddy · Aug 2, 2004, 10:56 AM

badtz · Aug 4, 2004, 06:19 PM

Tyler ...

now that you've had time to play with your new setup ...

how is the portion that's software raid'ed? [raid 0?]

do access times improve? or just the STR? how's the improvement with STR compared to a single disk [in your circumstance]?

Tyler McAdams · 02:00 AM

Originally posted by badtz:
Tyler ...

now that you've had time to play with your new setup ...

how is the portion that's software raid'ed? [raid 0?]

do access times improve? or just the STR? how's the improvement with STR compared to a single disk [in your circumstance]?

It's not as bad as I thought it would be... I wanted a hardware raid setup... but this is working just fine for my needs... application launch times and anything that takes a lot of "sustained disk bandwidth" usage is improved. These disks move at 7200rpm... although slow for for top of the line ultra3/160 (10k to 15k) the info is striped across 4 drives... which means everything is pretty much at the outer tips of the disks, which improves access time greatly.

Xbench is showing a 40% to 50% improvement in uncached reads and writes. The disks don't have a ton of cache but the burst rate for each drive is 160mb/s (ultra160).

I've got a cron job that rsync's everything to a backup drive (40gig ATA) every evening just in case. 6 drives total in the case, and no power issues so far. I could not be happier. The hardest part was actually getting the wiring to stay in place.

badtz · Aug 5, 2004, 02:44 AM

I've read that although STR improves in RAID, access times decrease tremendously....

not true? true?

also, is there any way to tell how 'taxing' it is on your system to have OS X do the RAID? [other than that, are there any other benefits/issues with doing software or hardware RAID?]

l008com · Aug 5, 2004, 02:48 AM

I would imagine that, your seek time is only as fast as the slowest time. So if you have one really fast drive and one slow drive, if your using a RAID 0, each drive is going to run as the slowest drive's speed. I try to match my drives as good as I can, both of mine have 8 MB Caches, similar seek times, and are both 7200 RPM.

Tyler McAdams · 06:05 AM

Originally posted by badtz:
I've read that although STR improves in RAID, access times decrease tremendously....

not true? true?

also, is there any way to tell how 'taxing' it is on your system to have OS X do the RAID? [other than that, are there any other benefits/issues with doing software or hardware RAID?]

STR will improve and access times also since your cache grows and the data is faster/easier to access. STR is the big thing you want to increase with raid... it's the biggest factor... A good raid drive maker is Medea... they're known for making raid arrays that have a consistant steady STR... which is big when you're doing any kind of HD work... etc.

The STR difference in a regular situation would be your startup time. Man, the difference between my raid drive and the ATA 40gig backup at boot time makes the ATA drive look ridiculous!

I'm not sure how taxing the software raid is... but I do know that the rule of thumb with these things is hardware is always better and faster than software. I'm sure there is some way to display statistics on the array... but I'm not too sure how... perhaps somebody here does?

I do know that Attotech has a software raid that has more features and probably performs a little better than Apple's raid. But with hardware raid you get features like optional cache memory and battery backup for secure data during a power outage...etc

Tyler McAdams · Aug 5, 2004, 05:45 AM

Originally posted by l008com:
I would imagine that, your seek time is only as fast as the slowest time. So if you have one really fast drive and one slow drive, if your using a RAID 0, each drive is going to run as the slowest drive's speed. I try to match my drives as good as I can, both of mine have 8 MB Caches, similar seek times, and are both 7200 RPM.

Yeap, all of these drives are exacly the same model quantum atlas... all taken from the same server... all from the same seller. If you mix and match drives you can put stress on the array.. once again, so I'm told.