PDA

View Full Version : Comments- Drive and RAID Performance Database



ricks
05-16-2002, 02:56 AM
This forum topic for discussion on the Drive and RAID Performance Database.

ricks
05-16-2002, 03:19 AM
Hi all,

We're finally getting closer to a usable DB. I about have a list of what to send in to get your results posted. I think I'll continue with the benchtest screenshot requirement, it seems to alleviate the common errors, especially typos, that plague a database like this.

I know it's a pain to send them but it gives me the ability to check my typing after entry as well as checking the senders figures.

My one format question at this point is how to keep the headers in sight as the DB gets bigger and more pages in depth. Maybe have the data scroll but not the headers? That's probably the next step.

Please feel free to email me and we'll discuss your benchtests and what I need with them. I have a list, just not in a form that should be published here yet. Single drive performance as well as array performance will be represented in the database. Feel free to send your drive performance even if you have no RAID component. Different drives, computers, OS versions, drivers, host cards, all this should be part of the DB.

Gregory, your success with installing your host card in the B&W 66mhz graphics slot needs to be here also.

Thanks to everyone for all your previous suggestions, it has helped considerably.

Rick

[This message has been edited by ricks (edited 16 May 2002).]

NorbyWells
05-17-2002, 01:37 AM
I am thoroughly intrigued by the data displayed in the 15k Cheetah vs. 10K Atlas III. The Atlas' sustained read numbers, even when used as a single drive with the ATTO UL3D, are staggering.

Here is why I am so interested: I am setting up a new Dual Gig to be my primary machine. I have already purchased the ATTO Express PCI UL3D based on my extensive readings of the Gurus web site. I was contemplating purchasing two G2 15k Seagate Cheetahs (18 Gb each) and stripe them internally using one channel of the UL3D. This striped array would be primarily used for Photoshop work and future video dabbling. I am also interested in a fast boot/application drive and was going to place a G2 15k Seagate Cheetah (36 Gb) on the UL3D's second channel and install the drive above the the original 80 Gb ATA drive. The 36 Gb would become the primary boot/app and document disk and the ATA would be the secondary doot drive and used for daily back up. I figure that I could sync the entire contents of the 36Gb to the 80 ATA on a regular, if not daily, basis.

Given the information in the Cheetah vs. Atlas chart I am considering purchasing the a 36Gb Atlas III in place of the 36 Gb Cheetah due to the Peak and Sustained read numbers which seem like they would help boost boot and application drive activity rates.

Any thoughts? Am I all wrong on this? I have, in years past, experimented with striped arrays and narrow drives using FWB (sorry for swearing) but have never attempted anything at this level of performance. Given the drive installation mentioned above, should I be concerned about any heat issues with the 3 SCSI drives and 1 ATA drive configuration..... any additional cooling needed?

Other info that may impact your thoughts: Dual Gig G4, 1.5 Gb Ram. I am making an effort to use OS X almost exclusively which rules out SoftRAID until they release their OS X version. I figured that I would stripe the two 18 Gb Cheetahs using Apple's Drive Utility for the time being and then reformat/restripe when SoftRAID X is released.

Any and all insight would be greaty appreciated before I throw down the plastic for these drives. If I am making any grave mistakes would someone please slap me back into line.


Norby

norbywells@mac.com

P.S. If and when I do get this all configured, I would be more than happy to provide data per Guru specs for the RAID/Drive database.

TZ
05-17-2002, 08:25 AM
Norby,

Stacking the Atlas should be safe, but i wouldn't put X15 above the ATA drive. Rather than X15 you could use a 10K Cheetah and get nearly the same level of performance (but less heat and noise, and see 75MB/s sustained reads and 50+MB/s sustained writes).

You could put the ATA drive in a FireWire case, it would be bootable and hot swap if you want to keep your Cheetahs etc inside. And stay with all X15's. Looking back, having all drives of the same make etc makes it easy if you want to modify or add to your RAID setup.

I tried Atlas 10k III, 10K Cheetah, and a pair of X15's. The Atlas died after couple months. The 10k is very good and solid, 50MB/s SW, 60MB/s SR (I didn't have a pair to raid, that is for a single drive). Single X15 came in at 58-60MB/s SW, 115MB/s striping two. Just another drive option to look at.

Gregory

ricks
05-18-2002, 05:19 PM
Norby,

I've been pondering your questions on Atlas vs Cheetah. That's an interesting way of looking at it. I didn't get Cheetahs for cost reasons at the time, I couldn't afford the 15k Seagates and the new 10k Cheetah wasn't around yet, the AtlasIII's seemed a good compromise and I wanted a chance to test them out.

Reads are Very impressive on the AtlasIII's. I would love to see them on the new Xserver, 4 of them might just come close to maxing the 533MB/sec bus in the read catagory, but writes would still be around 200, big difference.

I have no idea what the long term reliability is with the AtlasIII's either. I haven't heard much bad about them and they've been around longer than the current 15k Cheetahs. That said, trying to say the Cheetah's are better long term is stretching things since neither has been around that long. However, historically, Seagate is THE best at reliability.

I think you takes your pick. That's why we put together a database, from looking at numbers all day it all became just a mass of junk in my semi useless brain, the graphs really made it easy to see.

I like the AtlasIII's for one other reason, they are the coolest drive I have ever seen. And I include all the IDE models as well. AtlasIII's do not get hot. Amazing. I ran two of them stacked for a while as I was testing in different computers, never noticed any significant heat from them stacked even. That isn't a recommendation to do that long term though. I ended up moving all 4 to a Burly external just to clean things up and leave more room for ATA drives.

Rick

Damien
05-18-2002, 07:37 PM
Perhaps you have a boo boo...

you list Kaye as getting 88MB a sec using the apple driver (2x120GXP ATA) while Softraid is only doing half that? Is that right?

In my recent install of 2 10K Cheetahs into my 9600 it was my experience that the apple driver completely blows chunks and in most of the other listings in your database this seems to be proven out. But Kaye gets the Apple driver to work well? Maybe those drives are just afraid of what Kaye will do to them if they don't perform.......

kaye
05-18-2002, 08:18 PM
Damien,

The numbers are correct but let me explain why. First, getting 88MB a sec using the apple driver (2x120GXP ATA) is on the Acard 6880M in HardRAID mode (HR0) and the Apple Driver is recommended by Acard. That config is now my boot drive for both OS9.2.2 and OS10.1.4.

While Softraid is only doing half that? Is that right? That is what the ATTO test says, but. In fact, the ATTO test is confused by the fact that the sustained reads graph goes up to about 94MB/s and stays there for nearly the whole test, about 6MB of the 8MB test. So the ATTO test sees that whole long 94MB/s as Peak Read. Never seen a graph like that. Note the MB5 numbers for both tests. The SoftRAID config actually performs equally to the HardRAID config. I could have used SoftRAID for the HardRAID config instead of the Apple driver, but then no OSX bootability. I also keep a pliars close by just to keep everyone in line.

Once again, you are the first to notice that. k

Damien
05-18-2002, 08:33 PM
So Softraid was so fast that it confused the atto benchmark util..... LOL!

I noticed this now because I just had experience with just how bad the Apple driver really is :-) Stable, compatible and SLOW!

Both Softraid and HDST were more than twice as fast as Drive Setup (10K cheetah, OS 9, 9600, 2940U2W card)

NorbyWells
05-19-2002, 10:01 PM
Thanks to ricks and Gregory for sharing their thoughts on my impending drive purchase. Due to the heat issue it appears that both of you agree that the Atlas is the cooler of the two drives and would be "safe" to stack above the original ATA 80 Gb drive that came with the machine.

Gregory, when you state that the Atlas died after a few months was it replaced under warranty?

It's just that I am so dang impressed with the Atlas sustained read numbers that I am thinking it would make an excellent boot drive.

Norby

norbywells@mac.com

TZ
05-20-2002, 09:41 AM
Norby,

The Atlas is still working. FWB reports an error but maybe it's being conservative and it was a recoverable error. So until I hear from FWB the drive is being used for alternate backup. And it gets 147MB/s peack read, 117Mb/s SUSTAINED read. Writes are 52-51MB/s for peak/sustained. Not shabby.

HOWEVER, the new Seagate 10k Cheetah is also quiet, cool, and 50MB/s drive. Both sell for about the same, too. Less than $200.

Can't go wrong either way.

I spent the last 36 hrs running tests, LLF, verify media and drive. I'm just glad that I've collected as many disk utilities as I have. Each one has been useful and paid for itself more than once.

Watch out if you ever do need to or choose to use Intech HDST as it defaults to options you don't want. Being overly conservative and cautious. Especially disabling wide 16-bit mode AND did not enable the option to disconnect/reconnect to allow for chained I/O's. (is that what I want to say?)

I'd jump over to Seagate and Maxtor and Storage Review to review the heat, noise and other features. The Atlas - to me - does have a high pitched noise that often people can't - or don't - hear, I don't hear all the time (mostly in the AM when everything is quiet and things sound louder after sleep).

Gregory

magician
05-31-2002, 01:45 PM
gents.

i am staggered by the beauty of what is transpiring here.

http://macgurus.com/infopop/emoticons/icon_smile.gif


http://www.macgurus.com/ubb/magician2.gif

kaye
03-31-2003, 02:11 PM
bump

speters1
04-03-2003, 03:25 PM
When it comes to photoshop scratch, which speed test numbers are most important? I am also wondering if 4 Atlas 10kIII's on 1 channel would be faster than 2 drives on 2 channels?

ricks
04-03-2003, 03:35 PM
?In my testing I was unable to test all four AtlasIIIs on a single channel because I didn't have the cables. But in testing two drives there was significant advantage to having them split between two channels.

?With the newer UL4D there might be no advantage to multiple channels since each channel is capable of saturating the PCI bus. Someday I might have the moola to test that but at this point there is no purpose in me upgrading to the UL4 from the UL3D since the UL3D is faster than my computer can use anyway.

Rick

speters1
04-03-2003, 06:30 PM
So waht numbers are most important for photoshop scratch,sust read, sust write, peake read, peak write?

TZ
04-03-2003, 08:23 PM
All of it. Storage Review builds a file pattern for testing various real world uses, workstation, file server, web server, they are all different. Capturing, reading, random reads, sequential reads. ATTO is basically not testing very much. Its what everyone uses, but it is nearly "junk."

On my B&W, with one Atlas 10K-4, the sustained reads were 110MB/s, the peak read was 145MB/s which held for 6MB of the 8MB file test segment. Which means that it was testing the drive's built in cache which allocates 6MB of the 8MB for reads and pre-fetches. You would need Intech or another more complex program and test to learn anything real.

However, I do buy various drives, 15K Cheetahs, Z15 IBMs, 10K Cheetahs (various models), and a number of Atlas 10k II, III, and IV. The Atlas drives are my preferred, and the 10K IV would be great to RAID, but they are good just as a single drive - affordable doesn't hurt.

You use to need to "short stroke" a drive and create your fastest stripped raid on the outer third of the drive, that use 4, 6, 8 drives and thereby get 220MB/s sustained writes. The max the Mac could deliver.

Most drivers have some default custom configurations for video, audio, server, even for photoshop sometimes, by tweaking the cache segments and other mode pages settings. SoftRAID, FWB HDT, ATTO ETP, they all have options. FWB HDT use to have an option to create a photoshop scratch volume - you can use different size memory pool, segment size, etc and then people will test until they get the best overall performance for their use.

Learning a little, then playing with various drives, drivers, and even controllers. Some people have used dual Miles2 cards to create a dual channel setup in 9600, using one upper and lower bus master slot.

Ultra320 I've found already to offer superior performance to Ultra160. To get the most, though, I'd need to sell my UL3S and get a UL4S.

The Atlas drives tend to show higher sustained reads than Cheetahs. They've been building 10K drives with 8MB cache longer, too I'm pretty sure.

ricks
04-03-2003, 11:03 PM
?I do not know how to quantify Photoshop acceleration from a RAID array scratch disk. And believe me, I have tried!

?I built the four drive AtlasIII array for exactly that reason 9Photoshop scratch space) and used ever test utility I could lay my hands on to make similar and repeatable benchtests on various computers with other testers. That's why we ended up using and relying so much on the ATTO test, it both made sense in that the displayed data fit our model. And it was repeatable in other computer models and with various arrays combinations.

?The data is severely limited, I'll be the first to admit. ATTO tests only up to 8 MB file sizes. That means a large SCSI array doesn't even NEED to activate the heads to platters, ATTO tests only caching.

?But cache is part of the overall performance of a drive/bus/array.

?In the end all I can say is that my four drive, two channel array makes Photoshop sit up and say "how fast did you want?" Progress bars FLY http://www.macgurus.com/ubb/kickass.gif Performance is WAY above what I get with just the Acard RAID designated scratch. And SCSI s still king for reliability.

?I would probably today NOT build a four drive external. I would get the very fastest two drives from Seagate and hook them via a UL4D on one channel internally. Lower cost and plenty of performance, maybe even close to my four drive. And this internal setup would cost a bunch less than half what my four drive external did in 2002.

?The Dual UL4D would allow me the luxury of expansion externally if I ever need that in the future. That's the only reason I would choose the UL4D over the UL4S.

Rick

speters1
04-04-2003, 12:58 AM
Here is what just got. ATTO ul3d and 2 Atlas10kIII 160, and 2, 4 connector internal cables and terminators all for $400. Thsi is goung to be used for photoahop scratch. I was considering getting 2 more drives and and getting an external case from hypermicro. Getting 2 more Atlas's and an external case is going to be cheaper than buying 2 new 15k Cheetahs. I can't put them inside because I already have 3 drives in there. So, in my case, do you think I am doing the right thing (going with the 4 drives in external case)? Onelast thing, can you put 4 drives in one case and hook up 4 drives to 2 channels? Do the make a case that has 2 connections outside?

TZ
04-04-2003, 10:53 AM
Does PS even benefit from 15K drives that much? I didn't think it created random disk patterns that would. One thing you probably want to do, and some disk drivers will, is to use hfs+ with something other than the default 4k block size, something like 64, 128 or 256 (256 is only available when using SoftRAID). That is what FWB HDT PS scratch volume was doing in part.

You external RAID box with cable and terminator, $149 cable, $79 terminator, then the cost of box ($110+), plus cost of two drives.

Atlas 10K-3 drives: expect to get about 110MB/s per channel for two drives. You can put three on one channel, but you can't get more than 220MB/s no matter what you use, though it will feel fast... !

The Atlas and Cheetah 10K drives are capable of 80% saturation of a channel during reads, and delivers 70MB/s... an X15.3 only gets 75-77MB/s (and that was in 73GB drive test, 18 or 36GB drives have fewer platters and read heads, and you pay TWICE as much for 15K drive.

* Four drive RAID:
Extra case, cable, and terminator for external channel $325+.
Two 10K-3 Atlas drives $109 ea.

If you are ready to spend $530 just for two more drives and equipment, you could buy two 36GB Cheetahs ($179 ea) that deliver 140MB/s and 73GB of storage. If you need the extra 70MB/s, three drives... that extra little bit costs a LOT.

- G.

I've spent a fortune a couple times over learning to set up my first, second, and third RAID boxes. I wish I knew years ago what I learned by doing, and of course, drives just keep getting better. I knew it was important to do it right the first time, but I didn't know what that meant. And I couldn't afford to do it right in one fell swoop.

[This message has been edited by Gregory (edited 04 April 2003).]

speters1
04-04-2003, 03:12 PM
Hypermicro has external cables that are $27, and terminators that are $20. Are these ok?

TZ
04-04-2003, 04:08 PM
I throw those things into my junk box. I've got 'em 'cause I didn't know any better. Believe me, you do not want them.

I had one terminator that forced everything to 40MB/s. Suppose to be active LVD/SE terminator. I've seen stuff listed that only supports 40MB - at least those were honest.

I've tried rounded "Ultra320" cables from HMS that looked solid - and worked once under Jag, but not under OS 9, and ended with having to reinitialize (thought I'd lost the drive, damaged the boot blocks, something serious).

We swear by Granite Digital around here. And to get the best, most reliable, they just plain work. Initio has some Ultra2 cables that use to come (still do?) with the Miles2 kit, but they will tell you that they are okay but not the best.

ATTO does sell LVD terminator, but it costs more than the 2015 from Granite www.scsipro.com (http://www.scsipro.com) - ATTO data (external) cables are okay and test well, so you can save a few dollars there at www.attotech.com (http://www.attotech.com) online store. free shipping.

SCSI Pages of DOOM! (http://www.macgurus.com/products/drives/mgscsiadapters.php) has the lost archieved pages on cables etc.

If you were using Ultra2 or older technology, maybe you could get away with the el cheapo stuff.

That's part of why the first SCSI drive costs so very much, even w/o the cost of the drive (which have gotten rather affordable, the new Atlas 10K IV was half the price of the Ultra320 10K III) the "entry fee" is steep.

The single and dual drive cases from HyperMicro are good and well-built and worth the $75 or $110, I use quite of few of each.

The very first active terminators for SCSI from APS were $99, then gradually less, but they helped, they worked, were for narrow (but complex, or longer cable length) SCSI systems (multiple drives, multiple scanners, zip, etc).

I thought I was crazy, and very very lost, until I found this site.

I have some extra Granite terminators, a box of junk cables, and some single drive cases I was trying to sell, including the 39160 (works in 9.2 only), SoftRAID, even the cable that came with the 39160 isn't up to snuff with these Cheetahs etc I use. So make me an offer, I've got a thread on the For Sale forum looking to sell "SCSI Stuff." I'll even throw in my junk box for $1!

Gregory

TZ
04-04-2003, 04:29 PM
The external cable may be okay. Yes, I bought a pair of those, too. Ended up getting Granite's along with ATTO's. I think the reason I did was to save money, but ended up having 'sporadic' problems with a mirror RAID.

I would look at ATTO LVD TRM (http://store.yahoo.com/atto/ter68pinplvd.html) $97, and the VHDCI (http://store.yahoo.com/atto/cabverhigden.html) $95. If it uses the wider 68-pin ext. cable 68-pin cable (http://store.yahoo.com/atto/cab68pinfinp1.html) $65.

I just bought a cable tester ($49) from Granite to help sort throw when it is a bad cable, and when it is something else.

Granite SCSI Products (http://www.granitedigital.com/catalog/index.htm#scsiproducts)

- G.

I'm going to go do some testing with some cables this weekend and see what I come up with.

[This message has been edited by Gregory (edited 04 April 2003).]

ricks
04-04-2003, 05:36 PM
Steve,

?I don't see any problem at all with your UL3D and two channels in a single enclosure. The wiring can be kinda fun. Here's a picture of my backplane where you can see each external data cable coming in, each hooks internally to two drives via a ribbon cable with a panel mount connector on each end. One end for data in and the other either to connect to more drives or to hook up a terminator which is visible in the picture.

http://homepage.mac.com/rstephens/.Public/Backplane.jpg

?Here's the same case showing drives in drive coolers and the cables coming from the backplane to the drives and back to the backplane.

http://homepage.mac.com/rstephens/.Public/Internal.jpg

?Hope that helps you to see what you'll need in hardware.

Rick

[This message has been edited by ricks (edited 04 April 2003).]

speters1
04-04-2003, 08:06 PM
I have just one more question. As I stated before, I already have 2 Atlas 10kIII. I want to get to more to use for my stripped array. I will be using express raid. When i go to buy the other two drives I know I need to get the same maodel, size, ect. My drive is KW18L011 REV 03-D. Do I have to mach that exactly? I see googlegear has a KW018L2,KW18J2, KU018L2. Will any of these work with my other 2 drives? What is the difference between these?

ricks
04-04-2003, 08:58 PM
I personally have no idea what effect if any mixing those drives will have on your performance. Probably none, but unless someone has specifically done exactly what you're asking there is just no telling.

Rick

TZ
04-05-2003, 08:57 AM
I think you are talking about the $97 models? they should be identical, or identical enough. But use Apple RAID (free) or SoftRAID 3 when it comes out. EspressStripe is junk, crap, hard to use, and somewhat trouble.

I just ran ATTO on my new Atlas 10K4 and even in my begie on UL2D, that one drive gets sustained write and reads of 65MB/s, peak read 90MB/s (can't see, the machine is busy reinstalling Jaguar). ie, better than I got on B&W with UL3S. And much better than my first Atlas 10K II (or my current 10K III, or my X15.2) RAID setups.

It helped when you could just buy it all at once from MacGurus online store and have all the drives, cables, and know it would work. Use to be if you wanted solid SCSI RAID, you bought Cheetahs, but I for one, think Atlas drives deserve more recognition.

- G.

TZ
04-07-2003, 01:05 PM
Benchmarking (http://www.lostcircuits.com/advice/satacuda/10.shtml) from Lost Circuits is essential to understanding this topic -

speters1
04-09-2003, 10:23 PM
I want to have a 4 drive raid (4 atlas 10kIII 160), ATTO ul3d. Two drives set up interanlly on channel 1 and two drives externally on channel 2. If I add a cheetah 15k 320 internally on channel 1, will it not run as fast as it should, because of the other 4 drives? Should I save some money and get a 15k 160?

TZ
03-28-2005, 07:00 PM
I want to have a 4 drive raid (4 atlas 10kIII 160), ATTO ul3d. Two drives set up interanlly on channel 1 and two drives externally on channel 2. If I add a cheetah 15k 320 internally on channel 1, will it not run as fast as it should, because of the other 4 drives? Should I save some money and get a 15k 160? I don't think you understand bandwidth or the real difference between specifications and performance. A Ultra160 controller supports about 140MB/sec on a channel. So two 10K IIIs "eat up" 110MB/sec and you can never get more than about 140-144MB/sec.

Ultra320 specification would benefit slightly from UL4D (on G5) but otherwise you need another controller. You would be better with all 15K drives.

also, I would / should have advised against using 10K IIIs internally on any system. they just add too much heat.