PDA

View Full Version : New stripe speed problems.



kleima
04-03-2003, 09:39 PM
I just installed three new Seagate Cheetah SCSI320 10,000 RPM drives in an array in my Toaster case (I only had three drives slots, hence only three rather than four).
They are testing (in Toaster autoconfigure) at only 37 mb/sec! Now I knew that the 320 drives would be limited by the 160 controller, but this is not an acceptable speed!
The other stripe set (four 18 GB SCSI160's) achieve 88-90 mb/sec.
My controller is the Adaptec 29160. It's true I have a Medea array (external, and not SCSI160), and this could be slowing things down, but the Medea's speed is 55mb/sec. And, I left it off and retested the speed to check this theory and the speed of the new drives is still 37mb/sec!
Any help would be appreciated!!

ScorpioProd
04-03-2003, 11:57 PM
First off, when you boot up, do all three of your new drives show up as a 160 connection in the Adaptec BIOS. It will show a list of the drives during booting.

Second, you did stripe them all together in Windows, right? And do they all show as "healthy" when you do this?

SBowie
04-04-2003, 04:18 AM
Using the wrong kind of cable or terminator can do this.

kleima
04-04-2003, 12:05 PM
Yes, they are all healthy.

I did notice that where there was a "160" beside the each of the 4 18GB drives (when SCSI bios is booting), there was a "40" beside each of the new 73GB drives, so maybe they aren't being recognized as SCSI160.

I do have it terminated at the end with an LVD terminator, however, I am using a cable that has six connectors - one for the controller, four for the drives, and one for the terminator - so since there are only three drives, the first drive connector is open. Could this cause the problem?? This connector shouldn't be terminated, should it??

SBowie
04-04-2003, 01:30 PM
Empty connectors don't matter, but an LVD terminator is not the same as U160. Assuming you are using proper U160 cable (twisted pair ribbon), this is likely your problem.

kleima
04-04-2003, 01:36 PM
I am using twisted pair cable (for SCSI 160), but this is the terminator that came with the cable!
What kind of terminator should I use??
(It actually says LVD/SE on it.)

kleima
04-04-2003, 01:40 PM
Actually I just checked the terminator on the other stripe set (which achieves 88-90mb/sec) and it has the exact same kind of terminator, including LVD/SE written on it.:confused:

SBowie
04-04-2003, 01:45 PM
All of the U160 terminators I've seen state U160 right on them. But wait a second - you're using a 29160? I think one of the internal 68pin connectors on that card is U160, the other is only Ultra Wide! Maybe that's the core of your issue?

kleima
04-04-2003, 02:24 PM
You may be right, I think I remember something to that effect from when I bought it. I'll see if I can rustle up the manual. There are three internal connectors (one 50pin & two 68 pin), but I think it is just 2 channels, so two connectors share one channel? I do have a ribbon cable going from the 50 pin to an external adapter to which the external Medea (which is U wide, not SCSI160) is connected. But, I would think that the new drives would at least make the speed of the Medea (55mb/sec) if that whole channel was only Ultra Wide.

So, if that is the case, I could put all seven internal drives on the one channel, right?? But, would that require a custom cable, or is there a way of chaining two cables?

SBowie
04-04-2003, 02:46 PM
If memory serves, only the 39160 is truly dual-channel. The 29160 is only single-channel, but uses some clever technology to keep legacy protocol devices from slowing down the others.

A single cable shared by the U160 devices would likely be fine. Whether you can daisy chain your cables or not depends on whether the connectors permit that more than anything else. you will still, of course, want the (proper) terminator to occupy the last spot.

ScorpioProd
04-04-2003, 10:42 PM
As Steve said, the 39160 is dual-channel, the 29160 is only single channel. It uses a bus segmenting technology to split the bus into U160 and legacy speeds without compromising the speed of the other segments.

There are THREE internal connectors on the 29160. ONLY the one 68-pin connector on it is U160. The other 68-pin connector is ultra-wide with a limit of 40MBs, and then there's the 50-pin connector with its limits. So that's your problem if you're plugged into the 40MB/s connection.

As for terminators LVD/SE is what U160 ones that I have say on them.

kleima
04-15-2003, 04:23 PM
Hey, thanks for all the pointers!! I got a custom cable done with connectors for all the drives, both the original and the new stripes, and connected them all to the same channel (the 160 channel) on the 29160. When I hooked it up everything worked perfectly, and I tested the speed and the new drives clock out faster than the old (even though limited to 160 and with only three of them vs. four of the other!). They measure at 97mb/sec! (The four 160's hit 89-90).
Thanks again!

dwiener
04-16-2003, 05:57 PM
I'm also putting together a new system with 3 73gb u320 10000rpm cheetah drives. I'm using the onboard controller on a Supermicro X5DA8 motherboard. Autoconfigure gives me a speed rating of 65mbs on this striped set. I'm using a u320 twisted pair cable that supermicro ships with the motherboard. All controller settings are at default. What do I need to change here?

Dennis

kleima
04-16-2003, 10:33 PM
I'm not familiar with the onboard controller on that MB. Is is 32 or 64bit? SCSI 320 or 160?
Based on some of the troubleshooting I did, check to see if your SCSI bios displays the drives as 40 or 160 or 320 etc.
Are there other devices attatched to the chain? Like a tape backup? If there are (and they are U-SCSI instead of SCSI 160), then they can drag the performance down on the whole chain. (Same way with legacy hard drives.)
Make sure that drive indexing is unchecked on all video drives. I have found that while autoconfig says it is disabling drive indexing, that's not necessarily the case. Do it manually. (Right click on the drive in Window Explorer and choose Properties, General Tab and uncheck the "allow drive indexing" down at the bottom.)
Try these things to begin with.

SBowie
04-17-2003, 07:31 AM
Again, bear in mind that the figures Autoconfig reports are reduced from true maximum throughput because it takes the loading of the Toaster into account. I'm not sure that this is any different even though your particualr config does not place the Toaster's own demand on the same bus.

Try running Disktest.exe instead, and I think the figures will be more what you were expecting ... bearing in mind that a 3 x 10k drive array, while adequate, is not going to set the world on fire, Dennis.

dwiener
04-17-2003, 07:51 AM
Where will I find disktest.exe. This is a new install under windows xp. I couldn't find it anywhere on the 3890 disc.

Dennis

SBowie
04-17-2003, 08:40 AM
Originally posted by dwiener
Where will I find disktest.exe. This is a new install under windows xp. I couldn't find it anywhere on the 3890 disc.

Dennis I'll e-mail it to you offlist, Dennis.

kleima
04-17-2003, 09:48 AM
Maybe so, but my three SCSI 320s' (10,000RPM) on a 160 chain are getting 97 and his only 65. So, his speed is too slow even if he has it on a 160 controller or a 32bit bus. If that's the speed he gets on a 64bit bus with a 320 controller, then he's really running slow. (My 97mb/sec was the speed that autoconfig reported.)

dwiener
04-17-2003, 10:18 AM
disktest was giving me speeds of around 54-65mbs on the striped drives. I have decided to try a low level format on all three drives. I have no idea how many hours that will take. Then test all drives formatted individually before striping as a set again.

Dennis

SBowie
04-17-2003, 10:19 AM
While there's substantial variance from one model of drive to another, those findings are more or less consistent with what I've seen with systems running 3 drives. Yours would be the exception, more in the range of what I would expect from 15k drives.

It's possible to capture the output of Disktest and massage it into a spreadsheet, then analyze the throughput over the entire stripeset volume. It would be interesting to run yours through that. It's not uncommon to find regions where there is 25-30mps variance below peaks, not to suggest that Autoconfig (which is really a brief test) would report these or that this would have anything to do with the differences between your system and Dennis'.

SBowie
04-17-2003, 10:21 AM
Originally posted by dwiener
disktest was giving me speeds of around 54-65mbs on the striped drives. I have decided to try a low level format on all three drives. I have no idea how many hours that will take. Then test all drives formatted individually before striping as a set again.

Dennis Don't forget, for a 'clean' Disktest, run it before the Toaster initializes (or kill it). This has a significant bearing on the throughput reported.

dwiener
04-17-2003, 10:22 AM
Again I'm using the onboard u320 controller on the X5da8. Could that be the problem?

Dennis

dwiener
04-17-2003, 10:32 AM
Well I'm in the middle of a low level format on the first drive in the set. Hope it takes a few hours and not days. The Adaptec utility gives no progress indication. :-(.

Dennis

SBowie
04-17-2003, 10:35 AM
Using the onboard U320 controller is the ideal solution, Dennis, so no - that's not your problem. As you might guess from the dialogue above, I'm not of the view that you actually have a problem (assuming you have turned off drive indexing, etc.) What you're seeing is about what I would expect.

In my current system, I've got 4 older U160's on the same controller on the same motherboard. My throughput speeds top out around 109 ... when the stripeset is empty. However over the entire surface, it would be more in the neighborhood of an average of 87-90, with regions that drop into the high 70's. As I said, this particular stripeset is older, and I have seen higher numbers with more recent edition drives -- I get a fair number of systems passing through here -- larger drive cache has a lot to do with it ... but I don't find anything unusual in your numbers.

dwiener
04-17-2003, 10:43 AM
these are brand new u320 10k cheetahs, with 8mb cache. I'll try the disktest again after this low level format completes. argh.

dwiener
04-17-2003, 12:55 PM
Just tried a single drive and was still seeing only 64mbs throughput. I'll strip the remaining 2 and see what I get from those.

dwiener
04-17-2003, 02:59 PM
Well I restriped just 2 pair and got over 120mbs with disktest. Went and did all 3 and am now getting 195. Autoconfigure gave a reading of 178. Don't know what was went awry the first time I striped the set but all seems fine now. When runnig disktest I see the occaisional hiccup down to 40-50mbs. Is this normal? Do you think it worthwhile to run the diskcontroller error check utility on the drives and low level format them? Would this gain anything for my system?

kleima
04-17-2003, 03:19 PM
Now that sounds more like it!! With the Cheetahs, and the 320 controller, you should definitely be getting more that 65mb!

The drives I have been speaking of in the thread are the same as yours - 320 Cheetahs with 8 mb cache @ 10,000RPM.

It is normal for drive performance to vary, although the more drives striped together, the less dramatic should be the peaks and valleys.

SBowie
04-18-2003, 08:51 AM
Well eat my shorts, I'm quite surprised. I'd be interested to see the results of a complete Diskest analysis if you took time to do one and save the output, Dennis. I would not have beleived you could get that kind of performance from a 3 drive setup.

Paul Lara
04-18-2003, 09:48 AM
Originally posted by SBowie
I would not have beleived you could get that kind of performance from a 3 drive setup.

You certainly can, if they're Ultra 320's!!

SBowie
04-18-2003, 11:17 AM
Not to quibble with you, Paul, but I don't actually think the throughput rate of a given U320 drive is inherently any faster than the rate of the U160 version of the same drive ... at least this was never the case before.

To put it another way, the higher rate of the controller normally means that the overall ceiling for the bus has been raised, but one would not expect to get significantly different performance from an individual drive.

kleima
04-18-2003, 05:27 PM
The maximum burst data transfer rate for a SCSI 160 is 160mb/sec. The maximum burst data transfer rate for a SCSI 320 is 320mb/sec. So, yes 320's are faster than 160's.
I don't know the sustained rates off the top of my head for each drive, but I would expect them to have a similar difference.

dwiener
04-18-2003, 05:46 PM
Steve

I finally got rid of the brainfarts and remembered how to redirect the output of disktest to a file. It should be finished as we speak, but its on a machine at home and I'm at work now. Unfortunately I forgot to shutdown winrtme before I started it so I'll have to do it again when I get home later tonight. I'll email you both text files when I have them done.

Also refresh my memory. Is there a command line syntax for doing a shorter disktest? I seem to recall such a thing but can't remember what it was. There doesn't seem to be any built in help to the file as disktest /? just starts the test running.

Dennis

SBowie
04-18-2003, 06:04 PM
Originally posted by kleima

I don't know the sustained rates off the top of my head for each drive, but I would expect them to have a similar difference. I wouldn't, but then I've been wrong once today already.:eek:

In the past, protocol changes of this sort have had more impact on the throughput ceiling of a series of drives than they have on the individual speed of a single drive. This was discussed at length back when we all shifted to U160, and proved to be more or less true. There's little doubt in my mind that the 8meg cache has more bearing on the speed of a single drive than does the change from U160 to U320 with a similar drive.

dwiener
04-18-2003, 09:36 PM
Steve

I came to the VT from the original Framegrabber. I bought that to be able to do compositing easier with my old system, a FAST Video Machine with DPR. The FAST system had its own SCSI-2 controller built into its capture card and had 2 single and 1 wide scsi bus. Striped sets (a pair of drives on each of the single buses) when the would yield 8-9mbs, fast enough to capture at 5:1 compression and playback 2 streams of video real time. As drives got faster 3:1 compression was not a problem. Currently people who upgraded to the original LVD drives were able to get 15mbs on a single drive on the wide bus, fast enough to handle 2 streams of playback (audio and video), all on a scsi-2 controller. Don't know if its the electronics, or the cache, but they're fast enough to show improvements even with old technology and keep that investment relavent.

Dennis

SBowie
04-19-2003, 12:57 PM
Sure... I'm not arguing that faster drives aren't coming down the pipe as time moves on. Just that the controller basically sets the upper limit. No matter how fast the drives get to be, no combination will exceed the controllers top spec. In the examples you mention, earlier drives replaced with faster ones are still below the ceiling set by the controller..

Throughput is basically a question of "where are the bottlenecks." 'Raising the ceiling' by moving to a new protocol allows faster drives, RAIDs, etc., to approach the ceiling set by the controller. But it won't do much to speed up an existing drive design taken individually.

When new protocols appear, mfrs. often adapt existing drive technology to it, but of course doing so doesn't mean the drive speed for a given updated model will double. Modifying disk density, cache, rotational speed, and the like WILL have a significant bearing. So the "10K.6" series Seagates with the 8 meg cache are definitely faster than their predeccesors with the 4 meg cache. Clearly, I hadn't realized how much so!

Last summer, before U320 controllers were generally available, Seagate began offering these U320 8 meg cache Cheetahs. Since a single one of these drives would never hit the ceiling imposed by the U160 interface, in that setup (single U320 drive on a U160 controller), the drive would not hit a bottleneck, even though connected to a U160 controller.

In fact, one drive would in all likelihood provide identical performance on a U160 or U320 controller. Neither would 2 of such drives hit the ceiling, but presumably 3 or 4 might have. A lesser drive, such as the 4 meg cache 'ES' model, might not hit it until one racked up a half dozen in a stripeset.

Let's take another specific example: the same Quantum/Mator Atlas series of 10k drives are available in both U160 and U320 models. The performance of the drives, taken individually, is for all intents and purposes identical -- but racking up a stripeset of them, the user will obviously hit the ceiling sooner in a system using a 160 controller with 160 drives than with the 320's on a 320.


Anyway, I'm pleased as punch that you're getting such blistering speeds. Looks like you hit the 'sweet spot' with that drive, and it makes me want to hook up some U320 15k Cheetahs's to my system to see what that might do!

stuart_tetley
04-19-2003, 09:35 PM
Steve,

I'd be interested in getting hold of disktest.exe as well.
I have a 4x U160 stripeset which autoconfigure sometimes reports as delivering 133Mb/sec, and other times as low as 82.
I've also been having problems with drives fail on me - 2 of the original 4 have died within the first 30 days.

Any other advice welcomed. The drives are 4 Quantum/Maxtor Atlas 10K IIs, attached to an Adaptec 39160 (2 drives on each channel.) Are the 10K II's known to be fragile.

Thanks


Stuart

SBowie
04-20-2003, 06:20 AM
E-mail me offlist, Stuart, so I can send it as an attachment. I haven't used the Atlas drives much in recent years, but they used to be fairly reliable. I never liked Maxtors much, but I think the Atlas series continues to be pretty much as it was when Quantumn made them.