PDA

View Full Version : G4/Lv 0 Adaptec Array is Killing Me



42
07-31-2001, 07:30 PM
I posted more than six months ago with problems about my 6-disk Cheetah array. It's been working okay for the past 2 years, but a couple of times now, it has cropped up with problems of one sort or another. The last set of troubles (6 months ago) mostly went away with the purchase of a DG diagnostic terminator. This *after* the Adaptec rep suggested I terminate two or three of the drives in the array (which I tried without apparent impact). Anyway, the DG guy also said my cabling and enclosure sounded solid to him. And things did get better, although we were still getting a crash every day or two (from which we easily recovered with little or no file damage; we also tried slowing the Adaptec down to see if the crashes would go away, and they never did). But lately, things have been getting worse again. For no apparent reason, the array's been hanging during the nightly backup, freezing at least once or twice a day, and has now started giving me those SoftRaid bad media messages. (Yep, I ran all the normal utilities, did restarts, shutdowns, rebuilds, and so on.) But the last straw was when SR reported a missing drive, even though the message went away following a restart... yechh.

I ain't tryin' to squeeze more speed out of this thing. When it's working, it's plenty fast. What I want, crave, NEED, must get... is *reliability.*

So. What I did *this time* was reformat (SoftRaid222) all six and recreate the SoftRaid volume. Then I decided to take a clue from their pdf guide and decided to play with that "maximum current I/O" setting. It seemed logical to take that to the highest setting from what I read (on page 19). Then I moved all the data back onto the array and tried to clean it up with Norton before putting it back online. But Speed Disk crashed when I tried to optimize, with or w/o media check. Found that Speed Disk worked okay when I pulled that MaxCrnt I/O setting down to 512, but backups still ran slow, much slower than before. So now I have reinstalled the driver again with the default setting, but I'm just wondering what to do next...

I'm thinking maybe it's time to dump the Adaptec. Your thoughts would be welcome. I hear people raving about the ATTO UL3D and the Miles2 (though mostly in the context of vintage Macs) and not that many raves about Adaptec.

Anyway, here are some questions (assuming the basic hardware config is okay--see below):

1. What should the "maximum current i/o" be set at for my config? Or is it not nice to play w/ mother nature...

2. What is the best host controller for my drive/enclosure/Mac config?

3. Do I have too many drives in this array? Is there something else I'm missing?

Here's the hardware rundown:

G4/AGP/450/256/20 (the Apple part number is M7825LL/B)
Note: ASP reports under 'attributes' heading: "Low power mode disabled because of PCI cards in slots: SLOT-D (I have no earthly idea what meaneth this but thought I'd include it in case someone else does)

Kingston Data Silo DS-400

Six Seagate Cheetah 18LP drives

SoftRaid 2.2.2

OS 9.1

ASIP 6.3

Thanks in advance for your thoughts.


[This message has been edited by 42 (edited 31 July 2001).]

kaye
08-01-2001, 12:31 PM
42,

I vaguely remember your problems before. All six Cheetahs are the same model exactly? What model? No adapters installed? I would consider Granite cabling to your Silo and inside it. That said, if you are going to purchase an ATTO UL3D, which IMO would be the best for your G4 (the Miles2 is best with vintage Macs, not yours) the ATTO external connectors are a different animal, .8mm Champ VHD (Very High Density) connectors in order to physically fit the backplane. If your Adaptec is dual channel, I suspect it is not, it may have the same type of connector.

These cables are expensive as are the internal cables. And with a UL3D you will want to split those six drives, three per channel, and therfore two external Granite cables and two internal Granite cables for the separate channels. So you need to decide which card, existing Adaptec or UL3D, before getting any cables.

Six drives on one bus may be too much, I don't know. But I'm thinking it would be OK with the very best cabling. But then your Adaptec may be bad. I know magician has run eight X15 drives off of a UL3D, four drives per channel though, not six on a single bus. He did that with a Sawtooth and the drives in an 8-bay Burly enclosure.

I'm going to post this and leave the Maximum Concurrent IOs for another post. Besides that one did I miss any other questions? k

kaye
08-02-2001, 11:01 AM
42,

Your note "Note: ASP reports under 'attributes' heading: "Low power mode disabled because of PCI cards in slots: SLOT-D (I have no earthly idea what meaneth this but thought I'd include it in case someone else does)" I'm going to guess is an Energy Saver warning that whatever is in Slot D (I don't know how your slots are listed) is preventing low power mode. You will have to look in ASP to see what is in Slot D. Magician does not recommend allowing Energy Saver to put an array to sleep anyway.

Your question about Maximum Concurrent IOs, I did a lot of testing with the SoftRAID Drivers Parameters window in various configurations on my PowerTower Pro and S900 and got some valuable info from Mark James of SoftRAID, as I recall. What I tested (BTW, these tests took a long time and they were done on a vintage Mac with slower bus speed unlike your machine):

3x 10k Cheetahs ST39204LW striped, 1x Miles2 in the S900, SoftRAID 2.2.1 and 2.2.2. Best results as measured by ExpressPro Tools 8MB and no Disk Cache, rest default, and by MacBench 5, Max Conc IOs 256, Memory Pool Size 991. Write Acceleration ON and Read Acceleration ON.

What was puzzling is that the SoftRAID printable manual (maybe they have a newer one now) showed default Memory Pool Size of 552 but the default in the window showed 804. Also, after increasing the Max Conc IOs above 256 to 384 or 512 and clicking OK, on reopening the window at another time, the slider would be back at 256. Mark James, if I recall this correctly, said that 2.2.1 and 2.2.2 would only allow max 256, so you can set the slider to anything above 256 but it goes back to 256. The Memory Pool Size does stick and the amount you set does make a difference. I still set both to what I want regardless.

2x X15s striped, 2x Miles2 (1x X15 on each Miles2) in the PTP, SoftRAID 2.2.1 and 2.2.2. Best results as measured by ExpressPro Tools 8MB and no Disk Cache, rest default, and by MacBench 5, Max Conc IOs 256, Memory Pool Size 991. Write Acceleration ON and Read Acceleration ON.

4x X15s striped, 2x Miles2 (2x X15s on each Miles2) in the PTP, SoftRAID 2.2.1 and 2.2.2. Best results as measured by ExpressPro Tools 8MB and no Disk Cache, rest default, and by MacBench 5, Max Conc IOs 768, Memory Pool Size 1369. Write Acceleration ON and Read Acceleration ON.

I now have reorganized my X15s into an 8-bay Burly enclosure with six X15s in it. I have organized this burly, with magicians help, so that I can run dual bus (2x Miles2, 3x X15s per Miles2) or dual channel (1x UL3D, 3x X15s per channel) with my PTP or single bus (1x Miles2, 6x X15s) with either the PTP or S900. Six X15s floods any of these configs. They work but I need a late Apple G4 machine, with faster bus and faster memory, with the UL3D in dual channel to fully explore what the Burly can do and what to set for Max Conc IOs and Memory Pool Size. k

42
08-06-2001, 12:42 PM
Hmmm... Kaye, your two posts make me wonder if I should dump the G4 I have in favor of a later one, as well as get rid of the 2940U2W in favor of an ATTO-powered dual-channel setup. Your info about the max concurrent I/Os is also highly appreciated. Next question... can you give me any idea what I will need to spend to convert to the ATTO UL3D (new cables, diagnostic terminators, etc.) in a dual-channel config? And would these *have* to be implemented on a newer G4, or could I use the current beast if I run out of money? Thanks again for your wonderful insight and help...

kaye
08-07-2001, 10:11 AM
42,

What are the model numbers of the drives (STxxxxxLW or whatever)? I don't think you need a new G4. But before I detail out what you may need, gotta know which Cheetahs and if they are all the same. k

42
08-07-2001, 12:20 PM
Yes, the drives are all identical; six Seagate Cheetah model # ST318203LW, firmware 0001. The rest of the hw/sw config, again, is:

G4/AGP/450/256/20 (the Apple part number is M7825LL/B)
Kingston Data Silo DS-400
SoftRaid 2.2.2
OS 9.1
ASIP 6.3

Thank you once more... for your thoughts on this mess o' mine.

kaye
08-08-2001, 12:12 AM
42,

Somehow I missed your last post but I have been trying to juggle several house duties, a plumbing leak outside, transferring about 17GB of data from my old main machine to my new main machine, and setting it up with my old printer and scanner.

Anyway, I am worried about the possibility that one of those Cheetahs might be intermittently going bad. But, for now, I think the UL3D would be the high end answer. It would also allow splitting the drives into three per channel. This is an expensive solution as you will see. If you think of an alternative, say it. Your Kingston looks great and top of the line. I checked, looks very high quality.

1. ATTO ExpressPCI Dual-Channel Ultra3 SCSI Host Bus Adapter $459.99 ea http://www.macgurus.com/shoppingcart/showrampage.cgi?attoboards.html

2. 2x GD4911 Ext. 68-pin .8mm to 68-pin MicroD 3-ft. Cable $159.99 ea, $319.98, http://www.macgurus.com/shoppingcart/obj_show_page.cgi?mgscsicables.8mm.html

3. 2x GD2682 Quad 68-pin Int TPO LVD Ribbon Encl. Cable $148.99 ea, $297.98, http://www.macgurus.com/shoppingcart/obj_show_page.cgi?mgscsienclosures.html

NOTE: what I don't know is whether this cable is long enough but it fits with ease in an 8-bay Burly. I can measure my GD2682 if necessary to compare with your Kingston.

4. 2x GD6299 68-pin MicroD LVD Active Diagnostic Termination $99.00 ea, $198.00, http://www.macgurus.com/shoppingcart/obj_show_page.cgi?mgscsiterminators.html

5. 2x GD1636 Remote LED Indicator (for use with Terminators) $19.00 ea, $38.00.

Questions, comments? k

kaye
08-08-2001, 09:12 AM
42,

Looking at my post and having a new day fresh prospective I would start with this:

1. ATTO ExpressPCI Dual-Channel Ultra3 SCSI Host Bus Adapter $459.99 ea

2. 1x GD4911 Ext. 68-pin .8mm to 68-pin MicroD 3-ft. Cable $159.99 ea

With the above you could hook up the ATTO in single channel to your existing Kingston Data Silo DS-400 configuration. See what the result is. Then decide about the other items, part of them or all. The RAID would work better in dual channel but getting the above would allow you to see if your problem is licked. k

42
08-08-2001, 02:03 PM
Sounds great, Kaye. Thanks for the pricing and info. Before I order, I want to follow up quickly on one comment you made:

<BLOCKQUOTE class="ip-ubbcode-quote"><font size="-1">quote:</font><HR>Anyway, I am worried about the possibility that one of those Cheetahs might be intermittently going bad.<HR></BLOCKQUOTE>

'Way back six months ago, when I made the initial posts, I reported odd findings, namely that the various utilities I was using (FWB, Silverlining, HD Setup, etc.) gave spotty results with several of the drives (occasional failures of one kind or another--sequential read/write, etc.). But eventually, all 6 did successfully take a low-level format. I talked this over on this forum, as well as with the Seagate tech, read the knowledge base at FWB, and consensus at the time was that my drives were probably fine. (Note that the whole megillah is plugged into a substantial UPS, so I've been ruling out the 'catastrophe' scenario, where a huge spike or whatever blitzed all 6 drives.)

Now this time around, drive 6 flunked Silverlining AND FWB, put then passed both. Because of my earlier experience (when my problems were virtually eliminated by adding the DG terminator), I decided to assume the drives were okay, since results were erratic (and seemed to improve), as they were/did before. Was I wrong to make that assumption?

Is there a way to test these drives so that I can rule out the SCSI/controller angle? Hmmm... I guess I could pop them one by one into the G4 itself, although I don't think it has LVD capabilities w/o using the Adaptec. But I could plug them individually into the Adaptec and at least rule out cabling and termination issues, right? What do you think?

kaye
08-08-2001, 04:24 PM
I have always used the MacBench 5 Disk and Publishing Disk tests, done multiple times, as the most stressful application to test drives on. If MacBench freezes during the test, something wrong. magician has said repeatedly that a low level format is the most stressful thing you can do to a drive. I have never had a low level format fail on a PCI machine, but I am careful about only using a utility that is claimed to be compatible with the OS I am running. Of course, if the SCSI card or cable or terminator is flakey, then the failure of a low level format could be the result of the drive or any of the other hardware variables. Hard to know where the trouble lies. Were you wrong to rule out the drive/drives? I don't know.

I don't think that trying one drive at a time with the Adaptec is going to rule out cabling and termination, at least not any more than six drives. Could be just one connector, or could be the cabling with terminator with six drives or the Adaptec with six drives. Maybe I'm missing something. If you see a crystal clear test to try, do so. I'm still thinking to try the UL3D and single Granite external cable, #1 and 2, I mentioned in the last post. If you have problems, then it has to be the Kingston internal cable or a drive.

You aren't in the Bay area, are you? k

42
08-08-2001, 05:27 PM
I like your thinking... it makes sense to me to grab the ATTO dc card and cabling and see what we see. If problems still exist, I figure I'm money ahead by having improved at least *those* components. (My next step at that point would probably be to recreate the array with different drive combinations--yechh.) But if the ATTO single-setup works well, then I'd probably go for the rest of the stuff and set it up for dual-channel.

How do I go about ordering this stuff? Website? Phone?

Thank you again for your studied counsel. This forum is now the first place I turn for hardware stuff, especially RAID/SCSI, I can tell you that!

42

PS-Location-wise, no, I'm in a little hamlet just south of L.A. (Carson).

kaye
08-08-2001, 07:19 PM
Your choice for ordering, either by the website (from the links I posted) or phone. If it gets complicated or you are unsure about anything, then an email to magican is the best bet so that he can check it over.

When I was ready to order the Burly, I just emailed my guesstimate of the parts list and he emailed back exactly what I needed. I then emailed back the order and called to provide my credit card info. Many ways to do this. k

42
10-18-2001, 10:35 AM
Kaye... quick update: I finally did do the things we settled on--replaced the Adaptec 2940U2W with an ATTO dual channel; replaced all cables with the Good Stuff (GD); added diagnostic terminators to both channels. Left all the drives alone but re-initialized after plugging in to the new config. Now I'm a week into running the array without a crash or hiccup of any kind. (Prior to changes, I was crashing several times a day.) So I'm guardedly hopeful that this will prove to be the fix. Also, the ATTO dual channel setup is much faster.

One note of interest... the Kingston DS400 required some backplane mods before I could connect the GD internal and external cables through it. The Kingston comes with a row of plates, one of which has two 'champ' connectors (which allows a cable and a terminator on the outside, and each end of an internal cable on the inside). I had to hunt for quite a while to find someone who sold the normal 68-pin connectors for the DS400. When they arrived, turns out they didn't fit the holes on the Kingston, so I had to get our metal shop to modify those plates. Bit of a nightmare.

But for now, all is looking well. Now, on my way to build a new subsystem for another department. Thanks tons for your help...

kaye
10-19-2001, 12:26 AM
42,

Glad that it is running well. Must be nice to have a metal shop. Expensive solution but it should be reliable and, of course, fast. Thanks for the report. k