Page 1 of 3 1 2 3 LastLast
Results 1 to 20 of 44

Thread: Miles2 SCSI questions..

  1. #1
    Join Date
    Jan 2001
    Location
    Prairie du Chien, WI
    Posts
    98

    Default

    I thought I'd better move our discussion off anarch's thread, I have a question for you Kaye.

    I plan on adding more HD's to my (2)miles2's besides the 2 15k cheetahs I already have going. I would like to stick with two addtional 15k cheetahs, and stripe all 4. (I need lots of fast storage for movies, and am aware my write speeds will not greatly increase, just love 15k cheetahs like Louie.) Will I have to get granite cables to pull this off?

    I read in your post to anarch that more than one cheetah on a miles2 original twisted cable will cause it to puke. Did I read you right? Or am I just misreading again? Could I get away with a 10k on the miles2 chain with the 15k without going granite?

    Also thinking that if this "write through" problem isn't solved soon, I would be better off getting a 10k and booting off it and writing to the 15k Cheetahs so I can uncheck "write through", and regain my processor speed. Is it a fair trade-0ff?

    I remember in your previous posts that you have an additional 10k HD on your Miles2 chain and you said it was the last drive on the chain. Did you mean the self-terminated end of the chain, (farthest out on the Chain from the miles2)?

    Let me know your thoughts...
    Terri

    [This message has been edited by tekcan (edited 19 March 2001).]

  2. #2
    Join Date
    Aug 2000
    Location
    Concord, CA
    Posts
    7,056

    Default

    Terri,

    If you add more X15s, I think you will need to go Granite. I did need to. But YMMV, so you might want to try the Initio LVD cables/terminators first before springing for the Granite stuff. Just to increase sustained reads tho, pretty expensive to add two more X15s. Plus, with four drives and all the cabling and more 4-pin power cables, it becomes a real headache to stuff everything in there and still get the cover on the box.

    If I recall correctly, I was able to add a 10k Cheetah (ST39204LW) to one Miles2 with one X15 and the Initio stuff. Again, YMMV. Two X15s on one Miles2, I had to go Granite.

    Any time you RAID, whether with one Miles2 or two Miles2, put the non-RAID drive last on the physical chain, whether it is a boot drive or not. Both Louie and I proved to ourselves that the system does not like looking past a non-RAID drive to access a RAID drive or drives. Yes, I have a 10k Cheetah (ST39204LW), a boot drive, on the same chain from one of the two Miles2 cards, each Miles2 driving two X15s, and the 10k Cheetah is last, next to the terminator, on one of the chains. Strictly speaking, this config is not symmetrical, but it works for me with Granite and the 10k Cheetah last on the chain.

    If you don't mind, give Louie your email address. He has some additional info for you. k

    [This message has been edited by kaye (edited 20 March 2001).]

  3. #3
    Join Date
    Mar 2001
    Location
    Dublin, CA USA
    Posts
    119

    Default

    Kaye, we have confirmed it is a cache coherency issue with the Mac OS and the Dual G3 Processor that will not allow us to flush the cache, and causes the SCSI Bus to hang. The only work arround for this is to change the CPU cacheing option. We will try and work with the CPU upgrade people, but Winston is not real hopeful.

    Bob McWilliams

  4. #4
    Join Date
    May 2000
    Location
    Cordova, TN
    Posts
    2,428

    Default

    Bob:

    I'm confused! You say "the Mac OS and dual G3 processors" are at the heart of the problem and Winston is not hopeful.

    I happen to be one of several on these forums that have been waiting patiently for a fix and hearing nothing but optimistic reports until now. What is most confusing is that none of us that I know of that have the problem have "dual G3 processors". We have single G3 processors, dual Miles U2W cards and multiple X15 Cheetahs striped in RAID 0.

    All of us in this situation have spent a couple of thousand dollars or more for two Miles U2W cards, Granite cables and terminators, and X15's because "that was the way to get the most performance out of a legacy, six-slot Mac"; a philosophy preached here. Many of us have touted this combo to others who have also ponyed up the high dollars. We all thought a fix was forthcoming.

    As it stands for me, all those dollars are wasted without a fix. By having to enable write-through, my overall performance is no better than it was when using single UltraNarrow drives hosted by a BlueNote.

    I am one of those that first brought this problem to light on these forums in November. Others began reporting the same problem. No obvious concern was shone for any of our situations except from Kaye. Kaye went to a great deal of trouble to duplicate the issue on his machine and loaned it to you for two months. Each report of the two reports from you to Kaye sounded positive including the one from last week which indicated that you might even have a fix posted last Friday.

    It's not your fault the problem exists, but I firmly believe we all deserve a better explanation of the discrepancies between your earlier reports and the one today.

    As much as I love upgrading old Macs and as much as I love this forum, if this issue is not fixable, I will seriously consider getting out of the upgrade field and quitting this forum because I've mislead people.

    If any of you feel that this communication should have been private to Bob, I'm sorry. I felt that it had to said here because he hasn't answered two earlier emails from me.

  5. #5
    Join Date
    Oct 2000
    Location
    Colorado Springs, CO
    Posts
    936

    Default

    Louie,

    I totally understand what you are saying and firmly believe that the post you wrote SHOULD be public domain. It is an issue that we have all dealt with, either reading about it or doing it or staying away from doing it (that's me). And I am just as interested in this as you and some others. Granted I have almost nothing at stake here, but I (for one) would absolutely hate not having you here. There is a big issue here, and I believe that hardware manufacturers have an obligation to the devout elitists that preach the gospel of high-end-upgrades. The Gurus are some of the finest troubleshooters I have seen. They (we) ride the thin line sometimes and find things out that we sometimes wish we didn't. And lots of times are even more compelled to love what we do.

    So here's to hoping and praying that they get on the ball that we all thought they were on.

  6. #6
    Join Date
    May 2000
    Location
    Cordova, TN
    Posts
    2,428

    Default

    I've been trying to think of any dual processor G3 machine and I don't think there were any. We had dual 604's and dual G4's put no dual G3's.

    And, DSM, thanks for the moral support. This problem has been eating on me since Nov. and I've convinced several others to buy the same expensive goodies because I thought a fix was a hand. That hurts worse that spending my own money which I couldn't afford. You are right about finding out things that you wish you hadn't. I've been there several times in the last 2-3 weeks.

    [This message has been edited by Louie (edited 20 March 2001).]

  7. #7
    Join Date
    Oct 2000
    Location
    Colorado Springs, CO
    Posts
    936

    Default

    I remember reading on the XLR8 and Moto site that the G3-series doesn't even benefit from MP. It doesn't support it within the OS, and MP carriers populated with G3's would run only one processor. I have no idea what he was saying.

    I hope he can clear it up a whole bunch before we all drop off the deep end. A statement like that NEEDS to have more info to back it up.

  8. #8
    Join Date
    May 2000
    Location
    wherever I hang my hat
    Posts
    3,575

    Default

    guys, maybe I'm just not understanding something here.

    write-thru, as I understand it, is required when running G3's faster than 500MHz in machines which originally shipped around five years ago?

    so....if you go slower than 500MHz....generally write-thru isn't necessary. Right? Of course, a lot of this stuff is case by case. But to be in the 500Mhz G3 club with some vintage machines, like the PTP, you need to compensate for a cache coherency problem in MacOS by enabling write-thru, right?

    now, when you throw a Miles2 and an X15 into the mix, or even a pair of each, then you need write-thru enabled to successfully boot off the array, right?

    am I missing any of this so far?

    now... machines configured in this way, or shall we say, PowerTower Pro's configured in this way, with 500Mhz G3's....and dual Miles2 with at least two x15 Cheetahs....such machines deliver 75MB/sec sustained reads and writes, right?


    When Louie writes:

    quote:


    As it stands for me, all those dollars are wasted without a fix. By having to enable write-through, my overall performance is no better than it was when using single UltraNarrow drives hosted by a BlueNote.




    I'm wondering if I'm crazy? I really don't get it.

    I certainly don't see how Louie has misrepresented anything here, and we all know that he would never do anything of the kind deliberately. Does the CarrierZIF G3 500 rock in the PTP? In my opinion, it certainly does. Is there a faster upgrade available for that machine? Not that I am aware. So, enabling write-thru inflicts a performance hit. Ok. But how are we all measuring that hit? And in what context? Call me an optimist, but I'm pleased that this hardware runs at all! My 9600 doesn't appear to suffer this problem, but it has others. You take the good with the bad, and you find the right solution for you. At least, that's the way I see it.

    Now, high-tech stuff changes every day. So today, we understand the write-thru issue better than yesterday. Have we disseminated falsehoods? Not as I see it. I have told the truth as I knew it every day. Knowing Louie as I do, he has never knowingly lied to a customer. We tell the truth here, even when it hurts. I had a customer swear at me in email last night, no kidding, because he didn't like what I had to say about PC133 SDRAM. He said it lost me a sale. Well, ok, then. Whatever. I can live with myself, and that's more important.

    So, I just don't get the drama here, and I don't understand the emotions that are at play here.

    the way I see it, whether there is a fix or not, there is no other way to get a PowerTower Pro to run at 500Mhz, and no other way to get a PowerTower Pro to deliver sustained reads and writes around 75MB/sec.

    these machines, as far as I am aware, are actually faster in numerous benchmarks than my own 9600. Believe me, that doesn't make me hate my machine, or think that the money I spent upgrading it was wasted. No way. I use it every day, all day long. And I love it.

    so seriously, now.

    What the hell is it that I am missing here?

  9. #9
    Join Date
    Oct 2000
    Location
    Colorado Springs, CO
    Posts
    936

    Default

    I think it is that Mac users were told a fix was in the works and they would be able to scrape every bit of performance out of the combo that they could. Some machines have problems with the setup, and Initio made a couple statements that things were in the works. K, back me up. I was under the impression (from the threads) that Initio had K's machine in order to better get an understanding of what was going on. They recognized the issue, no? Then said they had a fix being worked on?

    I guess it all boils down to some folks being told that they could get something, then being told they can't. After months of being strung along. And after MANY recommendations for the companies involved. And the performance just isn't the same on similar machines. Not that things have to be equal on a 9600 and a PTP. But it'd be real nice if they were (both ways). Especially after spending TONS of $$ on upgrades. I would love to get a new Mac. But not for a while. And til then, I want to get every tiny little bit of performance out of my machines.

    I guess the drama is stemming from all that "Indian-giving". If I was Louie, I'd be a bit upset. And I want to know what K has to say.

    But Magician....what MP G3 machines is he talking about? And how does that have any relation to a single-processor dual-Miles setup? Let's start with clearing up the statements....

  10. #10
    Join Date
    May 2000
    Location
    Cordova, TN
    Posts
    2,428

    Default

    Magician:

    Cooksey himself, the man that wrote the software for XLR8, said right here on these forums, that the Write-Through box in the Control Panel doesn't open to warn you that may need it until the processor speed is 510 MHz. When I started this mess, I had a 466 processor running at 480. Then I bought a 500 hoping it would help the situation. It didn't. I've never run over 500. No one has ever even implied to me that write-through would be necessary over 500 MHz and I've never seen it written anywhere. I never heard the term "cache coherency" till Bob used it today and now you.

    Cooksey also said that Write-Through costs 5-7%, I think. That depends on how you measure. It causes all memory thruput over the backside cache size (1MB) to be written to RAM which takes a terrible toll.

    I've several benchmarks that show that having to enable Write-Through costs as much as was gained by the expensive RAID. I see these results by running the same tests booting from an UltraNarrow drive and turning off write thru. I've spent $2000 for zero overall performance gain.

    I tried to explain this to you back in Nov-Dec and your reaction was just like tonight's.

    If you don't understand my concern then I don't know how to explain it. I spent $2000 that I couldn't afford to have the highest performing PTP possible. No one said or wrote that I would have to cripple my machine with write-through. For my money, I got zero gain.

    You're right, I never intentionally misled anyone, but I did encourage others to go ahead and do the same thing that I did thinking that there was a fix coming right away. That's why I'm upset and your attitude that I should not be is a significant part of it.

  11. #11
    Join Date
    May 2000
    Location
    wherever I hang my hat
    Posts
    3,575

    Default

    well, Louie, you and I are entitled to hold different opinions, and just because I don't assign as much gravity to some issues as you do doesn't mean that I don't hold your opinions in respect and esteem.

    you seem to be upset because I don't see this the same way that you do. Well, you shouldn't interpret my different viewpoint as anything more than that. It certainly isn't a comment on my esteem and respect for you, for your life accomplishments, and for the good work you do every day here, helping regular folks out with their problems. Just because you and I don't always agree doesn't mean that we can't value our fellowship, or even kid one another about these things.

    That said, I'm still trying to understand, if only because you want me to do so. You're upset because you run some benchmarks when you boot from a narrow drive and get the same results as when you boot from an array with write-thru enabled? Ok. I think I get that much. Do you not reap performance gains anywhere else? Or are you focusing on a specific, finite set of benchmarks at the expense of an overall assessment?

    if you are saying that the overall performance of a PTP with Miles2 and X15's is equivalent to that of a PTP with single-ended narrow drives and a Bluenote, as long as write-thru is enabled, I would find it hard to believe. Still, since you are Louie, I would have to consider that statement very carefully, and confirm it, rather than just dismiss it. While I do not always share your perspective, your skills and expertise demand respect and consideration.

    So. If you can persuade me that overall performance between an X15'd PTP is the same as that as a PTP with a Bluenote and a SE narrow drive, what then? Is there still not value to be gained using that hardware? For example, Terrible Terri has been using her upgrades to capture video and make movies, and is apparently even considering another pair of drives, even though K has alerted her to the usual downsides, if I am not mistaken. Does this mean that her machine is not much more usable than it was? Or that she is not reaping even more value from her machine?

    merely because write thru needs to be enabled on some PTP's does not, in my opinion, lessen the value of the upgrades in those machines, depending on what you are doing with the machine, and what sort of performance enhancement you are seeking. By the same measure, the superior PCI performance of the PTP over the 9600 is merely something to be aware of--there's not a lot any of us can do about it. It's built into the machine. I still find the 9600 to be the best machine for working as I work. And this is not the first weird thing we have learned about PTP's, as we all well know. Steve Jobs would have us all discard our machines every other year, or even annually. I don't know about anyone else, but I can't afford to do that. I've been riding this 9600 for a few years now, and my next machine will need to deliver much better performance to entice me away. No way I expect my next machine to deliver this much value or longevity. I think Apple will make sure it never does that again.

    this is merely my opinion, at this point, and I see no reason why anything I believe should upset you. You know that you have utter license to speak the truth as you know it on these forums, and all that I have ever asked from you and the other moderators is that you not get us sued. Telling the truth, as we know it, at all times is the best protection against that sort of thing, as other companies can't sue if we are merely stating facts.

    As I read Bob's statement, he is obviously confused between G3's and G4's, as G3's have never been amenable to SMP, and this issue of write-thru in PTP's has never been related to SMP or G4's as far as I am aware. I do think that cache coherency is part of it, and by that, I mean the requirement to write data thru the cache to main memory to ensure that the contents between the two correspond. Corruption could obviously ensue if incorrect data were retreived from the cache. You do see issues with cache coherence when using dual-G4's, particularly in machines that don't handle speculative processing correctly, but that is irrelevant in this case, as I understand it.

    I do recall reading somewhere that write-thru is automatically toggled when processor speeds exceed 500MHz. I'll have to search around to find that source, or maybe we can get Cooksie back in here to say something intelligent about it all. I could almost swear it was in his post here. We'll have to search the archives for it, I guess.

    But what I do get from Bob's statement is that he has talked to Winston, their driver programmer, and fixing this issue will not apparently be a matter merely of changing some parameters in the Miles2 firmware. It will require some assistance from the daughtercard companies, as well. As you can imagine, getting something accomplished within one company is difficult enough. Getting something accomplished when two different companies must collaborate is ambitious, indeed.

    That said, I can think of no two companies that listen to users better than Initio and XLR8, so we shouldn't assume that this is the last chapter. If we can make a compelling case to these guys, who are, after all, focused on profit, that this deserves manhours and effort from their programmers, maybe something can actually be accomplished. The wild card in that scenario, of course, would be Apple. It may be that the problem is so deeply rooted in the way that MacOS works that nothing can ultimately be done. I think we can all assume that Apple will have little interest in listening to anything we have to say. They have historically ignored us, and frankly, that hasn't always been a bad thing.

    If that proves to be the case, as I understand this issue now, I still wouldn't agree that you had deceived anyone. Certainly no one here even had a glimmering that toggling write-thru was anything more than a troubleshooting measure that improved stability in certain circumstances. I stumbled on the write-thru option when dealing with audio engineers working with XLR8 daughtercards. These guys could get nowhere with Sonnet boards. And their XLR8 boards were finicky, but only when using audio applications. Toggling write-thru, for some reason, delivered rock-solid stability, they were happy, I was happy, and I moved on to other pressing matters, which as you know, arise daily in this business.

    Anyway, you know that you don't need my agreement to be pissed off about anything. If this issue pisses you off, fine. If you need me to do something, or assist you, just let me know. You and the other moderators should never fear that I intend to muzzle you. Speak freely and clearly. Sure, we should have some manners, and we all know we should keep our dirty laundry within the family, but your mission here is to help people the best you can, using all your experience and expertise. If that means that some upgrades should not be recommended in some cases, fine. How many times have we warned folks not to buy ATTO boards for use in vintage Power Macs? How many times have we warned six slot owners to be careful where G4 upgrades are concerned? You're entitled to be upset if you need to be, and I sincerely apologize if my inability to appreciate this problem as you do has caused you stress. Don't put that on me! I have too many other things going on, and I have no desire to alienate anyone, least of all my moderators.


  12. #12
    Join Date
    Jun 2000
    Location
    Nederhorst den Berg., Netherlands
    Posts
    144

    Default

    The problem has nothing to do with running over 500 mhz on a legacy machine nor using dual G3's, hmm dual G3's??

    Furthermore I want to emphasis its not just a bootproblem, once you have the problem and not boot of the dual miles/cheetah raid, write back enabled, the raid will be useless causing random freezes when accesssed.

    Underneath a memory bandwith test comparing write back vs. write thru.
    The performance penalty with write sizes up to 1 MB (the size of the backside cache of course) are obvious:

    Write back
    +---------+----------------+----------------+
    | Size | Write | Read |
    +---------+----------------+----------------+
    | 4 KB | 781.0 MB/s | 794.5 MB/s |
    | 8 KB | 794.5 MB/s | 781.0 MB/s |
    | 12 KB | 794.5 MB/s | 781.0 MB/s |
    | 16 KB | 794.5 MB/s | 794.5 MB/s |
    | 20 KB | 708.9 MB/s | 649.0 MB/s |
    | 24 KB | 677.6 MB/s | 555.2 MB/s |
    | 28 KB | 658.3 MB/s | 485.0 MB/s |
    | 32 KB | 658.3 MB/s | 475.1 MB/s |
    | 40 KB | 639.9 MB/s | 480.0 MB/s |
    | 48 KB | 640.0 MB/s | 480.0 MB/s |
    | 56 KB | 667.8 MB/s | 465.4 MB/s |
    | 64 KB | 640.0 MB/s | 465.5 MB/s |
    | 128 KB | 640.0 MB/s | 460.8 MB/s |
    | 256 KB | 640.0 MB/s | 460.8 MB/s |
    | 384 KB | 640.0 MB/s | 443.1 MB/s |
    | 512 KB | 640.0 MB/s | 460.8 MB/s |
    | 640 KB | 639.6 MB/s | 426.4 MB/s |
    | 768 KB | 548.6 MB/s | 360.0 MB/s |
    | 896 KB | 522.6 MB/s | 359.3 MB/s |
    | 1024 KB | 500.9 MB/s | 360.0 MB/s |
    | 2048 KB | 134.7 MB/s | 87.3 MB/s |
    | 3072 KB | 132.6 MB/s | 86.9 MB/s |
    | 4096 KB | 134.7 MB/s | 86.3 MB/s |
    | 5120 KB | 133.9 MB/s | 87.2 MB/s |
    | 6144 KB | 132.6 MB/s | 86.9 MB/s |
    | 7168 KB | 135.0 MB/s | 85.9 MB/s |
    | 8192 KB | 134.7 MB/s | 86.3 MB/s |
    +---------+----------------+----------------+

    Write thru
    +---------+----------------+----------------+
    | Size | Write | Read |
    +---------+----------------+----------------+
    | 4 KB | 781.0 MB/s | 794.5 MB/s |
    | 8 KB | 794.5 MB/s | 781.0 MB/s |
    | 12 KB | 781.0 MB/s | 794.5 MB/s |
    | 16 KB | 781.0 MB/s | 794.5 MB/s |
    | 20 KB | 303.2 MB/s | 677.6 MB/s |
    | 24 KB | 170.7 MB/s | 576.0 MB/s |
    | 28 KB | 153.6 MB/s | 485.0 MB/s |
    | 32 KB | 151.1 MB/s | 480.0 MB/s |
    | 40 KB | 147.7 MB/s | 465.4 MB/s |
    | 48 KB | 149.1 MB/s | 480.0 MB/s |
    | 56 KB | 147.7 MB/s | 465.4 MB/s |
    | 64 KB | 146.3 MB/s | 465.5 MB/s |
    | 128 KB | 144.0 MB/s | 460.8 MB/s |
    | 256 KB | 142.2 MB/s | 443.1 MB/s |
    | 384 KB | 142.2 MB/s | 460.8 MB/s |
    | 512 KB | 142.2 MB/s | 460.8 MB/s |
    | 640 KB | 142.1 MB/s | 442.8 MB/s |
    | 768 KB | 138.8 MB/s | 371.6 MB/s |
    | 896 KB | 136.9 MB/s | 370.9 MB/s |
    | 1024 KB | 138.8 MB/s | 349.1 MB/s |
    | 2048 KB | 137.1 MB/s | 86.3 MB/s |
    | 3072 KB | 137.5 MB/s | 86.9 MB/s |
    | 4096 KB | 137.1 MB/s | 87.3 MB/s |
    | 5120 KB | 138.9 MB/s | 86.2 MB/s |
    | 6144 KB | 137.5 MB/s | 86.9 MB/s |
    | 7168 KB | 137.5 MB/s | 86.9 MB/s |
    | 8192 KB | 137.1 MB/s | 87.3 MB/s |
    +---------+----------------+----------------+

    Ton

    (pm 9500/metabox G3 500/1mb running @ 480 mhz/OS 9.1/ 1 gig memory interleaved)]

    BTW: While composing this post I missed mag's last reply

    [This message has been edited by Ton (edited 21 March 2001).]

  13. #13
    Join Date
    May 2000
    Location
    wherever I hang my hat
    Posts
    3,575

    Default

    ok, I understand better, now. I am aware that it is not possible to use the array without write-thru enabled in some cases.

    what we're all dealing with here is a MEMORY BANDWIDTH issue?

    interestingly enough, this was in a 9500--not a PTP. So it is more a Tsunami thing, maybe, than just a PTP thing, huh?

    what app was used to capture these results?

    I'm almost impressed that you could even get the machine to run with the cache in write-back mode.

    BTW, Ton, your replacement array shipped today. You should have it shortly. Don't forget the check the power switch!


  14. #14
    Join Date
    Jun 2000
    Location
    Nederhorst den Berg., Netherlands
    Posts
    144

    Default

    To magician:
    As I recall correctly I was the one discovering write thru was "the cure" for the dual Miles2 / Cheetah raid problem after an extensive thread here on this forums last august or september (not quite sure about the date). Anyway I'm quite sure no one "adviced" me trying this option.
    At first I was glad I had a usable machine again, now still hoping for a definate cure.

    Why are you impressed I can run this machine write back enabled? before I installed Miles 2 cards ALWAYS ran write back enabled.

    Ton

    (I'll keep an eye on the powerswitch this time )

    [This message has been edited by Ton (edited 21 March 2001).]

  15. #15
    Join Date
    Aug 2000
    Location
    Concord, CA
    Posts
    7,056

    Default

    Ton, I think you are correct that you discovered the cure by enabling write-thru. I had to go back thru the archives and my notes to recall all of my experiences.

    Starting from scratch, meaning 2x Miles2 and 4x X15s and Initio firmware 1.06b and SoftRAID 2.2.2, I had no problems either with booting, from any drive or the striped RAID, nor did I get any freezes with large size file transfers with the RAID, all this with write-thru DISABLED and running xlr8 G3-500 at 540/270. The only thing that changed, when I went from OS8.6 to 9.0.4, I had to use System Picker to select a boot drive.

    Just to get it out of the way, in the above and following tests, I never had the no-boot problem (freeze after the xlr8 extension loads that Louie describes) in any configuration that I tried. I was never able to duplicate that and I don't know why.

    In an effort to duplicate Louie's and Ton's other problem, the freeze during large size file transfers with write-thru DISABLED, I tried other RAID drivers besides SoftRAID, HDT, Anubis RAID, ATTO RAID, as well as dual ATTO UL3D cards instead of Miles2. I won't bore you with all of the details, suffice it to say that I could now duplicate the RAID freeze during large file size transfers with write-thru DISABLED. What did that tell me? Namely that I could not point the finger at SoftRAID.

    Now I live close to Initio but not ATTO. So, in an effort to get the dual Miles2 to choke, I started playing with Initio firmware versions and SoftRAID versions. My purpose was to be able to take my PTP to Initio with the problem of large size file transfers puking/freezing on the RAID with write-thru DISABLED. Actually I was near the end of my rope trying to make this happen with dual Miles2.

    As it turned out, I could duplicate the puke with Initio firmware 1.05 and SoftRAID 2.2.1 with write-thru DISABLED, 2x Miles2 and either 2x or 4x X15s. So that is the configuration that went to Initio with 4x X15s.

    Some comments:
    1. Initio firmware 1.06b is only necessary, so far anyway, for Seagate drives with firmware 0002. I could use Initio firmware 1.05 because I made sure there were no Seagate drives with firmware 0002 in PTP.

    2. I have great sympathy for Louie's and Ton's position, and essentially that is why I went thru this to at least duplicate one of their problems. I was never able to duplicate the boot problem.

    3. Initio Bob and Winston have essentially two problems which they feel are parallel or similar, and I think Bob is confusing them, while Winston is actually working on the problems. This is my supposition anyway. These two problems are our problem with single G3 and dual Miles2 requiring write-thru ENABLED and a problem with dual G4 (they have a 533 DP). Both problems relate to the OS not properly handling cache flush. Or better stated, the OS does not completely flush the cache before more stuff floods in, causing a hang. Bob and Winston have been to several seminars and meetings at Apple begging for a solution, and I guess Winston has been in touch with Apple PCI engineers over this issue.

    My interpretation of what they were trying to attain, as a temporary measure, was to somehow ENABLE write-thru only during large size file transfers and only for a small amount of time to allow the cache to flush completely. So normally write-thru would be DISABLED, but if the Miles2 cards detected an incomplete cache flush, write-thru would be ENABLED until the cache was completely flushed.

    4. Unfortunately, with the problem with the OS, this has turned into a much harder nut to crack. And why my PTP worked before I started all of these changes, I don't know, and Louie and I have pondered this to no end.

    5. One other tidbit, going back to SoftRAID 2.2.1 (on the RAID only) cured the necessity of needing System Picker with OS9.0.4 for me.

    There is some stuff that Bob asked me not to repeat because it is in the Apple domain at this point. If I have given any secrets away, Bob, I did so unintentionally, and only meant to address this issue that we are confronted with and to clarify that there are two issues with cache flush. If there are more issues or questions I have not addressed, post. k

    [This message has been edited by kaye (edited 21 March 2001).]

  16. #16
    Join Date
    Jan 2001
    Location
    Prairie du Chien, WI
    Posts
    98

    Default

    What a fantastic thread! We are finally airing our laundry in the open as the term, "forum" implies! I think we were all holding back because we thought if we were nice, we would get our fix.Ha! Because of the rhubarb, Bob has posted in anarch's thread that he has a renewed interest in getting our problem fixed.I don't generally cross-post, but I'm not supposed to type 'til my shoulder heals, so I copy-pasted my reply to him here:

    "Go for it Bob! We really need this fix, I know of at least a dozen people waiting to upgrade thier PTP's and 9600's, some are already settling for "brand X's" SCSI card since they don't think you'll come through for us! We also have several machines at work that will get the upgrade if a fix comes through.
    It's not just a few machines we are talking about here, for every person who posts, there are 25-50 who just read along and learn. Macgurus success as a vendor proves it."

    Hey Guys, This problem is happening to 9600's as well as PTP's. It's happening with 466 G3's as well as 500's. It's happening with 10k Cheetahs RAIDed, as well as 15k cheetahs. I even recall reading it happens to single HD's.(please speak up if you are one of these to confirm it) Now that we're talking, this problem is showing itself to be bigger than we all first thought.

    Ton came up with one work around, and Kaye hasn't experienced the problem at all.
    Perhaps there is another work-around, if Kayes machine works, maybe we should look into why it does. We both have the same board (05), but I have to enable "write through". We know it's a cache problem, but not a consistant problem across similiar machines.

    Lets keep it up and out in the open, if your machine doesn't work unless write-through is enabled, lets hear from you and document it for initio & xlr8. If your machine works, try to figure out why and pass it on.

    Here is the email I sent to Louie, I think I should share my support of Louie with all of you.

    I'm glad I got the miles2 & 15k cheetahs, Louie, and value your advice
    highly. I knew the risks going in and thanks to your help, knew what to
    do when things went wrong. There was a chance I wouldn't have to enable
    "write through" and I took it.

    We are far from finished with this battle yet, me may soon discover a
    work-around that gets us fully up to speed. I want you to be there
    fighting for us. I don't think you realize how important you are to the
    forum, PLEASE don't even think about giving up.
    Wanted to personally let you know how I
    feel about your advice and help with my upgrade.

    I have to stop typing, I'm on heavy drugs for pain, so it's easy to abuse myself when the pain is covered up...

    Keep up the good work!

    Terrible Terri.

  17. #17
    Join Date
    Aug 2000
    Location
    Concord, CA
    Posts
    7,056

    Default

    Terri,

    Do you have the boot problem that Louie describes with write-thru disabled? Or do you have only the freeze with large file size transfers with write-thru disabled? Or is your problem different? k

  18. #18
    Join Date
    May 2000
    Location
    wherever I hang my hat
    Posts
    3,575

    Default

    at this point, I am running beast with a 500Mhz XLR8 CarrierZIF G3, a single Miles2 (firmware 1.06b) with dual X15's and a 36GB Cheetah backup drive. MacOS 9.1. I have MAChFire FireWire and DualPort USB PCI cards installed, and dual Radeon and one Rage 128VR.

    no boot problems, no write-thru required.

    I have Radeon problems right now, but that's it. When I was running Nexus previously, beast was fast and solid as a rock. I rebooted him about once a month. The machine literally has not been shut-down at anytime for the past three years unless I was shuffling hardware around. Ok--one time I recall I shut-down because we had a thunderstorm and the UPS's were screaming. That's it.

  19. #19
    Join Date
    Jan 2001
    Location
    Prairie du Chien, WI
    Posts
    98

    Default

    I freeze up on boot from the RAID 15k cheetahs. When booting from my original HD, 2.1gig Seaghate, I get random crashes. Checking write-through eliminates these problems.

    I would like to point out that although I am very happy with my upgrade, I do expect initio and/or xlr8 to address the problems we are having. I think I deserve full processing power, considering my investment, as was led to believe that a fix was in the works.

    Mag, are you using granite cables?

    I also note that both you and Kaye aren't having the boot problem, and you both have an addtional HD on the SCSI chain. Could there be a connection? Perhaps adding another HD slows the chain just enough to allow boot-up?

    Would like to hear more on this.

    Terri

  20. #20
    Join Date
    Jan 2001
    Location
    Austin, TX, Los Estados Unidos de Mac
    Posts
    672

    Default

    Mag is not using dual Miles2s, just one. That would seem to be the reason he's not experiencing the write-through problem.

    Kaye, on the other hand, seems to have found the Holy Grail of PowerTower Pros- a machine that is somehow immune to the write-through problem, and for that matter, can take on a 500MHz G3 overclocked to 540!!!!

    [This message has been edited by the_anarch (edited 22 March 2001).]

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •