Wide SCSI IHD on BlizzardPPC SCSI woes!

[Floid]

Roj wrote:
I have a hunch termination isn't the only problem here though. The drive is a 16-bit drive, but putting the 68->50 pin connector limits it to 8-bit access. Assuming the Blizzard is the same as the CSPPC, get to the preboot menu and force the drive to 8-bit rather than Auto or 16-bit.

I have a hunch termination is the problem, but in a different way.  Is this adaptor a simple mechanical adaptor, or a *high-byte terminator?*

Drives have long been *rumored* to be smart enough to operate in narrow mode without special adaptation, and one would think an onboard terminator would help guarantee this  - but on the other hand, remember that *both ends of the bus must be terminated,* and even with an onboard terminator, that's only providing half the expected resistance.  I've had this exact problem attaching some IBM drives to a narrow PCI controller (those were low-cost units surplus from a RAID somewhere, and the onboard termination option was left off their boards) ... switching from $6 "wide-to-narrow" adaptors to $15+ versions specifically including a high-byte terminator (yes, active) allowed things to function.  Sadly, I forget the best vendors to search; eBay is probably the most affordable route, as long as you're purchasing from someone who's sure about what they're selling.  (Remember, the keywords "high-byte terminator," "high-byte terminating" HD68M/50F adaptor good; "wide-to-narrow"/"narrow-to-wide HD68M/50F" bad - or rather, that promise really says nothing about whether it will actually work, and sadly, low prices usually indicate it's *not* properly constructed with a terminator inside.  When I had my troubles, I started with adaptors purchased from either Dirt Cheap Drives or Megahaus - can't remember now, but both are huge, respected disk resellers.  They were advertised for putting wide drives on narrow chains, but nope, no termination!)

Also remember that SCSI is, in general, rather picky about its cabling, so designing a topology that reduces the length of 'tees'/'stubs' off the bus is a good idea.  In other words, if you plan to have many wide drives, use a single HD68F/50F (or M) high-byte-terminating adapter to attach a proper wide cable (terminated on the end by a proper, full-width terminator, which you can find in one of the drives) - don't use individual adapters on each drive if they'll be sitting in the middle of a chain.  Similarly, if you're going to get into the cheap-surplus SCA drives, it's a good idea to find a proper backplane, rather than using many adaptors off a regular cable.

[Floid]

Okay, some FAQs, vendors, etc:

http://www.cablemakers.com/scsitermination.htm
http://www.ramelectronics.net/html/wide-narrow-scsi-adapter.html
http://www.cablestogo.com/product_list.asp?cat_id=109 - Note how specific models are marked 'high-line terminating;' that means the rest aren't, despite the cost!
http://www.provantage.com/buy-22063539-adaptec-cables-converters-scsi-external-shopping.htm - Adaptec makes and sells a number of adaptors; notice that some are specified as high-byte terminating.  There're a bunch of FAQs and diagrams somewhere on Adaptec.com, too.
http://www.pc-pitstop.com/scsi_adapters/ - Best deal in USD I've hit so far on Google.  Yet again, note the high-byte models, and how you can't tell the difference without test equipment!

There's a great, graphics-heavy article out there somewhere, but I can't find it tonight.

Remember, the idea is to terminate the upper lines off the drive for the *drive's* sake.  They seem to look for appropriate termination *first* (maybe the bus transceiver chips just don't work at all with some lines floating... Or maybe the drive 'smartly' checks for proper termination to avoid burning them out, before negotiating to communicate with the host controller narrowly... Technically, devices could talk to eachother with 16-bit transfers on a wide segment even when the host controller is narrow!, but I don't know if it works that way in normal configurations.*) ... Whatever's going on, if they don't get it, some will refuse to 'talk' at all.

As to 'active' vs. 'active negation' - assuming an "Ultra"-clocked bus, if you're terminating the drive against itself (those upper 8 lines connect to nothing but terminators), it probably doesn't matter at all; no sense paying extra for signal quality on lines that won't carry signals after you've convinced the drive to wake up.  If you have more than one "Ultra"-wide device connected to each other on a wide chain, then it can't hurt to have 'active negation' - if they need it, one of the ultra-wide drives' internal terminators would offer it for one end; if the terminator at the end of the narrow segment doesn't support it, it probably won't be any improvement to have it on the terminator capping off the high byte.  (On the other hand, if you *have* it on the high-byte, you'll only need to buy/try one more improved terminator for the narrow end if you notice bus errors or data corruption - From my experience, an unlikely problem, but it's something to be aware of.)

With those IBMs, I thought the same things others have said - "maybe they're LVD," "maybe they're defective," "they're Fast drives; maybe they won't negotiate for the clock speed of the Ultra bus," "maybe they just won't work in narrow mode at all" -- but as soon as I swapped in the proper terminating adaptor, there they were, working great!  (All this with a normal, 'full-width' active terminator at the end of the wide end of the chain, of course.)

Once you understand the concepts involved, it's easy to make the right decisions about these things - the hard part is finding someone who'll sell you the proper hardware!  

(Hmm, if it says "wide to narrow" or "narrow to wide," and you always need high-byte termination when transitioning between widths, wouldn't you think they'd include it?  No, they count on you having one of the rare drives that does the right thing itself... or they just don't understand the technology at all, even though they're happy to take your money.)

*Non-'normal' configurations include such things as multiple host controllers on one bus - good ol' SCSI is a highly extensible architecture!

[Floid]

Dragster wrote:
Well, finally my troubles are gone, the drive got recognized and it's working now

Congratulations!  Good it hear it all worked out.

What happened? Well...

After doing a lot of tests without success, I set the SCSI ID of the drive to unit 0, removed the termination jumper and put the jumper for the "force narrow" option, then I went to the SCSI menu of the blizzardppc and put the controller itself as unit 7 (the last one) and voilá, it worked!!!

So now the chain is termination (active with the controller if I understand correctly the blizzppc manual), then the HD as ID 0, a Plextor 40X cdrom as unit 1 and a Yamaha CDRW as unit 2 with the termination jumper ON.

Okay, if I'm reading this right, you've got one *narrow* chain:

 [Yamaha Narrow Terminator]
          [Yamaha CDRW]
                   |
         [Plextor CD-ROM]
                   |
    [IDC50-to-HD68 Adaptor]-[Fujitsu HD in narrow mode]
                   |
    [Host controller in BPPC]
   [BPPC Narrow Terminator]

(Yes, the spacing is ugly - the board's eating my formatting - but I've made it so it stays accurate.)

Anyhow, drawn that way, you can see it looks pretty good.  You have one narrow chain, with narrow terminators on each end.  The terminator in the Yamaha is probably active; I wouldn't worry about that.

The 58-to-60 adapter creates one small stub/'tee' electrically, but it's supposed to; you can't help that.  It only becomes a big deal when you try to wire an entire chain with adaptor stubs on each drive, as when some people try to use 7 SCA drives with adaptors rather than a backplane.  It's like a dip in a road; cars are built to handle them, you shouldn't even notice one, but cover the street with them and you might spill your coffee!  (Stubs create a few added reflections, and maybe a little impedance mismatch.  The terminators are designed to 'absorb' that - the connectors on the cable and on the drives are already small stubs themselves.  It's when you overload things - creating a lot of longer, weird stubs, by using a lot of adaptors on each drive, or doing something silly like trying to split the cable with a Y-adaptor - sort of like trying to drive a car 'around' a tree - that your shock absorbers give out!)

There's no rule for how many adapters you can use; it's just that each one adds a little more risk.  If you ever add a second or third wide drive, *then* I'd worry about getting a single high-byte terminator and a regular wide cable:

[Wide terminator in wide drive]
[Wide drive]
||     <- Wide (two 8-bit channels) cable.
[Wide drive]
||
[Wide drive]
||
V     <- HD68F-to-IDC50M high-byte-terminating adaptor
|      <- Regular 50-pin narrow cable
[Narrow CD-RW]
|
[Narrow CD-ROM]
|
[Host controller]
[Narrow terminator in host controller]

...as you can see, that doesn't create *any* added stubs, and direct transfers between the wide drives might execute with full 16-bit widths.

Thanks again for all your useful answers.

So.. do you think this setup is donde correctly or should I remove the passive tertrmination jumper from the cdrw?

Looks beautiful.  One improvement could be to move the wide drive to the *end* of the chain, with its terminator on (and taking off the terminator on the Yamaha, once it moves to the middle of the chain) - but since Fujitsu were nice enough to give you that "force narrow" jumper, something I didn't have on the IBMs, chances are the drive is terminating its unused lines properly anyway -- like I've said, it shouldn't really matter, because they're unused!

So:  "If it ain't broke, don't fix it."  --  But it won't hurt to keep a printout of this thread around, so you know what to try if you ever have trouble.