vsh@uunet.uu.net (Steve Harris) (11/14/89)
Some time ago, I queried the net if anybody had experiences with "cheapo" SCSI disk drives. I received two kinds of responses: 1) No, but let me know what you learn. 2) We're using 350 MB disks from XYZ which sell for $2500. By "cheapo", I meant under (or close to) $1000. Which probably means about 80 MB capacity. Which probably also means targeted for the PC market. After shopping around, asking folks at work where they got their PC or Mac disks, we identified a vendor, Hard Drives International (HDI) of Tempe, AZ, as having very good prices, a reputation for quality, and a 30 day free return/replacement policy. However, they have no experience with Sun computers. This means, insofar as support goes, you're on your own! Here is a sample of the SCSI disks they have (price includes enclosure and power supply, capacity is in MB, times are in milliseconds): Capa- Seek Transfer Manufacturer Model city Time Rate Price ------------ ----- --- ---- ----- ----- MiniScribe 9380S 340 16 10 1895 MiniScribe 3180S 160 17 10 1275 Seagate 296N 80 28.0 10 599 Quantum Pro80S 80 19.0 10 799 Quantum Pro105S 105 19.0 10 899 Micropolis 1375 153 16.0 10 1295 Micropolis 1578 332 16 10 1795 (Note: these prices change, the Pro80S just went down $150!!) DISCLAIMER -- I have no association with these people other than as a satisfied customer. Their phone number is: 800-234-3475. You get a cheerful recording and wait for the next available sales rep. (usually not more than a minute). If you will be buying for your company, ask for Brian in Corporate Accounts -- he knows his business particularly well, and was most helpful (give him my name so he knows putting up with me was worth his while). We ordered the Quantum Pro80S and the MiniScribe 3180S. The Pro80S is the little brother to the Pro105S, which Sun uses in the SPARCstation and the 3/80, so I was confident it would work. I called MiniScribe and they said they were "pretty sure" their drive would work. To make a long story short, I never was able to format the MiniScribe, and returned it. We'll try the Micropolis 1375 next. I had all kinds of problems with the Quantum. But I finally succeeded in formatting the drive after the manual arrived and I found the actual number of blocks on the drive. The entry in the /etc/format.dat file is WRONG. Here is the entry I made that works. disk_type = "Quantum ProDrive 80S" \ : ctlr = MD21 : fmt_time = 1 \ : cache = 0x5c : trks_zone = 6 : atrks = 0 : asect = 1 \ : ncyl = 802 : acyl = 2 : pcyl = 804 : nhead = 6 : nsect = 34 \ : rpm = 3662 : bpt = 16896 # partition table for 11MB root, 25MB swap, and 45MB usr partition = "Quantum ProDrive 80S" \ : disk = "Quantum ProDrive 80S" : ctlr = MD21 \ : a = 0, 22644 : b = 111, 51204 : c = 0, 163608 : g = 362, 89760 FORMATTING THE DRIVE: After correcting the format.dat file, I used the format utility under SunOS 4.0.3 on a both a SPARCstation1 and a 3/50. On the SPARCstation, the drive at SCSI address 1 is sd1, on the 3/50 it is sd2 (see below). In the next couple paragraphs, I will refer to this as drive 1 (as on the SPARCstation), substitute "drive 2" if on a 3/50, 3/60, etc. Format goes out and looks for drives attached to the system. If it doesn't find sd1, either the addressing or the wiring is wrong. When you select drive 1, you may get messages to the effect that there are discrepancies between the size reported by the disk and the size reported by the label. Ignore these, you're going to reformat the disk anyway. You will also get a message that no defect list was found. Go into the defect sub-system and request the original defect list, the drive will report some number of defects to the Sun. When you exit the defects sub-menu, commit the new list. (It's my impression that the disk is smart enough to take care of its own defects, and that this is why the format does not find a defect list. I don't know if it matters whether or not you extract the original list from the disk, it doesn't show up if you format the disk again.) During the course of formatting the disk (which takes about 1 second!), format reports the following error message: Error for command 'format' Error Level: Fatal Block: 3 Sense Key: Illegal Request Warning: Using default mode select parameters. Warning: Drive format may not be correct. Block 3 (0/0/3), Fatal non-media error (illegal request) failed THIS IS A SPURIOUS ERROR MESSAGE, THE FORMAT DID WORK CORRECTLY! (Thanks to Donna at the Sun hotline for telling me to ignore this error message, and for being so helpful. As she noted, the real test is whether or not the label command succeeds.) Next issue the label command. It should work with no errors. Exit format and run newfs to build your filesystems. (more: see below) THE HDI SHOEBOX: To get the free-standing "shoebox" from HDI, you have to order the "Mac kit", but need to specify a generic SCSI interface. It was explained to me that the Mac SCSI interface is non-standard and requires a different EPROM than the Generic, or PC, interface. Later I spoke to Quantum who said this is not the case (at least for the ProDrives) -- I didn't fully understand what the difference was but you could use either style equally well. For other drive manufacturers this may not be true, so order the Mac kit with the generic SCSI interface. The box has two female "CHAMP"-style SCSI connectors on the back (like a Centronics printer connector). However, when you look inside the box you see that they do NOT form a horseshoe loop with the drive in the middle -- the ribbon just runs from the first connector to the second connector to the drive. This means you cannot daisy-chain these units! (You would have to modify the cable to make a horseshoe). Since I do not plan to daisy chain, I just connect to the lower CHAMP connector, the one at the end of the cable (I'm not a SCSI expert, but it's my understanding that you must avoid all "T" connections to the SCSI bus -- by connecting to the middle CHAMP, the little strand of ribbon cable to the lower connector would constitute a T and signal quality could be compromised -- SCSI experts please correct/amplify these remarks). CABLING TO THE SUN: HDI supplies a cable to connect the "shoebox" to your Mac. At one end is a 50 pin CHAMP connector, at the other is a 25 pin "D" connector (like RS-232). It's useless! You will have to buy or make your own cable -- I make mine with unshielded ribbon cable and 50-pin CHAMP and "D" IDC (Insulation Displacement) connectors. For connecting to the SPARCstation, I built a CHAMP to "edge-connector" ribbon cable, opened up the box and connected the external disk in place of the second disk. Some time ago, there was much discussion of how Sun screwed up the SCSI interface on their 3/60s -- pin 26, which is supposed to supply +5 volts for termination power was grounded. This meant that any other device which chose to supply +5 on that line was likely to melt-down its power supply, or worse, its firmware. Since, up till now, I have always connected to Sun shoeboxes, I have not been concerned with this problem. However, now that I am connecting to another vendor's shoebox, I must take precautions. As a simple, if ugly, fix, I just cut wire 26 of the ribbon cable at the Sun end, causing line to float. (Actually, I made a short piece of cable with line 26 cut which I can insert between the sun and the cable to the disk drives). Strangely, when I tested for continuity between pin 26 and ground on the 3/60, there was none. Still, better to be safe... BTW -- the Quantum drive does not have a jumper to select the source for termination power (a la CDC). According to their tech hotline, the hardware can detect +5 on pin 26 and uses that if it's there, otherwise it uses +5 from its local power input! DRIVE ADDRESSING: Note that the SPARCstation's disk addressing is peculiar -- logical disk 0 (sd0) is at SCSI address 3, and logical disk 1 (sd1) is at SCSI address 1. On the 3/50, SCSI address 1 maps to logical disk 2 (sd2). This is due to the assumption that there can be up to two drives on each SCSI controller (the SCSI standard allows up to 8 devices per controller); the SCSI controller at address 0 (my primary drive) could have disks 0 and 1, the controller at address 1 (my secondary drive) could have disks 2 and 3. In fact, with embedded SCSI, there is only one device at each SCSI controller address, but the Unix kernel is configured to allow for two. The HDI shoebox has a clever little address setting switch on the back of the box -- there are two little levers and a window displaying a digit, press the "+" lever and the digit increments, press the "-" lever and it decrements (you have to "flip" the lever to be able to press it). Unfortunately, the value displayed in the window does not match the value set on the three address jumpers of the drive. I checked for continuity and found that setting the switch to 1 does not jumper address A0. Furthermore, when you examine the wiring of the plug that mounts on the disks address jumper pins, you see that it can only jumper pins 0 and 1 -- there is only one wire running to the pair of connectors for jumper 2. (I assume this setup works okay for Mac's.) I just disconnected the plug from the disk and set the address manually (the three address jumpers are labeled: A0, A1, A2). You have to open the box up to do this, but once the plug has been removed you can access the address jumpers by removing a metal plate from the base of the shoebox. While you're removing the plug, remove the parity-enable jumper (labeled "PE" -- it's the middle of three, just above the address jumpers). Apparently the disk doesn't work correctly with the Sun if parity is enabled (this is at the recommendation of Quantum -- I noticed that the two Pro105S drives in our SPARCstation also has the PE jumper removed). (For what it's worth -- I just succeeded in formatting, labeling, and newfs-ing a Pro80 with parity enabled.) MANUALS: The drives do not come with owners manuals -- you can get one by calling the manufacturer's technical support line. HDI will give you the manufacturers' 800 numbers. Quantum's is 800-367-1984. Over the course of mucking with these disks, I have looked at a number of manuals. Most are fairly good, but usually you have to dig to find the information you need. The Quantum manual is exceptional, far superior to all the others I have seen. It is well organized, well written, readily comprehensible (insofar as a technical manual can be readily comprehensible), and professionally typeset (not a Xerox of an nroff printout on a dot-matrix printer!!). There are a large number of diagrams and tables which make the information much easier to understand. It has as good a summary of SCSI as any I have seen. The day the manual arrived, it took me about 30 seconds to find the information I needed to diagnose the source of my problems (the erroneous format.dat entry)! It's my SCSI reference manual now. CACHE: I did not attempt to format the drive using the standalone diag utility which comes with SunOS 3.x, but I bet it would work. However, it seems likely the on-board cache capability of the drive would not be enabled -- there is no provision for this capability in diag. From reading the disk manuals, it seems this capability requires a a bit be set in the SCSI "mode select" command -- it is NOT on by default. I assume that the "cache = 0x5c" parameter in the format.dat file results in the appropriate bit being set. === MILD FLAME === I do wish Sun would publish definitions of these parameters (cache, trks_zone, atrks, asect) so that we could take advantage of them. For example, the MiniScribe manual speaks in terms of alternate sectors per cylinder (not per track); how do I modify my format.dat entry to accommodate this capability? Why is there no entry in section 5 of the Command Reference Manual for format.dat??? === END FLAME === SUN HOTLINE: I'd like to take a moment to commend Sun on improving the hotline. Six months ago my request for assistance could have taken several days before I received a response, and it's unlikely the person at the other end would have been much help. They have come a long way from those dark days -- the response was within 2 hours, and Donna was most helpful in resolving this problem. I'm sure there are still problems, especially with more esoteric aspects of the operating system. Nevertheless, I was impressed with the improvement. FOLLOW-UP: Although my formatting took place under SunOS 4.0.3, our application currently runs under 3.5. I figured, no problem, I'll just load 3.5 (from my master tape -- I refuse to go through suninstall) while I still have the disk mounted under 4.0.3, halt, set the address jumper to 0, plug it in to a standalone 3/50, and reboot. No such luck. I got the message: Boot: sd(0,0,0) and nothing else. It just sat there. The EEPROM saw the disk, but was unable to find the boot program. Back to the drawing board. So, I set up a 3/50 with a cartridge tape drive and this disk on the SCSI bus. Booted from 3.5 tape, loaded diag, relabeled the disk, loaded the standalone copy, copied the miniunix to sd(0,0,1), and booted sd(0,0,1)vmunix -as. Standard stuff. Worked fine. Reloaded 3.5 from my master tape, sync'ed the disk, and rebooted. No problem. Clearly, something under 4.0.3 is incompatible with 3.5. My guess is newfs, but how the boot EPROM loads the boot program from the root file system is a mystery to me. Any ideas? -- Steve Harris -- Eaton Corp. -- Beverly, MA -- uunet!etnibsd!vsh