From rjn@hpfcrjn Sat Jun 10 20:28 MDT 1989 Received: from hpfcrjn.HP.COM by hpfclw.HP.COM; Sat, 10 Jun 89 20:28:00 mdt Received: by hpfcrjn.HP.COM; Sat, 10 Jun 89 20:23:39 mdt Date: Sat, 10 Jun 89 20:23:39 mdt From: Bob Niland Full-Name: Bob Niland Message-Id: <8906110223.AA00650@hpfcrjn.HP.COM> To: Chris_Van-Woerkom@hpfcrjn, Dave_Kinsell@hpfcrjn, Don_Boxley@hpfcrjn, Doug_Johnson@hpfcrjn, Gil_Sandberg@hpfcrjn, Jeff_McNair@hpfcrjn, Jerry_Nelson@hpfcrjn, Jim_Brokish@hpfcrjn, Jim_Geer@hpfcrjn, John_Sontag@hpfcrjn, Mark_Edwards@hpfcrjn, Mike_Schubert@hpfcrjn, Paul_Febvre@hpfcrjn, Paul_Perlmutter@hpfcrjn, Pete_Lord@hpfcrjn, Rob_Harris@hpfcrjn, Robert_Quist@hpfcrjn, Ron_Rogers@hpfcrjn, Steve_Johnson@hpfcrjn, Tony_Pilarinos@hpfcrjn Subject: Apollo/HP ctg. tape issue Status: R From: Bob_Niland To: Subject: 1/4-inch cartridge tape, Apollo vs HP; format backgrounder With the acquisition of Apollo, the merge of SPU product lines and associated mass storage, we bring into collision two conflicting 1/4-inch cartridge tape formats. And, the incompatibility goes beyond the inability to read each other's tapes. The purpose of this memo is to provide the background information required by those making decisions about cartridge tape drives on future WG products. Please forward it to interested parties working the merge issues. I will provide copies to Apollo's storage engineering team on June 12th. -------------------------------------- One minute summary: QIC ("Quarter Inch Committee") 60 and 120 Mbyte cartridge tape drives are available half-high embedded-SCSI versions. At first glance, this looks like an attractive device for a Merge'040 computer. Apollo currently uses QIC, and presumably we shall carry QIC forward for at least DOMAIN configurations. However, inserting a current HP cartridge in a QIC drive leads at the very least to frustration, and potentially to permanent damage to the HP tape. Chances of merging the tape formats, using existing media, are exceedingly low, and the implications for selling an "HP/Apollo" workstation with a QIC drive are serious. Attempting to use a QIC tape in an HP drive results in: tape rejected. Attempting to use an HP tape in a QIC drive results in: 1. No data exchanged. 2. An apparently damaged tape (rewind scenario). 3. A probably destroyed tape if a manual (reposition spill scenario). 4. A definitely destroyed tape (write scenario). There are also implementation considerations for supporting QIC drives under HP-UX. I have included some initial proposals at the close of this memo. -------------------------------------- Cartridge Tape Specification Overview: .====================.=========================.==========================. | Characteristic | HP Cartridge Tape(note) | Apollo (QIC) Ctg. Tape | |====================|=========================|==========================| | Mechanical form | 3M DC600 | 3M DC600 | | factor | | | | Merchandising name | | QIC-24 or QIC-120 | |--------------------|-------------------------|--------------------------| | Typical supplier | HP | any discounter | |--------------------|-------------------------|--------------------------| | Vendor designation | DC600HC, DC615HC, | DC600A, DC615A | | | DC600XTD, DC615XTD | | | Tape pre-format | Full-track 3M HCD-75 | | | | or modified HCD-134 | | | Data format | MFM | NRZI | |--------------------|-------------------------|--------------------------| | EOT/BOT | Delimited by pre-format | Optical tape sense holes | | Rewind position | Right spool empty. | Left spool empty. | |--------------------|-------------------------|--------------------------| | Capacity | 67 or 134 Mbytes | 60 or 120 Mbytes | | Number of tracks | 16 or 32 | 9 or 15 | |--------------------|-------------------------|--------------------------| | Access types | Random (re-writeable) | | | | or serial | Serial | | Error control | Read-while-write | Read-while-write only | | | plus XOR ECC. | | `=========================================================================' Note - By "HP" cartridge tapes, I refer to the 88140L/LC/S/SC media used in the 7908..14, 7942/46, 9144A and 35401A drives, plus the 92245L/S media used in the 9145A, but NOT the 92242L/S tapes for the 9142A "Thor" drive (which is yet another format). Issues: * The HP and QIC cartridges are mechanically identical. An HP cartridge will fit in a QIC drive, and vice-versa. * Because we (HP) have never merchandised our nearly unique format, there is nothing printed on HP, 3M DC6xxHC or QIC media, or HP or QIC drives, to suggest that interchange might be unwise. * We also have NO NOMENCLATURE to describe the situation. We never gave the HP/HCD format a name. As you can see from the footnote above, the existence of the 9142A PCT-format tapes means we can't even say "HP tapes" to describe the HP/HCD family. Even if we could, DMK will no doubt soon be selling "HP" QIC tapes to our Apollo installed base. The introduction of the 9145A finally forced us to print "16 track" and "32 track" on some [new] drives, but we have never published the abbreviation "HCD", nor identified our media as DC600HC and DC615HC. Some obscure data sheets do mention "MFM". * Most QIC customers apparently do not buy their tapes from their computer vendor. We therefore have NO opportunity to control what is printed on those tapes. + Most HP customers do buy their tapes from HP. We DO have an opportunity to control what is printed on them, but a significant change in legends might raise concerns in the installed base, particularly on the HP3000, where they may have no awareness of the existence of other standards. So what is the problem? * "HP" tapes are preformatted by 3M or a 3M licensee. A full-track factory write head lays down fixed physical records on the tape. No HP (or HCD) drive ever writes on these "keys", only in between them. Our "format" and/or "mediainit" processes merely perform read/write tests, spare bad logical sectors and update CS/80 logs on the tape. The read/write heads in the drives are either 1/16- or 1/32-track. We could not re-write keys in the field even if we wanted to. * These keys also do not extend to the physical BOT/EOT sense holes in the tape. Optical sensing was too unreliable in the late 70's, when the "Linus" drives were developed, so all HP drives use the keys to denote BOT/EOT. (QIC tapes are more like traditional 1/2-inch 9-track media. They have no pre-formatting. QIC drives can only rely on the sense holes for BOT/EOT.) * If the customer puts a QIC tape (blank or written) in an HP drive, the HP drive will search for the nearest key, fail to find one, time-out, buzz, release (unload) the tape and illuminate the FAULT indicator. No data lost, no positioning errors, but no data is read or written either. * If the customer puts an "HP" tape in a QIC drive, the drive will rewind it to the wrong end of the tape (by HP conventions) and OUTSIDE the keys region. If the tape is removed from the QIC drive and re-inserted in an HP drive, the HP drive, as in the preceding example, will search for a key, not find one in a reasonable time, and reject the tape for fear of spilling tape if the search continues. It will buzz, release and FAULT. The tape may be MANUALLY wound, in the COUNTER-INTUITIVE direction, to reposition it inside the keys. If re-inserted in the HP drive, it will properly load (after adding an extra minute to the load time because it was rewound to the wrong end). Now please note that in the construction of DC600 media, the tape is not affixed to the reels... * Given a mis-positioned cartridge, should the customer manually wind in the intuitive {i.e. incorrect} direction, the tape unspools. Disassembly and re-spooling of DC600 tapes is a nightmare for the inexperienced. We don't even publish instructions (although we used to for the earlier DC100A {HP 98200A} tapes). I have re-loaded cartridge tapes. Not fun. * If, on the other hand, the tape was left in the QIC drive, two more scenarios obtain. The QIC drive has no clue that this is an HCD tape. 1. The HP data structures just look like noise. So any attempt to read an HP tape simply fails or returns garbage data. If the tape is removed after a read attempt, the rewind scenario above prevails. 2. A QIC drive will happily WRITE on an HP tape. This destroys both existing data AND the HCD preformatting keys. If the preformatting is ever lost, the tape can NEVER be used in an HCD drive again. (This problem can also happen if the customer degausses the tape.) What happens back at the HP drive? Apart from the rewind/time-out problem, as soon as the HP drive detects a trashed key, it will FAULT and reject the tape. Only 3M can (re)format an HCD tape, and I doubt that they offer such a service. These scenarios are not hypothetical. This memo was derived from an article I post to Usenet periodically, everytime a real customer asks why their SUN workstation "ate" their HP tape. It happens. SUN uses QIC, too. To my knowledge, no one makes a QIC drive that can recognize and harmlessly reject HCD media, much less read or write both QIC and HCD formats. I doubt we could obtain and qualify a drive before 9.0. A dual-format drive would likely be expensive, defeating the main objective of having such a device in the first place. It may be possible to have 3M modify media made for HP such that it will not insert in a QIC drive, but this is unlikely and may increase the price of our already-expensive tapes to an unacceptable level. And, of course, it would not affect the millions of cartridges already in circulation. -------------------------------------- Assuming that we desire to support QIC under HP-UX (and we have been asked), there are further obstacles. We would require the following. This is summarized from my recent "Convex/Archive" big deal memo. * Serial SCSI read/write of QIC tapes. This has already been prototyped. Paul Perlmutter has a modified version of the HP-UX 6.5 SCSI driver that has been informally tested with the Convex drive. This provides rudimentary backup/restore, but without recovery capability, it is not a complete backup solution unless an 'install' device is also present in the configuration. If customers wish to integrate their own QIC drives, and assume the risks, we should not impede them. * Serial SCSI "recovery system" capability, if not full serial SCSI "install/update". This capability is needed to allow at least recovery of a completely crashed disk at a customer site. This functionality breaks down into several sub-components: - Serial SCSI boot. The current revision C boot ROM only knows how to boot from direct-access HP-IB and SCSI devices (i.e. disks and disk-emulating tapes like the 9144A). Serial SCSI boot has been requested for revision D, but could easily fall off the requirements list in light of the likely need to provide a merged HP/Apollo capability set for the Merge'040 computers next year. We are probably resource-limited on boot ROM enhancements, although it is worth pointing out that Apollo will need at least QIC boot on Merge'040, if not 8mm as well. - Serial SCSI secondary loader (boot extension). The boot ROM's responsibility ends after locating, reading and transferring control into the loaded binary code called the "secondary loader" (SL). The SL knows how to find /hp-ux on the boot media, read it in and transfer control (but only on direct-access media today). - Serial read-only recovery/install kernel. Our existing HP-UX kernel knows how to initialize itself from a read/write file system (i.e. conventional disk) and from a read-only file system (i.e. direct- access cartridge tape). It knows how to suppress swapping when read-only. HP-UX does NOT know how to initialize itself when the source devices is serial (not direct-access). This capability was requested for 8.0, to support DAT install/recovery, but has apparently fallen out because no software distribution is currently committed for DAT. The next generation HP-UX medium is CD-ROM, which is read-only direct-access, which we already support. - Supporting a full user-friendly "install" from a serial device. * Serial boot/recover/install definition. Series 300 not only supports no serial cold-load capability, it also has no definition for it; no media format, no record format, no control transfer protocol, no boot record validation, etc. This does exist on Series 800 for 9-track (HP-IB) tape boot/install/recover. Some effort would be require to invent or port a definition for Series 300 (as well as implementing it, of course). * To my knowledge, no resource estimates have been made for any of the above (except the boot ROM). -------------------------------------- Observations and Recommendations: * Cartridge tape is the least expensive low-end backup media available, and is likely to maintain this position for at least two more years. I don't believe we can avoid the above problems by simply designing cartridges out of future workstations. * Even if future workstations don't contain cartridge tape drives, we will be selling existing designs (and media for them) for some time to come, not to mention installed-base media sales. * Nomenclature: Regardless of what we do about drives, I suggest that we give the HP/HCD media format a name, names, or key off the track count, and print a warning on HP-supplied media. For example: "Use only in Hewlett-Packard 16- or 32-track drives." This also implies labelling our [Apollo] QIC drives as 9- or 15-track. * Educate the customers: The HP/HCD format pre-dates QIC by about five years and has distinct advantages in access and data reliability. I suggest that we add a section to the System Administrator manual describing the formats, the functional differences, and while we're at it, give the reasons for the tedious HP pre-load unload drive behaviour and outlining precautions. (We have never done any of this!). * Separate-but-equal tape drives: I suggest that future workstations including a QIC drive be merchandized ONLY for DOMAIN configurations, perhaps OSF configurations, but not HP-UX. * Converge on a new format: If we decide that a low-end stand-alone backup/recovery capability is needed, let us choose an entirely new mechanical tape format, such as DC2000 {e.g. Irwin}. There may be no interchange with older formats, but there is also NO EXPECTATION of it either, due to obvious mechanical differences. * Convert to QIC? We have supported incompatible formats before. The HP 9815/9915, 9825, 9835/45 and Series 80 all had mutually incompatible DC100 formats, much to the annoyance of customers. However, even then, the worst case scenario resulted in lost data, not destroyed cartridges. Regards, Hewlett-Packard Bob Niland 3404 East Harmony Road HP-UX mail: rjn%hpfcrjn@hplabs.HP.COM hpfcse!rjn Fort Collins HPDESK: BOB NILAND /HP4000/UX CO 80525-9599 Telnet 229-4014, AT&T (303) 229-4014 1UP4 atten: Bob Niland MS66 List: Don_Boxley Jim_Brokish Mark_Edwards (CPB) Paul_Febvre Jim_Geer Rob_Harris Doug_Johnson Steve_Johnson Dave_Kinsell Pete_Lord Jeff_McNair Jerry_Nelson Paul_Perlmutter Tony_Pilarinos Robert_Quist Ron_Rogers Gil_Sandberg Mike_Schubert John_Sontag Chris_Van-Woerkom EOF