From rjn@hpfcrjn Tue Apr 18 23:53 MDT 1989 Received: from hpfcrjn.HP.COM by hpfclw.HP.COM; Tue, 18 Apr 89 23:53:09 mdt Received: by hpfcrjn.HP.COM; Tue, 18 Apr 89 23:47:46 mdt Date: Tue, 18 Apr 89 23:47:46 mdt From: Bob Niland Full-Name: Bob Niland Message-Id: <8904190547.AA09142@hpfcrjn.HP.COM> Subject: Next generation backup Status: R From: Bob_Niland To: Chris_Van-Woerkom Doug_Johnson Tony_Pilarinos cc: Subject: Workstation backup (caution - contains HP ROMAN8 characters) It is currently perceived that HP 9000 workstations are at a competitive disadvantage with respect to backup. It is true that we have not had strategic plan for backup. This report: * summarizes the actual situation; * develops generic requirements for backup devices; * evaluates alternatives, and; * proposes that we: - proceed with support of DAT and MO for mid-range/hi-end backup - investigate DC2000 at the entry level - monitor developments in writeable CD This report examines mass-storage devices solely on a backup basis; however, backup devices and media are closely tied to software distribution (installation and update). This report was prepared in parallel with a separate report on "Next Generation Software Distribution". Request a copy if you did not receive both. The present report neither presumes nor excludes the possibility that the system backup device may also be used for software distribution. This report also focuses on HP-UX. The current lack of SCSI on BASIC/WS and Pascal/WS does not encourage alternative backup strategies for those systems. THE CURRENT SITUATION (HARDWARE): HP-UX 6.2 was the last version for which flexible disk 'install' media were available. HP-UX 6.5 is likely to be the last release for floppy 'update' media. With the deletion of floppy, the price of a functional (if impractical) entry-level stand-alone system increased by the difference in price between the 9122C#001 floppy drive ($1150.00 list) and the current entry device, the 9144A cartridge tape drive ($2600.00 list). In this age of "$10,000 workstations", an apparent increase of $1450.00 is causing some concern. Although the inability to fit the HP-UX install kernel in 780 Kbytes was the pivotal event, it has long been acknowledged that flexible disks are inadequate for practical backup. Sun only distributes software on floppy for the 386i. The new SPARCstation offers a built-in 3ø-inch floppy drive, but it is promoted as a DOS/interchange device, not backup, and apparently not software distribution. DEC distributes primarily on tape, has recently added CD-ROM, but does not promote floppy as a backup device. Apollo still distributes on 5÷-inch floppy (albeit 1.2 Mbyte). Two years ago an advertised system price would have included the software installation and backup device. Now that all major workstation vendors support a diskless capability, the install/backup device is becoming an extra-cost option. And quite often the device requires an extra-cost interface or adaptor to support it. Here are some representative prices for entry-level tape backup devices, (including interfaces): _____________________________________________________________________ | Vendor | Model | Capacity | I/F | List price | Comments | |ööööööööööööööööööööööööööööööööööööööööööööööööööööööööööööööööööööö| | Apollo | | 60 Mbytes | | $2000 | DN-3K,4K,10K | |öööööööö|ööööööö|öööööööööööö|öööööö|öööööööööööööö|ööööööööööööööööö| | DEC | PC4XT | 150 Mbytes | SCSI | $1820 | DECstation only | | DEC | TZ30 | __ Mbytes | SCSI | $4000 | | | DEC | TK50 | 95 Mbytes | | $3087..5200 | | | DEC | TK70 | 296 Mbytes | | $7600 | | |öööööööö|ööööööö|öööööööööööö|öööööö|öööööööööööööö|ööööööööööööööööö| | SGI | | 150 Mbytes | SCSI | $2500 | | | SGI | | 60 Mbytes | | $1500 | | |öööööööö|ööööööö|öööööööööööö|öööööö|öööööööööööööö|ööööööööööööööööö| | SUN | T650G | 60 Mbytes | | $2000 w/disk | | | | | | | $3000 alone | | °°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°° By comparison, the HP (WG) entry-level products are: _____________________________________________________________________ | Vendor | Model | Capacity | I/F | List price | Comments | |ööööööööööööööööööööööööööööööööööööööööööööööööööööööööööööööööööööö| | HP | 9144A | 67 Mbytes | HPIB | $2600 | 33 Kb/sec. | | HP | 9145A | 134 Mbytes | HPIB | $4055 | 33 Kb/sec. | °°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°° For interchange and in the mid-range we have the following devices. _____________________________________________________________________ | Vendor | Model | Capacity | I/F | List price | Comments | |ööööööööööööööööööööööööööööööööööööööööööööööööööööööööööööööööööööö| | HP |C1701A | 325 Mb x 2 | SCSI | $6435 | 300 Kb/sec. | |öööööööö|ööööööö|öööööööööööö|öööööö|öööööööööööööö|ööööööööööööööööö| | HP |35401A | 67 Mb x 8 | HPIB | $8150 | 33 Kb/sec. | |öööööööö|ööööööö|öööööööööööö|öööööö|öööööööööööööö|ööööööööööööööööö| | HP | 7979A | 45 Mbytes | HPIB | $13,400 | 160 Kb/sec. | |öööööööö|ööööööö|öööööööööööö|öööööö|öööööööööööööö|ööööööööööööööööö| | HP | 7980A | 170 Mbytes | HPIB | $23,200 | 600 Kb/sec. | °°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°° We do not appear to have any competitive problems here. The 35401A "Merlin" is unique in the industry. Those desiring 797x 9-track drives usually have an interchange requirement, and price is less of a consideration (also, several models of older remarketed drives are available at attractive prices). The C1701A "Shark" magneto-optic disk drive is scheduled for June CPL. At the high end we have the following devices. _____________________________________________________________________ | Vendor | Model | Capacity | I/F | List price | Comments | |ööööööööööööööööööööööööööööööööööööööööööööööööööööööööööööööööööööö| | HP |7980XC | 600*Mbytes | HPIB | $32,200 | 1.2 Mb/sec. | |öööööööö|ööööööö|öööööööööööö|öööööö|öööööööööööööö|ööööööööööööööööö| | HP |ûDATý | 1.2 Gbytes | SCSI | $5000 | 180 Kb/sec. | |öööööööö|ööööööö|öööööööööööö|öööööö|öööööööööööööö|ööööööööööööööööö| | various|Exabyte| 2.0 Gbytes | SCSI | $4...6000 | 256 Kb/sec. | |öööööööö|ööööööö|öööööööööööö|öööööö|öööööööööööööö|ööööööööööööööööö| | HP |C1700A | 20 Gbytes | SCSI | $29,900 | 300 Kb/sec. | °°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°° The 7980XC, with on-board data compression, is unique in the industry. Meanwhile, several hardware vendors are selling Exabyte drives (with HP-UX driver) to our customers. Enterprising HP SR's have also brought in Exabyte resellers on key deals. (Exabyte is 8mm helical scan cassette.) The C1700A "Jaws" magneto-optic autochanger will be on the July CPL. Although it provides a capability unique among workstation vendors (DEC's device is write-once, our is write-many), the media cost will likely prevent it from being widely seen as a strictly backup device. The HP DAT product (4mm helical scan tape cassette) will not be available until calendar 1990. Although it has several technical advantages over the current 8mm devices, DAT may not completely displace 8mm. Apart from having an 18 month head start, 8mm has slightly higher performance and doubled capacities are rumored for next year. Corvallis is using 8mm for site-wide backup. A copy of their report is available. HP customers are therefore using a mixture of: * ÷-inch single-play drives (9144A/45A) on stand-alone and server systems. * ÷-inch autochangers (35401A) on stand-alone and server systems. * Home-brew remote backup of networked non-diskless nodes, to both HP and non-HP (NFS) hosts. * Floppy backup, only on pre-6.5 systems (theoretically). * 3rd-party 8mm helical scan drives on stand-alone and server systems. * ø-inch single- and multi-reel backup on stand-alone and server systems. * Miscellaneous 3rd-party and arcane solutions (e.g. 7935H disk packs). THE CURRENT SITUATION (MEDIA): The following table positions our existing and imminent media (hp) against the competing and possible future media. ________________________________________________________________ | | | Unit | Per/ | | Media type | Capacity | Price | Mbyte | |öööööööööööööööööööööööööööööööö|öööööööööööööö|öööööööö|ööööööö| | 8mm cassette tape | 2000 Mbytes |$ 10.00 |$0.005 | | 4mm DAT tape (hp)| 1200 Mbytes | 10.00 | 0.008 | |öööööööööööööööööööööööööööööööö|öööööööööööööö|öööööööö|ööööööö| | Writeable CD | 600 Mbytes | 10.00 | 0.017 | est |öööööööööööööööööööööööööööööööö|öööööööööööööö|öööööööö|ööööööö| | ø-inch 9-track tape (hp)| 600 Mbytes | 20.00 | 0.033 | | 3480, HI/TC ø-inch cartridge | 300 Mbytes | 50.00 | 0.167 | est |öööööööööööööööööööööööööööööööö|öööööööööööööö|öööööööö|ööööööö| | ÷-inch "120 Mb" cartridge (hp)| 134 Mbytes | 39.80 | 0.297 | |öööööööööööööööööööööööööööööööö|öööööööööööööö|öööööööö|ööööööö| | 5÷-inch magneto-optic disk (hp)| 2x325 Mbytes | 249.00 | 0.383 | |öööööööööööööööööööööööööööööööö|öööööööööööööö|öööööööö|ööööööö| | ÷-inch "60 Mb" cartridge (hp)| 67 Mbytes | 35.60 | 0.531 | | ...in a 35401A magazine (hp)| 8x67 Mbytes | 329.80 | 0.615 | |öööööööööööööööööööööööööööööööö|öööööööööööööö|öööööööö|ööööööö| | DC2000 ÷-inch cartridge | 40 Mbytes | 30.00 | 0.750 | |öööööööööööööööööööööööööööööööö|öööööööööööööö|öööööööö|ööööööö| | 3ø-inch flexible disk | 20 Mbytes | 10.00 | 2.000 | est | 5÷-inch flexible disk (hp)| 1.2 Mbytes | 3.00 | 2.500 | | 3ø-inch flexible disk (hp)| 1.6 Mbytes | 6.30 | 3.938 | °°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°° The interesting thing to note here the Per/Mbyte price spread ratio of nearly 800:1. This implies that PC/workstation customers are not particularly sensitive to the price of media. We should expect this to change as workstations become more routine in major accounts, and purchasing agents start to notice the relatively high price of ÷-inch cartridges, and they get accustomed to the $10 price of 4mm and 8mm cassettes. ISSUES AND CONSIDERATIONS: * With the deletion of floppy, we no longer have have an entry-level backup device. Our least expensive device, the 9144A, has a list price of $2600.00. Because it is an external drive, we can't bury (or reduce) the price by building it in as our competitors do. * Is entry-level backup a problem? Can it be solved? Is it a problem worth solving, or would the resources be more profitably be applied elsewhere? * No one outside of PCG is studying the low-end backup issue. No peripheral division (with the possible exception of Guadalajara) acknowledges having the charter for such an activity. CPB (Bristol) is investing all their resources in getting DAT to market. They have none to spare on further cost-reducing the 9144A, or even repackaging it for Dis'N'Dat. * With the continuing SQRT(2)-per/year growth in the size of HP-UX, the target "entry level" system size keeps increasing. 1989 may be the last year that the minimum HP-UX system fits on one 67 Mbyte tape. * As we move toward a focus on larger accounts, networked systems and supported network backup, the need for local backup on small stand- alone systems may diminish. Early reports on the evolution of HP workstation topologies don't support this, however. The stand-alone requirement will be with us for some time to come. * As we sell increasingly larger servers and stand-alone systems, our maximum backup capability is stretched. Today, customers with systems in excess of 600 Mbytes are attaching 3rd-party Exabyte tape drives because we have no similar product [yet]. * Our backup plans need to complement our installation media plans. A separate report is available on "Next Generation Distribution Media". GENERAL REQUIREMENTS FOR A BACKUP DEVICE: Generally speaking, ease-of-use, media cost, and total time required must be such that the device will be actually purchased, and the indifferent or even slightly irresponsible part-time user/administrator will not be motivated to neglect the periodic backup task. The average administrator of an HP-UX system is also a user (perhaps the only user) and performs admin on a part-time basis. Incidently, your humble narrator is such a part-time admin (on two systems). Very few workstations have full-time administrators. If our backup strategy assumes a fully trained professional admin/operator, and we tell that to customers, we will not succeed in the workstation market. * Effortless access to the device - If the admin must walk two aisles away to borrow a tape drive (that may be in use, or already borrowed by someone else), or needs to stand around for 2 minutes unloading the tape someone left in it, this person will quickly find excuses to "forget" the task next time. Even on a permanently connected 9144A or 9145A, the 3 minute period required for the unload/reload cycle is long enough that the weekly archive backup ('n' tapes) is effectively 2n separate tasks. When I was in this situation, ordinary interruptions and distractions would often prevent the job from getting done in one 10-hour day. * The 8-hour rule - Backup of the host/server file system must take less than 8 hours (1 shift), including verification. For purposes of calculation, file system capacities range from 10 Mbytes (BASIC/Pascal), thru 80 Mbytes (entry-level HP-UX) up to 7 Gbytes (the practical maximum on Series 300, given current backup utility performance). However, this does not necessarily dictate a minimum backup performance in Mb/min, as different drives can be used for different configurations. Even on an 80 Mb system, 3 Kb/sec (1/10 of current device speeds) would meet the 8-hour rule. A 7 Gbyte system, on the other hand, would require 250 Kb/sec, which can be achieved on a Model 370. * Data integrity - Simple read-while-write verification is the minimum. Either the backup device or host software should provide some level of error correction. This is particularly important for the next five years. Host data compression, a relatively immature technique, is being used extensively, and the current Lempel-Ziv algoritm has NO capability to recover from data errors during 'uncompress'. A single bit error early in the first tape of a single- or multi-tape backup can render the entire balance of the archive unuseable. * Shelf life - Media must have a re-read life of at least 10 years. An HP Class B environmental range for the media is also presumed. * Interchange - Media must be portable between systems (at least on the same drive model). There is a fair probability that if an old archive is ever needed, it will not be read on the drive that wrote it. Drive-to-drive interchange between different vendors of the same type mechanism (e.g. 3M "Linus" and HP "Buffalo") is essential. This is important if we are to consider any new small-format tape drives. QIC drives, in particular, have a poor reputation for interchange. * A recovery procedure must exist for a totally lost file system. That is, a massive hardware failure (crash) of the root (or only) disk on a system will require that the replacement disk be initialized, have at least a minimal HP-UX system installed (a "recovery system"), and have the current archive restored. If the backup device is not bootable, or is bootable but is not "rootable" or has no recovery system capability, then a separate means must exist for bringing up a recovery system. This is mentioned because a CD-ROM distribution scheme, for example, may require a COMPLETE re-install, and then application of the backup. In this and similar cases, provision of a separate recovery system on the install media is indicated. * Recovery time less than 8 hours - Restoration of one or all files must take less than one shift, since this is an anticipated event. Recovery of a complete lost file system may take longer (for mediainit and newfs/mkfs via the recovery system), since this is an extraordinary event. * Single-file recovery - is mandatory. This is mentioned because some otherwise attractive schemes (e.g. compressed raw disk image) don't permit it unless a spare disk drive is available for complete image restoration. The problem here is that this type of backup can only be restored by overwriting your entire existing file system, destroying files added or changed since the archive was written. * Drive Cost: Although it is possible to develop value propositions in a vacuum, the realities of the backup market are: o Backup is like insurance and life preservers. Everyone recognizes that they need them, but they prefer to ignore them. The less visible the price (any price) of the backup device, the less annoyed the customer is. Imagine the annoyance if seat belts were an extra-cost (but mandatory) "option" on automobiles... o The price point spread for each type of device is firmly set by the marketplace and the competition. It is difficult to convincingly sell a pure backup device at a higher price based on features and value-added. This is partly because it calls customer attention to something they would prefer to ignore. o It is possible to value-price devices that have more than backup functionality (e.g. industry standard interchange); have leading-edge performance (e.g. 7980XC), or create a new area of functionality (e.g. "Jaws"). * Media price: 2 sets of archive media should be less than 4% of the purchase price of the disk drives they back up. Unit cost of media should be less than $250.00. At least one value proposition may be of use here: Consider "backup" to be a "service contract for data". Media price should be less than the disk(s) drive(s) hardware service contract on an annual basis. For return-to-HP service, this is typically 4% of purchase price per year. The minimum media usage model is: 2 complete archive sets in rotational use at any time; 1 delta set in use at any time; 1 archive set retired every six months. This is a start-up expense of 3 sets, plus 2 sets/year. Current ÷-inch tape, at 65/Mbyte, meets this 4% test. Since backup media are infrequently filled to maximum capacity, unit price is also a concern. DAT and 8mm are setting a price point below about $15 per tape. ÷-inch $40. Magneto-optic is $250.00. In addition, the media PRODUCT price must be at or below department signature authority. Currently, we are seeing no unit-price resistance at $219.00 for 88140LC (five 16-track ÷-inch) tapes. Exceeding $250.00 would risk encountering some sensitivity. * Fast mediainit (or none). If we required a full one-hour re-certify pass on our cartridge tapes before use, we would have a more serious customer satisfaction problem than we already have. Any future virgin media must be ready for use in a few minutes at most. * Write protect - There must be a way to manually prevent writing to the media. A sliding or rotating tab is ideal. This method must not rely on the application of an external tool or device (e.g. write-protect tab). Such appliances are easily lost, in which case the lax user will leave the medium enabled. Preferrably, write-protection will not employ the removal of part of the medium either (e.g. 9-track write ring). These are also easily lost, encouraging user to defeat the write-detect mechanism in the drive (I used to do this routinely on 9-track drives). * Convenient to label, re-label and store. * Media suicide prevention - New devices which physically accept our existing media MUST NOT be capable of destroying that media. This is mentioned because adoption of a full-size ÷-inch QIC drive would place us at precisely this risk. FSD chose not to the support the HP 9142A "PCT" drive for similar reasons. We have never adequately explained the "pre-formatting" used on our current ÷-inch cartridges, and most users are not aware that writing on them (or deguassing them) renders them permanently useless for future use on HP drives. Only 3M can re-format the full-track "keys". The drive read/write heads are not full-track. * Removeable media - This is an obvious requirement listed only for completeness. "WANT" CHARACTERISTICS OF A BACKUP DEVICE/MEDIUM: * Unattended - Operator-less backup in less than 8 hours is HIGHLY desired, and may be a must. By way of example, our late floppy-based systems do not meet this test. Virtually no one puts up with the hassle of playing "floppy frisbee" for several hours, not to mention the incredible expense on a dollars-per-Mbyte basis. This is also true for PCs. At the Fort Collins site, the only Vectras that are routinely backed up are those on an OfficeShare or MSNet server. Those which have access to a departmental loaner cartridge tape drive get backed up irregularly. Those which have only floppy are almost never completely backed up. Even systems with dedicated 9144A "Buffalo" drives are not routinely backed up if the system disk exceeds about 80 Mbytes. This requires tape swapping, which frequently does not get done in the real world of part-time workstation admins. * Single-medium per system or single-medium per drive is desired. Although an auto-changer meets the "unattended" requirement, a large capacity media is more desireable, since it reduces drive cost and media management. * Non-serial restore. A backup may require multiple media, but it should be possible to recover a single file known to be on medium 'n' without requiring reading all media 1...n-1. By way of example, the current 'cpio' utility allows media skipping with the re-sync option. On the other hand, the use of host software data compression with any backup utility will cause recovery failure unless all the media are read, and in the correct order. * Use of the backup media as a distribution media is desired. * Industry standards are desired. Conformance of the media and at least one logical format to a formal standard (e.g. ANSI) or a defacto standard (e.g. IBM) is an advantage. * Re-useability - If the media is write-once, this makes it more difficult to meet the cost/Mbyte/year requirements. However, write-once is not automatically disqualifying. * Reading of existing media is highly desired. This applies to new ÷-inch tape formats and new flexible disc formats (if any). * Self-labelling. THE ALTERNATIVES: The following table summarizes the apparent choices available. A detailed narrative on each medium follows the recommendations after the table. __________________________________________________________________ | Media | Drive | Media | Data |Capacity|Transfer| Seek |Preload| | Type | List$K | List$ | Modes | Mbytes |Rate/sec| Time | Time | |öööööööö|öööööööö|ööööööö|ööööööö|öööööööö|öööööööö|öööööö|ööööööö| | DAT/4mm| 5-7 | 10 | R/W | 1300 | 180K | 18s | ____ | | 8mm | 4-6 | 10 | R/W | 2000 | 1.2M | 280s | None | |öööööööö|öööööööö|ööööööö|ööööööö|öööööööö|öööööööö|öööööö|ööööööö| | ÷"(HP) | 4 | 48 | R/W | 134 | 66K | 60s | 120s | | ÷" QIC | 1.2-3 | 25 | R/W | 120 | 33K | 60s | None | | DC2000 | .9 | 30 | R/W | 40-200 | 50K | __s | ?___? | |öööööööö|öööööööö|ööööööö|ööööööö|öööööööö|öööööööö|öööööö|ööööööö| | MO | 6.5 | 250 | R/W | 330x2 | 300K | 100ms| None | | WORM | ___ | 170 | R/W | 300 | 300K | 100ms| None | | WriteCD| ? | 5 | RO* | 800 | 150K | 500ms| 4s | |öööööööö|öööööööö|ööööööö|ööööööö|öööööööö|öööööööö|öööööö|ööööööö| | Floppy | .5-1 | 8 | R/W | 1-4 | 35K | 85ms| None | | | ? | ? | R/W | 20 | - | - | - | |öööööööö|öööööööö|ööööööö|ööööööö|öööööööö|öööööööö|öööööö|ööööööö| | IBM3480| 6-50 | 15 | R/W | 200 | 3M | __s | None | | HI/TC | 3-6 | 25 | R/W | 300 | __ | __s | None | °°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°° TECHNOLOGY NARRATIVE: °°°°°°°°°°°°°°°°°°°° DAT (Digital Audio Tape) - Drives will be available in 5÷-inch full-height form factor. The 1.2 Gbyte medium is a cassette about 3/8-inch thick and having the profile of a business card. It is essentially a miniature VHS cassette. DAT support for backup is already committed for 8.0. Cost reductions and 3ø-inch form factor are planned for later versions. DAT may be a software distribution medium. Advantages: * Higher single-medium capacity than anything from HP today or planned. * Lower media price/megabyte than anything from HP today or planned. * Transfer rate matches today's backup utility performance. * Sony/HP "DDS" format will be an industry standard. * DDS incorporates both read-while-write and ECC. * Single-file recovery could be faster than 8mm. * High-speed tape duplication technology is available. Disadvantages: * Initial SCSI drive price (in "Dis'N'Dat") of about $5000 is at parity with 8mm, but too high to be "entry level". * 8mm stores more data, and has higher performance as long as the data rate sustains streaming, the tape has few defects, and few file marks are written. * The failure of DAT in the audio market could prevent economies of scale from reducing the drive cost. °°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°° 8mm - Drives are currently available in 5÷-inch full-height form factor. The 2.0 Gbyte medium is about the size of a pack of playing cards. It uses the same technology and media as 8mm video. "Hi-band" 8mm (Hi8) is now announced, and this is presumably providing the promised 2x capacity and performance increase. The drives are made by Sony. We currently reference 8mm as needed. I suggest we continue to do so (but no more than that). Advantages: * Highest capacity single-medium technology extant. * Slightly exceeds our backup performance, and will continue to in the 2x version. Disadvantages: * HP is committed to DAT. * Defacto, but not formal format standard. However, being state-of- -the-art generally relieves that obligation. * Sony may "discourage" 8mm data use once DAT is available (the 8mm and DAT are from unrelated Sony divisions). * No ECC, wastes tape during data starvation, bad media spots and writing file marks. Loss of 10-to-20% of space is common. °°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°° ÷-INCH CARTRIDGES (DC-600 form factor): HP - The HP9144A (16-track, 67 Mbyte) and 9145A (32-track, 134 Mbyte) are probably the last full-size ÷-inch drives that HP will build. Only the near term cancellation of DAT could change that, and there is no reason to suppose that this might happen. The capacity of the 9145A will go to 200 Mbytes this year with the introduction of 900-foot media. We should assume that this increase will appear on QIC drives as well. Non-HP - Any number of vendors currently offer 60 Mbyte-class and 120 Mbyte- class drives using comparable media. Capacities of up to 300 Mbytes are imminent. Embedded SCSI is common and the QIC standards are the only ones worth considering. The 3M HCD-134 (SCSI) drive would not be interchangeable with the 9145A. QIC Advantages: * Embedded SCSI (making "Dis'N'Dat" integration simpler). * 5÷-inch half-height available. * Industry standard format. * Capacities greater than 120 Mbytes. QIC Disadvantages: * Mechanisms only slightly cheaper than HP. * Error control only read-while-write (no ECC). * Poor reputation for drive-to-drive (much less vendor-to-vendor) interchange. * And the biggie: QIC uses the same mechanical cartridge as HP. HP cartridges employ full-track-preformatting and 16 or 32 data tracks, with random block re-write capability. QIC is blank (unformatted), employs 9, 15, 18 or 24 data tracks and is not randomly re-writeable. Writing on an HP preformatted tape on a QIC drive DESTROYS the tape for further HP-drive use, permanently. There is also no HP(HCD)-to/from-QIC interchange, with the customer dissatisfaction therein attending. I do not believe we could introduce a QIC (or HCD-134) drive without massive customer dissatisfaction problems. Our recent acquisition of Apollo (who use QIC) does not change this. To my knowledge, no one makes a drive that can handle both QIC and HCD, and that would be the only solution to the compatibility problem (and would not be a competitive drive in other respects). °°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°° DC2000 - This is the old DC100A mini-cartridge, loaded with ÷-inch rather than 1/8th-inch tape. Although these drives are very inexpensive, at half the factory cost of standard DC600 ÷-inch drives, the comparison to the 9144A must include the software installation device (e.g. CD-ROM) unless we are prepared to deliver software on DC2000. Advantages: * Drives are cheaper than current ÷-inch ctg. * 3ø-inch half-high is standard. * Embedded SCSI available. * One drive (40 Mbyte Irwin) already qualified on Vectra, with the 80 Mbyte in progress. Disadvantages: * Capacities are lagging conventional ÷-inch. * More expensive per Mbyte than our current products. * Drive-to-drive interchange over full HP Class B range unknown. * No read-after-write, Reed-Solomon ECC only. °°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°° MO - (Magneto-Optic) is a re-writeable cartridge disk technology. The medium is an ANSI-standard 5÷-inch cartridge which looks rather like the Sony 3ø-inch flexible disk, shutter and all. The media must be flipped over to use both sides; there are no twin-head drives yet. The stand-alone and auto-changer HP products, although not on CPL yet, are fully supported by HP-UX 6.5. Advantages: * Random access - makes disc-image backups practical. * Capacity and performance are more than adequate for the near term. * Excellent shelf-life and environmental stability. Disadvantages: * High drive and media price. * Not mass-producable, but the fast load-time (7 sec) and fast real-time write speed (10x tape) are attractive. °°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°° WORM - (Write Once - Read Many) is writeable optical technology, conceptually similar to punch cards and tape. Both drive and media are similar in price to MO. There are significant problems for some backup utilities in supporting WORM. No WORM development is currently under way in HP, although future MO mechs may support R/W, WORM and R/O ANSI media. °°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°° CD-ROM - (Compact Disc) is physically the same as the now-common home audio medium. Many ROM players can accept either ROM (Data) or audio discs, but none are shipping yet with the proposed CD-ROM/XA interleaved data/audio format. CD-ROM is R/O (read-only) and is therefore not a backup medium. However, at least two re-writeable CD technologies have been announced (phase change and dye-polymer). Both are claimed to produce a CD that is compatible with existing CD (audio, and presumably ROM) players. Unit availability is slated for 1990. Media pricing is under $10.00. Drive pricing is unknown. WCD-ROM has the potential to be a highly desireable entry-level backup (and distribution) medium. It bears watching, but today must be considered vaporware. °°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°° Flexible Disk - We just abandoned (788 Kbyte) floppy as a distribution medium. We could have continued to support the 1.6 Mbyte "HD" media, but would probably have merely postponed the problem for a year (and we would still have had to abandon the 788 Kbyte installed base). WG has never had a 1.2M byte 5÷-inch drive to support. Floppy technology is still being augmented. 4 Mbyte media are available in the market and 20 Mbyte media are announced. However, at present, above 4 Mbytes we lose compatibility with existing formats. Since we are retaining floppy support for SoftPC and low-volume interchange, compatibility is a MUST requirement. The 4 Mbyte media are too small to be worth pursuing. °°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°° IBM 3480 - The medium is a single spool of ø-inch tape loaded into a 4x4x1-inch cartridge. The tape is auto-loaded by the drive, once the cartridge is manually inserted. The drives are absurdly expensive. The 3 Mbytes/sec data rate is over 10x what we can generate today. The relatively low 200 Mbyte capacity requires an operator. The 3480 is a total mis-match for workstations. HI/TC - Using a cartridge nearly identical to the above, HI/TC was intended to be the "poor man's 3480". It has evidently failed and is no longer a factor in the market. °°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°° Conclusion: Prerequisite question: Do we need a low-end stand-alone capability? If the answer is "NO", we can probably proceed with DAT, MO and current ÷-inch products, and save the resources needed to develop an entry-level solution. If the answer is "YES", the most promising technology is DC2000. We may be able to build an 80 Mbyte backup device into an Apex or mini-Dis'N'Dat for under $400.00 factory cost. However, unless we distribute software on DC2000, we would also require an install device for a complete solution. The proposed next-generation (and entry-level) distribution device is CD-ROM. I suggest we investigate DC2000 technology, with a view towards delivering a solution including both CD-ROM and DC2000 for under $1800 list. If this price target cannot be met, we should consider distribution of HP-UX (AXE and PE only) on DC2000 tape. See the "Next Generation Software Distribution" memo for more details. The entry-level Install/Backup joint solution: = CD-ROM for install/update. Target price $800. = DC2000 for backup. Target price $1000. ================================================== = Total $1800 Please respond with your comments. I propose to circulate this report outside Fort Collins next week. Regards, Hewlett-Packard Bob Niland 3404 East Harmony Road HP-UX mail: rjn@hpfcrjn.HP.COM hpfcse!rjn Fort Collins HPDESK: BOB NILAND /HP4000/30 CO 80525-9599 Telnet 229-4014, AT&T (303) 229-4014 1UP4 atten: Bob Niland MS66