de5@STC06.CTD.ORNL.GOV (SILL D E) (07/18/90)
I recently asked: >Which is a better backup device/medium for a heterogeneous network >including Suns, DECs, and various other workstations and minis? > >How do they compare on: > > -price of drives > -cost of tapes > -capacity > -reliability of drives > -reliability of recorded tapes > -speed of dumps and restores > -vendor and third party support > -driver availability > -ease of use > >DEC seems to be pushing DAT's, and argues that they're inherently >better since they were designed to record digital data, whereas 8mm is >an analog video format. > >Thanks in advance. Summary if warranted. A summary seems to be warranted, as this is something of a hot topic. 1. Price of drives "The 4mm seems to cost substantially more (~1.5x) [than] the 8mm for initial hardware." --Don Rice "The price was 1/2 of a 8mm unit." --J. Eric Townsend "We are being quoted prices that buy two WangDATs for the price of one Exabyte." --Sakari Jalovaara Conclusion: Neither is clearly cheaper, shop around. 2. Cost of tapes "In bulk, from a music/audio supply store, about $12 apiece for name-brand [DAT] blanks." --J. Eric Townsend "DAT tapes cost $15.00 and seem to be going up in price." --Jim "Tapes appear to cost about the same (+/- 20%.)" --Sakari Jalovaara Conclusion: Neither is clearly cheaper, per tape, but remember to consider the difference in capacities (see next item). 3. Capacity "Currently, 4mm stores half as much as 8mm per tape." --Don Rice "Now Exabyte 2 GB DAT 1.3GB Future Exabyte 5 GB (EXB 8500) DAT 8 GB (achievable HP)" --Matthias Kandler quoting "Battle of the Backup", May '90 _Systems International_ "Supposedly a double speed and capacity 8mm tape is also right around the corner." --Bill Nowicki "The 8mm tape has a 2.2 GB capacity versus 1.3 GB for the 4mm. Exabyte is coming up later this year with a 5 GB capacity 8mm drive." --Eduardo Krell "8mm beats the DAT in storage capacity (something like 2GB vs 1.3GB with max available tape lengths.)" --Sakari Jalovaara Conclusion: The 8mm technology clearly holds the lead today. 4mm may eventually catch-up or surpass, or it may not. 4. Reliability of drives "Exabyte hardware tends to break a fair amount, but at least they tell you there is a problem. From the responses: `I've had several of them break: doors jamming, unable to read/write (they give you errors, don't worry), the little green light burning out.... I've only had about 10 tapes fail so far in the past 18 months [out of roughly 900]. Most of these tapes got stuck in the tape drives because the drives are so cheap.' `(Our Exabyte) drives have a hard time living on the SCSI bus with other peripherals. They seem to hang sometimes, forcing a reboot.'" --Nathan Wilson "Nobody's saying (knows?) anything about the reliability of DAT hardware." --Nathan Wilson "We have had a Wangdat DAT purchased from Apunix for three months with few problems. Once it refused to give back a tape, but holding down the eject button for 10 seconds got it out (undocumented feature)." --Jim "We have been using 8mm for about 8 months with no problems." --Jack Churchill "We have been using Exabyte (on a MicroVax 3600) since February this year and have had no problems. We regularly have to restore files and directories and also haven't had a problem yet. Then again on the DEC system we only average about 6 tapes a month. On our Apollo setup we use about 20 tapes a week and so far there have been only 2 problems. 1. Caused by Sony anti-static tapes, don't use them, they're for home movie cameras only. Get certified Exabyte tapes, preferably from Exabyte themselves. Whatever gunge is on the Sony tapes it can render your complete Exabyte unit useless. 2. The systems HAVE to be serviced by the [supplier] about every 2,000 operational (tape writing/reading) hours." --Anonymous Conclusion: Although there are more complaints about Exabyte hardware reliability, this could just be because there are many more of them in use and they've been around longer. No clear advantage should be assigned to DAT. 5. Reliability of recorded tapes Conclusion: There were no complaints about unreadable Exabyte tapes, which is a good sign. For DAT, it's too early to tell. 6. Speed of dumps and restores "According to DEC it looks like their DATs can do a 10 MBytes/min." --John Hascall "Random file recovery: Exabyte 23m3s 20m14s DAT 27m11s 27m14s Dump performance: DAT's data transfer rate is slightly worse than Exabyte's." --Matthias Kandler quoting "Battle of the Backup", May '90 _Systems International_ "DAT drives have very fast seek times." --Nathan Wilson "We are starting to test the 4mm DATs since their positioning speed is faster (although transfer rate is slower and supposedly a double speed and capacity 8mm tape is also right around the corner)." --Bill Nowicki "This weekend, we backed-up about 560MB in slightly over 1 hour on the [Exabyte] we bought under the Prime Computer name. This was on an EXL325 (25MHz '386 MultiBus II, SCSI on the motherboard, not on the MultiBus). Admittedly, we need to do some more tuning on the file systems to get them to run optimally. Unlike our QIC tapes, the [Exabyte] is so quiet that I cannot determine whether it is streaming continuously or not." --Daniel Graifer "[DAT's] seek/rewind time is outstanding." --J. Eric Townsend "The transfer speed of the 8mm drives is faster than 4mm (something like 240 KB/s versus 180 KB/s). The 5 GB drive will have double the transfer speed, so it will be something like 480 KB/s." --Eduardo Krell "The throughput we have gotten [4mm] is only about 90 Kbytes per second max, and the bottleneck seems to be the Sun 3/60 SCSI." --Jim "The only complaints [about 8mm] are the search and transfer speeds during restores when people (not me) are in a hurry to read a file." --Jack Churchill "A WangDAT manual says 183 kB/s sustained data rate; 8mm is the same (I can't find the manual, I think it says 180 kB/s.)" --Sakari Jalovaara "DAT scores big points in that it seeks (skips files) much faster. If you have a full tape with 8 files, "mt fsf 8" takes *forever* on the 8mm, DAT does it in something like 30 sec worst-case (I can't find the real timings in the manual.)" --Sakari Jalovaara "But even for the normal backups, I find [DAT] annoyingly slow, beating 150K/sec on a good day. (I'm partially convinced the slowness is psychological, since the tape moves so slowly as compared with QIC drives, say.)" --Colin Plumb Conclusion: 8mm seems to have the advantage in raw throughput, whereas 4mm has a definite advantage in seeking. 7. Vendor and third party support "Both are proposed as ANSI and ISO standards. Exabyte is de facto standard DAT has yet to establish itself." --Matthias Kandler Conclusion: Adequate support seems to be available for both, with a slight edge to Exabyte due to its maturity. 8. Driver availability "Exabyte distributors tend to market for only one brand of computer. Some of the DAT drives that I've gotten info on work with a truckload of different computers, Suns, DEC, HP, IBM, Apple." --Nathan Wilson "We use /dev/rst on the Sun to talk to the DAT." --J. Eric Townsend "Apunix supplies the driver [for the WangDAT they market.] --Jim Conclusion: I'm not sure about this one. Some require add-on drivers, others don't. Make sure drivers are available for your target systems. 9. Ease of use "Only problem with the DAT was figuring out new blocksizes and other stuff for our tar/dump/restore backups." --J. Eric Townsend "Our DAT is on a SCSI port on a Sun 3/60 running OS 4.0.3, and we upgraded to that OS from 3.5 to use the drive. Apunix supplies the driver. It is essential to enable scsi disconnects with scsi_disre_enable = 1 if you have an active disk on the scsi." --Jim "I'm playing with a WangDAT 1300 at the moment (image processing - need lotsa storage space), currently driving it with some backup software that thinks it's talking to a 1/2" tape drive and so sequential-reads everything - arrgh!" --Colin Plumb Conclusion: No clear advantage either way. 10. Inherent superiority of medium "Both use magnetic media to record flux changes based on whatever modulates them - in this case, it's your data." --Gary Bridgewater "Clearly, the underlying hardware is analog in either case and that digital controllers added to video-designed drives are not theoretically incapable of providing the same results as digital-from-the start products ... however, its also a pretty challenging task." --Steven Parkes "In a high performance drive, it all looks analog anyway, but the closer you put the 1/0 decisions to the drive, the better, assuming you really plan to transcribe digital data." --George Robbins "The frequency-modulation method used to record video signals on tape is very similar to digital recording: no bias signal is used; replay amplitude and waveshape are not important; zero-crossings are. I also understand (from the hi-fi audio press, so hardly from an unimpeachable source) that the pretty much identical metal dust tape is used for both DAT and 8mm video." --Dominic Dunlop "There was a long article in one of the design rags a couple of months ago about 4mm vs. 8mm, from the standpoint of coding methods. It claimed that 4mm uses the same error correction methods designed for human-listener audio playback, while 8mm uses algorithms specially tuned for computer data storage. The conclusion was that the 8mm algorithms were far superior for archival applications. Since this article was written by an Exabyte engineer, we can be sure it is unbiased and accurate, right?" --Don Rice "I remember seeing, some months ago, a press release from Exabyte (I'm on their mailing list) clarifying this issue. The *tape transport* is 8mm format, the electronics are Exabyte's own. I believe if you check (call Exabyte), you will find that 8mm tape drives are written in a digital format, not some analog adaptation." --Greg Tarsa "The read/write heads and associated electronics for an 8mm tape were originally designed for video signals stored in analog form, and great pains were taken to ensure linear response. Encoding digital data as ones and zeros was not the major goal. Circuitry external to the video section is required to keep the error rate down. DAT was designed from the ground up to store digital data only. It is fully optimized for handling bits, bytes, and blocks. So, it's not a question of what type of connectors are on the outside of the box, but rather what the insides are designed for." --Joe Smith "DAT's 90 degree wrap angle (Exabyte 210 degree) is said to reduce head and tape wear, and enable fast file search." --Matthias Kandler "Exabyte's error checking algorithm is supposed to be significantly better." --Nathan Wilson "It's not unreasonable to assume that video tapes, which are meant to have a small head flying by at an acute angle at a high speed, recording a video signal might not perform very well when given digital data. (To be honest, I don't have the slightest idea what the heads in an Exabyte drive look like)." --Ed Falk "The magnetic particles on the tape don't know anything about digital or analog. It is a characteristic of the iron oxide that it is easily saturated, and once the magnetizing signal goes beyond the linear range, the tape quickly reaches saturation. The magnetic heads used in helical scan recording are very similar to video-heads, and a very similar technique for data storage was pioneered by IBM during the 1970s, resulting in an archival storage technology with helical scan cartridges." --Robert Kinne "To the best of my knowledge, *all* video tape (quad scan, various old helicals, VHS, 8mm) uses FM to encode the video signal. What this means is that no information is carried in the amplitude of the signal, and all of the magnetization is fully saturated. This isn't actually digital recording, but it's a lot closer than most people seem to think. (Note that many 'digital' recording systems use some form of FM to actually record the information.)" --Craig Jackson "Our group here at UW have decided to go with 4mm because it seems to us it's the up and coming technology. Some of the reasons we decided this were that the published bit error rates are something like 100 times lower with DAT, the DAT transport mechanism is simpler, [therefore] less likely to break, and all the big players are siding with DAT (Sony, Phillips, DEC, HP..). As far as I know Exabyte is the only one making 8mm drives. I think [the] DAT drives fit the 5" form factor if not the 3.5" so we could see them on PC's and such." --Laurence Lundblade Conclusion: No clear advantage. Only time will tell. 11. Maturity of the technology "People on the net (including this site) are quite fond of their 8mm drives but they are all newish so there is little known about their long term functioning - compared to 9 track tapes for instance. DAT is even newer." --Gary Bridgewater "George [Goble] has spent about 2 years wringing out the Exabyte and helping them get it right, so it is NOT "newish". Further, I know of noone in the world who is better at breaking hardware than George, so if he's happy, I'll buy one sight-unseen. So, if you want to help someone debug their product, go buy a DAT drive. Otherwise, leverage the time spent by other folks who are very good at helping people debug their products and buy one already debugged!!" --Michael O'Dell Conclusion: Both are immature, but 8mm has a slight edge. OVERALL CONCLUSION Exabyte and DAT each have their pros and cons. In order to decide which is right for you, you need to match your particular needs to these pros and cons. Exabytes hold more and record faster (today), so they're good for making backups. Unfortunately, they seek very slowly, so they're not as well-suited to restoring files from backups, or other non-backup uses. The quick seek capability of DATs makes them good for tertiary storage: keeping large amounts of data on-line, but off the hard disk. It also makes multiple-dump files per tape more practical. THE CONTRIBUTORS: (thanks again, guys) Gary Bridgewater, Data General Corporation gary@sv.dg.com,r {amdahl,aeras,amdcad}!dgcad!gary Michael O'Dell, mo@messy.bellcore.com Steven Parkes, Coordinated Science Laboratory, University of Illinois steven@pacific.csl.uiuc.edu George Robbins, Commodore, Engineering Department {uunet|pyramid|rutgers}!cbmvax!grr, grr@cbmvax.commodore.com Dominic Dunlop, The Standard Answer Ltd., domo@tsa.co.uk Don Rice, University of Alaska Fairbanks fnddr@acad3.fai.alaska.edu, fnddr@alaska (bitnet) John Hascall, Project Vincent, ISU Computation Center john@iastate.edu / hascall@atanasoff.cs.iastate.edu Greg Tarsa, Tarsa Software Consulting tarsa@elijah.mv.com, {decuac,decvax}!elijah!tarsa Jonathan Krueger, Defense Technical Information Center jkrueger@dtic.dla.mil, uunet!dgis!jkrueger Joe Smith, BT Tymnet Tech Services jms@tardis.tymnet.com, jms@gemini.tymnet.com, {ames,pyramid}!oliveb!tymix!tardis!jms Matthias Kandler, Inst. fuer Informatik kandler@informatik.tu-muenchen.dbp.de Nathan Wilson, Teleos Research nathan%teleos.com@ai.sri.com Bill Nowicki, Legato Systems nowicki@legato.com Daniel Graifer, Franklin Mortgage Capital Corporation uunet!dag@fmccva.franklin.com Ed Falk, Sun Microsystems sun!falk, falk@sun.com J. Eric Townsend, University of Houston Dept. of Mathematics jet@uh.edu, Bitnet: jet@UHOU Robert Kinne, University of Colorado, Boulder bobk@boulder.Colorado.EDU Eduardo Krell, AT&T Bell Laboratories {att,decvax,ucbvax}!ulysses!ekrell, ekrell@ulysses.att.com Jim, Interet uunet!interet!jim Craig Jackson, DRI/McGraw-Hill dricejb@drilex.dri.mgh.com {bbn,axiom,redsox,atexnet,ka3ovk}!drilex!{dricej,dricejb} Jack Churchill, CSIRO Division of Exploration Geoscience, Remote Sensing Group churchill@decus.com.au Sakari Jalovaara, Helsinki University of Technology sja@sirius.hut.fi Colin Plumb, Array Systems Computing, Inc. colin@array.UUCP Laurence Lundblade, Networks and Distributed Computing, U of Washington lgl@cac.washington.edu Anonymous, large european electronics firm, research c/o de5@ornl.gov DISCLAIMER: This information is provided without any warranty. It does not represent the official word of the companies employing the contributors. -- Dave Sill (de5@ornl.gov) These are my opinions. Martin Marietta Energy Systems Workstation Support
eggert@twinsun.com (Paul Eggert) (07/18/90)
Dave Sill's nice summary wrote: "DAT tapes cost $15.00 and seem to be going up in price." --uunet!interet!jim We pay $7-$8 a tape for 8mm tapes (Sony P6-120MP) in small quantities. This isn't the ``certified Exabyte tape'', but our drive is still working after over a year, dumping about a gigabyte per day.