mhorne@ka7axd.WV.TEK.COM (Mike Horne) (10/04/90)
I've got a couple of SCSI hard drives that I'd like to install in my PC. I'm looking for SCSI controller recommendations, in particular: - I need a 16-bit SCSI controller that will be fast. - I need the corresponding SCSI driver software (I would assume that this would come with the board). I would also like to find a driver that will allow me to do system backups to a SCSI streaming tape drive. An Adaptec card has been recommended to me (I don't know the model number), but I'm also interested in others. Thanks in advance! Mike Horne mhorne@ka7axd.wv.tek.com mhorne%ka7axd.wv.tek.com@relay.cs.net ...tektronix!mhorne%ka7axd.wv.tek.com
larry@nstar.uucp (Larry Snyder) (10/08/90)
mhorne@ka7axd.WV.TEK.COM (Mike Horne) writes: >I've got a couple of SCSI hard drives that I'd like to install in my PC. >I'm looking for SCSI controller recommendations, in particular: > - I need a 16-bit SCSI controller that will be fast. > - I need the corresponding SCSI driver software (I would assume that > this would come with the board). I would also like to find a > driver that will allow me to do system backups to a SCSI streaming > tape drive. >An Adaptec card has been recommended to me (I don't know the model number), >but I'm also interested in others. The best SCSI controller available for AT bus machines is the Adaptec 1542B. By best - the 1542 yields the fastest throughput under 386 Unix (Interactive) and DOS, and with the public domain support utilities to re-program the devices - one can really make the 1542 shine as compared against the other controllers on the market. Several folks with the WD 1007 FAST controllers have reported problems with a lack of support - and slower throughput.. -- Larry Snyder, Northern Star Communications, Notre Dame, IN USA {larry@nstar, uunet!sco!romed!nstar!larry, nstar!larry@ndmath.math.nd.edu} backbone usenet newsfeeds available Public Access Unix Site (219) 289-0282 (5 high speed lines)
pjh@mccc.uucp (Pete Holsberg) (10/08/90)
In article <1990Oct07.173315.6532@nstar.uucp> larry@nstar.uucp (Larry Snyder) writes:
=The best SCSI controller available for AT bus machines is the
=Adaptec 1542B. By best - the 1542 yields the fastest throughput
=under 386 Unix (Interactive) and DOS, and with the public domain
=support utilities to re-program the devices - one can really make
=the 1542 shine as compared against the other controllers on the
=market. Several folks with the WD 1007 FAST controllers have reported
=problems with a lack of support - and slower throughput..
What's available to measure the throughput, Larry?
Pete
--
Prof. Peter J. Holsberg Mercer County Community College
Voice: 609-586-4800 Engineering Technology, Computers and Math
UUCP:...!princeton!mccc!pjh 1200 Old Trenton Road, Trenton, NJ 08690
Internet: pjh@mccc.edu Trenton Computer Festival -- 4/20-21/91flint@gistdev.gist.com (Flint Pellett) (10/09/90)
larry@nstar.uucp (Larry Snyder) writes: >mhorne@ka7axd.WV.TEK.COM (Mike Horne) writes: >>I've got a couple of SCSI hard drives that I'd like to install in my PC. >>I'm looking for SCSI controller recommendations, in particular: >> - I need a 16-bit SCSI controller that will be fast. >> - I need the corresponding SCSI driver software (I would assume that >> this would come with the board). I would also like to find a >> driver that will allow me to do system backups to a SCSI streaming >> tape drive. >>An Adaptec card has been recommended to me (I don't know the model number), >>but I'm also interested in others. >The best SCSI controller available for AT bus machines is the >Adaptec 1542B. By best - the 1542 yields the fastest throughput >under 386 Unix (Interactive) and DOS, and with the public domain >support utilities to re-program the devices - one can really make >the 1542 shine as compared against the other controllers on the >market. Several folks with the WD 1007 FAST controllers have reported >problems with a lack of support - and slower throughput.. I'd be careful about calling it the "best". It may be "best for the money" but there are (IMO) better controllers available, but for more money. For example, the PSI Hyper-Store controller. (There may be others that are even better than it, but I don't know of them.) This controller has several features that are rather nice. For example: the controller has a generic part and then small (4" x 4") media adapter(s) mounted on the end: it can have up to 7 media adapters at once, with various combinations for different types of disks. For example, you could install it with one SCSI media adapter, one ESDI, and one RLL, and then use all three types of disks in your machine with only one controller. If you stack 7 SCSI adapters, the one controller can connect 49 SCSI devices. Two years from now when the world decides SCSI-2 (or whatever) is the way to go, instead of buying a whole new controller and throwing away your old disk, or burning 2 slots for 2 controllers, on the PSI you can just add a SCSI-2 adapter. (You will have to make sure you have a short card next to this one, but who doesn't have at least one short ports card or mouse card?) The Hyperstore also has up to 4MB of memory on the controller for doing caching- not that big a deal in UNIX, but a big win in DOS. (Virtually all the benchmark programs report back a 0.4 ms (yes, 4/10th, not 4) access times because they never use up the whole cache. -- Flint Pellett, Global Information Systems Technology, Inc. 1800 Woodfield Drive, Savoy, IL 61874 (217) 352-1165 uunet!gistdev!flint or flint@gistdev.gist.com
larry@nstar.uucp (Larry Snyder) (10/11/90)
pjh@mccc.uucp (Pete Holsberg) writes: >=The best SCSI controller available for AT bus machines is the >=Adaptec 1542B. By best - the 1542 yields the fastest throughput >=under 386 Unix (Interactive) and DOS, and with the public domain >=support utilities to re-program the devices - one can really make >=the 1542 shine as compared against the other controllers on the >=market. Several folks with the WD 1007 FAST controllers have reported >=problems with a lack of support - and slower throughput.. >What's available to measure the throughput, Larry? we have several benchmarks available on nstar, one of which is the byte unix benchmarks from 6-8 months ago.. -- Larry Snyder, Northern Star Communications, Notre Dame, IN USA {larry@nstar, uunet!sco!romed!nstar!larry, nstar!larry@ndmath.math.nd.edu} backbone usenet newsfeeds available Public Access Unix Site (219) 289-0282 (5 high speed lines)
iverson@xstor.UUCP (Tim Iverson) (10/11/90)
In article <1990Oct07.173315.6532@nstar.uucp> larry@nstar.uucp (Larry Snyder) writes: >The best SCSI controller available for AT bus machines is the >Adaptec 1542B. By best - the 1542 yields the fastest throughput >under 386 Unix (Interactive) and DOS, True for Unix, very false for DOS. The non-intelligent boards are much faster than the 154xB series on a single-threaded OS (e.g. DOS) due to the high overhead on each command: about 2ms per command for the 1542B. There are also fewer integration problems since you don't have to worry about the 1st party DMA getting thrown for a loop by the 386 memory managers. Of course, as bad as the 1542 is for DOS, it's just about the best you can find for Unix - almost all support it with a native driver (for some it's the only SCSI board you can boot from), and it provides some very helpful features to boost performance under Unix. There are boards that do much better, but they do not have the widespread acceptance that the 1542 has. >Several folks with the WD 1007 FAST controllers have reported >problems with a lack of support - and slower throughput.. The WD7000 FASST board performs about the same as the 1542 under DOS; I don't know how well it does under Unix - I've never seen a driver for it. I last used one about 6 months ago, when I had to try three different boards before finding just one that worked and had to do the integration with a non-wd board because their own software (by Columbia Data) wouldn't work correctly with the 7000. > Larry Snyder, Northern Star Communications, Notre Dame, IN USA > {larry@nstar, uunet!sco!romed!nstar!larry, nstar!larry@ndmath.math.nd.edu} - Tim Iverson uunet!xstor!iverson
larry@nstar.uucp (Larry Snyder) (10/14/90)
iverson@xstor.UUCP (Tim Iverson) writes: >True for Unix, very false for DOS. The non-intelligent boards are much >faster than the 154xB series on a single-threaded OS (e.g. DOS) due to the >high overhead on each command: about 2ms per command for the 1542B. There >are also fewer integration problems since you don't have to worry about the >1st party DMA getting thrown for a loop by the 386 memory managers. I had a seagate ST01 here, and it coretested at 700kb/sec - while the 1542 always comes it about 960kb/sec (these values are of course with no cache) under PC-DOS 3.3 -- Larry Snyder, Northern Star Communications, Notre Dame, IN USA {larry@nstar, uunet!sco!romed!nstar!larry, nstar!larry@ndmath.math.nd.edu} backbone usenet newsfeeds available Public Access Unix Site (219) 289-0282 (5 high speed lines)
iverson@xstor.UUCP (Tim Iverson) (10/16/90)
In article <1990Oct13.231715.1548@nstar.uucp> larry@nstar.uucp (Larry Snyder) writes: >I had a seagate ST01 here, and it coretested at 700kb/sec - while the >1542 always comes it about 960kb/sec (these values are of course with >no cache) under PC-DOS 3.3 Well, you've chosen the slowest, most buggy of the dumb boards to compare it to; why don't you try something better (SDC800, Future Domain, etc.). If the 1542 turns in 960kb/sec, then a good dumb board should yield 1.1MB/s. BTW - I use a 1542B myself, but then I switch between DOS and Unix often, so the extra cost of the board is justified. > Larry Snyder, Northern Star Communications, Notre Dame, IN USA > {larry@nstar, uunet!sco!romed!nstar!larry, nstar!larry@ndmath.math.nd.edu} - Tim Iverson uunet!xstor!iverson
jdg@neptune.sisd.kodak.com (Jeff Gortatowsky CUST) (10/16/90)
I use a 1542A (don't know what the B rev gets you) and I get a CORETEST (v2.92) of 1.2mb/sec on my 386AT (320mb Imprimis Wren). I run DOS 3.3. But lack of an OS/2 driver hurts me. Also some tools which play with DMA don't work (for me anyways) on a high speed setting. Tools like FASTBACK and PC-TOOL PCBACKUP have to be set to medium speed. SCSI drives seem smart enough to make most "disk optimizers" redundant. All in all like SCSI. If only I had an OS/2 driver... -- Jeff Gortatowsky-Eastman Kodak Company | UUCP : ..uunet!sisd!jdg Systems Integration & Software Development | Voice : (716)-726-0084 (EST/EDT) ------------------------------------------- USMail: Rochester NY 14653-5810 Kodak makes film not comments. Therefore these comments are mine not Kodak's.
poffen@sj.ate.slb.com (Russ Poffenberger) (10/17/90)
In article <1990Oct13.231715.1548@nstar.uucp> larry@nstar.uucp (Larry Snyder) writes: >iverson@xstor.UUCP (Tim Iverson) writes: > >>True for Unix, very false for DOS. The non-intelligent boards are much >>faster than the 154xB series on a single-threaded OS (e.g. DOS) due to the >>high overhead on each command: about 2ms per command for the 1542B. There >>are also fewer integration problems since you don't have to worry about the >>1st party DMA getting thrown for a loop by the 386 memory managers. > >I had a seagate ST01 here, and it coretested at 700kb/sec - while the >1542 always comes it about 960kb/sec (these values are of course with >no cache) under PC-DOS 3.3 > You will find that the AHA1542 will outperform on transfer rate, but because of the intelligent "sector mapping" it does, there tends to be a little overhead when it comes to average access and track to track seek times. Russ Poffenberger DOMAIN: poffen@sj.ate.slb.com Schlumberger Technologies UUCP: {uunet,decwrl,amdahl}!sjsca4!poffen 1601 Technology Drive CIS: 72401,276 San Jose, Ca. 95110 (408)437-5254
iverson@xstor.UUCP (Tim Iverson) (10/18/90)
In article <1990Oct16.223855.4793@sj.ate.slb.com> poffen@sj.ate.slb.com (Russ Poffenberger) writes: >You will find that the AHA1542 will outperform on transfer rate, but because of >the intelligent "sector mapping" it does, [...] The AHA1542 does not do any "intelligent" sector mapping. Assuming that you're referring to the fact that it reports an artificial geometry in order to satisfy utilities that require one, there is no overhead to speak of converting the 64head/32sector geometry into a SCSI LBA. e.g. lba = (cyl << 11) | (hd << 5) | (sec - 1) Not only is overhead on this calculation negligible, but almost all SCSI HAs must perform the exact same calculation, so when comparing performance, this factor is a wash - both do it, and both take about the same time to do so. >[...] there tends to be a little overhead >when it comes to average access and track to track seek times. This difference between published times and coretest times *is* due to the geometry translation, but not due to calculation overhead. Because of the translation, there is usually more than one real cylinder for each artificial cylinder, thus coretest thinks it's seeking 1 cylinder when it may be seeking farther. This does not affect comparisons of coretest results obtained for the same or similar HAs. The per command overhead on the AHA 154x host adapters is due mostly to command queueing. The 154x HA is designed to perform well in multi-threaded situations. As a consequence, it does not perform as well as possible in single-threaded situations. Putting a 1542 in a DOS only machine is like buying a Ferrari for your grandmother - it can go 200+, but she'll never drive it faster than 35. >Russ Poffenberger DOMAIN: poffen@sj.ate.slb.com >Schlumberger Technologies UUCP: {uunet,decwrl,amdahl}!sjsca4!poffen - Tim Iverson uunet!xstor!iverson
larry@nstar.uucp (Larry Snyder) (10/19/90)
iverson@xstor.UUCP (Tim Iverson) writes: >Well, you've chosen the slowest, most buggy of the dumb boards to compare >it to; why don't you try something better (SDC800, Future Domain, etc.). If >the 1542 turns in 960kb/sec, then a good dumb board should yield 1.1MB/s. the ST01 was the only other dumb board I had.. >BTW - I use a 1542B myself, but then I switch between DOS and Unix often, so >the extra cost of the board is justified. -- Larry Snyder, Northern Star Communications, Notre Dame, IN USA {larry@nstar, uunet!sco!romed!nstar!larry, nstar!larry@ndmath.math.nd.edu} backbone usenet newsfeeds available Public Access Unix Site (219) 289-0282 (5 high speed lines)