lamaster@ames.arc.nasa.gov (Hugh LaMaster) (07/27/89)
I keep seeing all these ads for synchronous SCSI controllers with speeds in the 4-5 MB/sec. range. But I don't see the ads for 4-5 MB/sec. synchronous SCSI disks. Are there any reasonably inexpensive general purpose high capacity etc. synchronous SCSI disks out there with speeds of, well, what is available? Hugh LaMaster, m/s 233-9, UUCP ames!lamaster NASA Ames Research Center ARPA lamaster@ames.arc.nasa.gov Moffett Field, CA 94035 Phone: (415)694-6117
olson@anchor.sgi.com (Dave Olson) (07/27/89)
lamaster@ames.arc.nasa.gov (Hugh LaMaster) writes: >I keep seeing all these ads for synchronous SCSI controllers with speeds >in the 4-5 MB/sec. range. But I don't see the ads for 4-5 MB/sec. >synchronous SCSI disks. Are there any reasonably inexpensive general >purpose high capacity etc. synchronous SCSI disks out there with speeds of, >well, what is available? These speeds are typically the burst speeds over the scsi bus when talking about disks; devices that transfer from/to RAM might actually be able to sustain them. There are disks coming on the market now or in the next 2-3 months that are 20 Mhz drives and can deliver sustained 2.1Mb/sec on read, and 1.9Mb/sec on writes, when doing i/o on contiguous files with 128K or larger read/write size on the raw device; less, but not too much lower with sizes down to about 16K. Filesystem rates are typically lower... Some of these drives actually can deliver bursts from/to their buffer at 5Mb/sec as measured with a logic analyzer, for 32K to 64K at a time. (Some of these drivers are zone bit recording drives, so the throughput drops off as you near the end of the drive.) Of course, your host has to be able to handle it also; the new Silicon Graphics 4D25 (20 MHz Personal Iris) with the WD 33C93A SCSI chip is the machine on which the above throughput was measured. Dave Olson It's important to keep an open mind, but not so open that your brains fall out. -- Stephen A. Kallis, Jr.
george@wombat.UUCP (George Scolaro) (07/27/89)
In article <29254@ames.arc.nasa.gov> lamaster@ames.arc.nasa.gov (Hugh LaMaster) writes: >I keep seeing all these ads for synchronous SCSI controllers with speeds >in the 4-5 MB/sec. range. But I don't see the ads for 4-5 MB/sec. >synchronous SCSI disks. Are there any reasonably inexpensive general >purpose high capacity etc. synchronous SCSI disks out there with speeds of, >well, what is available? Hitachi have a new-ish drive: DK515C-78 (660 meg 5.25 in). The quoted storage is in mbytes (formatted of course). We checked a while ago and the 660 meg drive was around $2500 or so. It supports sync. scsi at 4 mbytes/sec, and sustained transfer off the media of 2.5 megabytes/sec (impressive). All I know about the drive is from technical docs. I haven't bought one yet, but plan to in the near future, for a home brew NS32532 system with an Adaptec AIC6250 SCSI chip (which supports sync scsi), I want MORE SPEED! BTW the drive also supports old/slow scsi at 1.5mbytes/sec. -- George Scolaro george@wombat (try {pyramid|sun|vsi1|killer} !daver!wombat!george)
lamaster@ames.arc.nasa.gov (Hugh LaMaster) (07/27/89)
In article <257@odin.SGI.COM> olson@anchor.sgi.com (Dave Olson) writes: >lamaster@ames.arc.nasa.gov (Hugh LaMaster) writes: >>I keep seeing all these ads for synchronous SCSI controllers with speeds >>in the 4-5 MB/sec. range. But I don't see the ads for 4-5 MB/sec. >>synchronous SCSI disks. Are there any reasonably inexpensive general >These speeds are typically the burst speeds over the scsi bus >when talking about disks; devices that transfer from/to RAM might >actually be able to sustain them. >that are 20 Mhz drives and can deliver sustained 2.1Mb/sec >on read, and 1.9Mb/sec on writes, when doing i/o on contiguous >Some of these drives actually can deliver bursts from/to their >buffer at 5Mb/sec as measured with a logic analyzer, for 32K to 64K I realize that the 4-5 MB/sec. is the SCSI bus rate, but, I guess what I am asking is this: IBM, CDC, and more recently, SMD-E drives have been around, (some of these for years) with speeds >= 3.0 MB/sec. CDC and Ibis have produced drives with data rates of 10-12 MB/sec (I know, parallel heads). The need and technology are clearly there, though not necessarily in the same place (price/performance-wise). My question is: now that "cheap" drives with data rates from a single head of 24MHz are available, why aren't there synchronous SCSI drives with at least such speeds available? The new high speed workstations, like the SGI 4D/25 mentioned by Dave Olson, are crying out for higher speed disks. I believe that there would be probably be a market for a dual-parallel-head drive which would effectively operate at 48MHz (2x24) and which could keep a 5 MByte/sec SCSI bus busy. About 4 of these would turn one of the new 20 MIPS workstations from various companies into a real "CDC 7600 on a desk". Hugh LaMaster, m/s 233-9, UUCP ames!lamaster NASA Ames Research Center ARPA lamaster@ames.arc.nasa.gov Moffett Field, CA 94035 Phone: (415)694-6117
markb@denali.sgi.com (Mark Bradley) (07/28/89)
In article <29289@ames.arc.nasa.gov>, lamaster@ames.arc.nasa.gov (Hugh LaMaster) writes: > In article <257@odin.SGI.COM> olson@anchor.sgi.com (Dave Olson) writes: > >lamaster@ames.arc.nasa.gov (Hugh LaMaster) writes: > > >when talking about disks; devices that transfer from/to RAM might > >actually be able to sustain them. > > >that are 20 Mhz drives and can deliver sustained 2.1Mb/sec > >on read, and 1.9Mb/sec on writes, when doing i/o on contiguous > > >Some of these drives actually can deliver bursts from/to their > >buffer at 5Mb/sec as measured with a logic analyzer, for 32K to 64K > > I realize that the 4-5 MB/sec. is the SCSI bus rate, but, I guess what I am > asking is this: IBM, CDC, and more recently, SMD-E drives have been around, > (some of these for years) with speeds >= 3.0 MB/sec. CDC and > Ibis have produced drives with data rates of 10-12 MB/sec > My question is: now that "cheap" drives with data rates from a single head > of 24MHz are available, why aren't there synchronous SCSI drives with at > least such speeds available? The new high speed workstations, like the > SGI 4D/25 mentioned by Dave Olson, are crying out for higher speed disks. > They are coming. Although I can't give names and dates, think in terms of multiple HDA's under the control of a single embedded controller with significant 'cache' on board. Multihead/surface disks in 5.25 or 3.5 inch form factor probably don't make sense, but there are ways other than that to get significantly faster raw data rates from disks. 24 MHz drives are just around the corner without ZBR techniques, as well. It is important to realize, too, that just because the disks are fast that the file systems out there in unix land today can really deliver the read performance that the hardware can provide. Advances are being made to take advantage of this faster technology too, but they must be developed together in order for it all to work. We'll have something for you soon. markb -- Mark Bradley "Faster, faster, until the thrill of IO Subsystems speed overcomes the fear of death." Silicon Graphics Computer Systems Mountain View, CA ---Hunter S. Thompson
tneale@aeras.UUCP (Tom Neale) (07/28/89)
In article <29254@ames.arc.nasa.gov> lamaster@ames.arc.nasa.gov (Hugh LaMaster) writes: >I keep seeing all these ads for synchronous SCSI controllers with speeds >in the 4-5 MB/sec. range. But I don't see the ads for 4-5 MB/sec. >synchronous SCSI disks. Are there any reasonably inexpensive general >purpose high capacity etc. synchronous SCSI disks out there with speeds of, >well, what is available? I'm not sure how you rate "inexpensive" but most of the newer high capacity (380 MB to 1.2 GB) 5.25" drives are sync SCSI capable. Also, the new 3.5" drives (200 MB and up) seem to have this feature. Yes, the drives are available and are shipping. The SCSI spec says that sync should run at 4.0 MB/sec. Some drives will go faster. To an OEM or system integrator these drives usually cost $3.50 to $5.00 per megabyte. You can usually double or triple that by the time it gets to the end user. -- Blue skies, | ...sun!aeras!tneale | | in flight: N2103Q | The hurrieder I go Tom Neale | in freefall: D8049 | the behinder I get. | via the ether: WA1YUB |
buck@siswat.UUCP (A. Lester Buck) (07/28/89)
In article <29289@ames.arc.nasa.gov>, lamaster@ames.arc.nasa.gov (Hugh LaMaster) writes: > >that are 20 Mhz drives and can deliver sustained 2.1Mb/sec > >on read, and 1.9Mb/sec on writes, when doing i/o on contiguous > > >Some of these drives actually can deliver bursts from/to their > >buffer at 5Mb/sec as measured with a logic analyzer, for 32K to 64K > > I realize that the 4-5 MB/sec. is the SCSI bus rate, but, I guess what I am > asking is this: IBM, CDC, and more recently, SMD-E drives have been around, > (some of these for years) with speeds >= 3.0 MB/sec. CDC and > Ibis have produced drives with data rates of 10-12 MB/sec > (I know, parallel heads). The need and technology are clearly there, > though not necessarily in the same place (price/performance-wise). If you really want to, just take an SMD drive at 3.0 MB/sec and run it through the NCR ADP-48 SCSI to SMD synchronous controller for a top speed of 4 MB/sec across the bus (minimum synchronous period 250 ns). This is the fastest SCSI disk subsystem which I have heard of - please let me know if anyone can configure something faster. > I believe that there would be probably be a market for a dual-parallel-head > drive which would effectively operate at 48MHz (2x24) and which could keep > a 5 MByte/sec SCSI bus busy. About 4 of these would turn one of the new > 20 MIPS workstations from various companies into a real "CDC 7600 on a desk". But if you keep the SCSI bus busy (continuously), you have removed a major reason to use SCSI in the first place. You now cannot add any more drives on this bus without a disastrous performance hit. Although the programming would not be as easy as with SCSI, you might as well have a host bus to SMD controller directly for each drive. To effectively and continuously use drives that have sustained transfer rates of several MB/sec will require either the "wide" SCSI or "fast" SCSI (or both) parts of SCSI-2. Of course, few host buses can sustain such data rates (and especially not four at once!), so this is all rather acedemic. -- A. Lester Buck ...!texbell!moray!siswat!buck
cdr@amdcad.AMD.COM (Carl Rigney) (07/28/89)
Ciprico's latest SCSI VME board supports the SCSI-2 wide standard. With a 32bit data path instead of 8bit it should reach 20 MByte/sec synch burst rates. I believe Imprimis either has or is about to have some drives that do wide SCSI. I suspect the problem will be finding hosts that can accept that kind of data input. CDC Wren V's (600 MB formatted, 16ms, 5MB synch) are selling for around $2500; I've also heard good things about the Hitachi that someone else mentioned. Ciprico said at Sun-Expo that they expect Fast SCSI-2 chips to come on the market in January, and they expect to have a board using them by April or so. Fast, Wide SCSI is 40 Megabytes/Sec! At that point you probably have to talk RAID to get the necessary disk speed. --Carl Rigney Glossary for those who are confused: SCSI-1 standard uses a 50-pin connector with 8 data lines. SCSI-2 defines a "wide" implementation that has a second cable with 8 or 24 additional data lines, doubling or quadrupling your data rate. SCSI-2 also defines a "fast" implementation that tightens up the timings to allow 10 MB synch on the old cables, or 40MB if combined with the 32-bit path. A "RAID" is a set of parallel disks with striping by word, typically, and usually a parity drive. By using 5 disks (1 parity) you can quadruple your data rate; by using 9 you get 8x. Some have the capability to lose a drive and keep going, recreating the replacement drive (after you swap out the bad one) off the Parity drive. (Others require downtime while it recreates the missing drive.) Mean time to Data Loss for RAIDS is astronomical because you have to lose two drives at the same time. Someone recently mentioned that NCR's "Introduction to SCSI" book is now being published by Prentice-Hall - I'd recommend it highly. I got mine at Techmart after the IEEE conference on System Design & Mass Storage last month; I don't know if they still have any.