rjkloost@THUNDER.LAKEHEADU.CA (05/16/91)
S and it is on all hours.. ) Thanks rjkloost@THUNDER.LAKEHEADU>CA [A[C[C[C[C[C[C[C[C[C[C[C[C[C[C[C[C[C[C[C[C[C[C[C[C[C[ [C[D [D[D[D[D[D[D[DU.CA
FelineGrace@cup.portal.com (Dana B Bourgeois) (05/17/91)
On a related note:
Seagate tells me they will be announcing/shipping a new (very large)
drive this fall. It is about 2.4 Gbyte unformatted. I asked because
I am spec-ing a Sun SPARC at work and want the largest drives I can
get my hands on. One of the specs intruiged me. Throughput. I have
seen this spec quoted several different ways and I am wondering about
it. Peak numbers are quoted and sustained numbers. Sometimes just
numbers. The units are either MegaHertz or MegaBytes/Minute. Now the
guy I talked to said the rotation speed is to be 3600 RPM and yet he
claims a thruput of 10 MegaBytes/Sec (I said 'minute' up above. So I
messed up! Shut up! --E. Murphy). Now I tried to convert that to a
sustained megahertz number and came up with 80 MHz so I think the 10 MB/S
is the SCSI burst speed not the hard drive device sustained speed.
OK, finally for my questions:
1. What is the fastest sustained throughput drive the Amiga
can handle? 20 MHz? 28MHz? What if any 'magic' is needed
to handle such a drive? Controller/memory/etc.
2. Sun and other workstation vendors already claim to use
SCSI 2. Does CBM have plans to support it? I'm not asking
about products. More design philosophy. Will SCSI 2 give
enough increase in speed to make a viable product as opposed
to a 'me-too' marketing product.
3. I see transfer rates being batted around in workstation/unix
magazines of 4-5 MB/sec. Used to be people on this thread
said the Amiga's 1.2 MB/s SCSI was one of the fastest SCSI
implementations going. I've seen Amiga numbers around 2 MB/s.
Have SCSI designs on other machines really tripled/quadrupled
in speed?
4. Assuming that sustained throughput is an important spec for
hard drives, anybody know a good rule-of-thumb for estimating
the speed of a drive given a few common factors like, average
seek time, rotation speed, max bit density/unformatted capacity?
Seems like you should be able to do this.
Thanks in advance
Dana Bourgeois @ cup.portal.com
"Why do I ask such dumb questions? All the good ones are already taken!"daveh@cbmvax.commodore.com (Dave Haynie) (05/21/91)
In article <42421@cup.portal.com> FelineGrace@cup.portal.com (Dana B Bourgeois) writes: >On a related note: >Seagate tells me they will be announcing/shipping a new (very large) >drive this fall. It is about 2.4 Gbyte unformatted. >OK, finally for my questions: > 1. What is the fastest sustained throughput drive the Amiga > can handle? 20 MHz? 28MHz? What if any 'magic' is needed > to handle such a drive? Controller/memory/etc. Specifying a frequency here is meaningless, because you're interested in bandwith. Bandwith is generally measured in bytes/second, which gives you numbers that are independent of system architecture. All things being equal, a system that manages 10MB/s could be a 64 bit RISC thing clocked at 1.25MHz, a 68000 clocked at around 20MHz. Anyway, in the A3000, the weakest link is the SCSI bus. It runs somewhere about 5MB/s in synchronous mode, which the A3000 SCSI hardware has no trouble keeping up with. Actual transfers from the DMA controller's FIFO to memory happen at around 20MB/s. > 2. Sun and other workstation vendors already claim to use > SCSI 2. Does CBM have plans to support it? I'm not asking > about products. More design philosophy. Will SCSI 2 give > enough increase in speed to make a viable product as opposed > to a 'me-too' marketing product. There is a SCSI-2 version of the A3000's Western Digital SCSI chip. Right now, they don't have a scsi.device for it. It's probably not going to buy you much anyway, though, since even with killer drives, you're going to be limited by the speed of the drive. The main advantage of the faster modes on SCSI buses is when you have multiple drives in place. > 3. I see transfer rates being batted around in workstation/unix > magazines of 4-5 MB/sec. Nearly all of the system benchmarks I've seen (in places like "UNIX Review" and "Personal Workstation") measure disk rate in KB/s. Most UNIX systems can sustain 350-500kB/s. A few really good ones go higher, but not many. At least until you get into RAID boxes instead of single SCSI drives. For example, the April 1991 issue of "UNIX Review" tests a few low cost UNIX systems. They get a maximum of 617kB/s for the Sun SLC, 436kB/s for the DECstation 2100, 295kB/s for the HP DN2500, and 423kB/s for a generic '486 PClone (all sequential reads, see page 49 for more details). > Used to be people on this thread said the Amiga's 1.2 MB/s SCSI was > one of the fastest SCSI implementations going. I've seen Amiga > numbers around 2 MB/s. Have SCSI designs on other machines really > tripled/quadrupled in speed? For the most part, no. You're either looking at the wrong numbers, or thinking of some pretty expensive workstations. Also, beware what you're looking at. Drive people like to talk in terms of peak transfer rates, rarely sustained transfer rates. Real meaningful benchmarks have to reflect what's happening through a machine's filesystem, which is what you're seeing with Amigas in most cases. The filesystem takes into account Disk, SCSI, host adaptor, device driver, and filesystem efficiencies. >Dana Bourgeois @ cup.portal.com -- Dave Haynie Commodore-Amiga (Amiga 3000) "The Crew That Never Rests" {uunet|pyramid|rutgers}!cbmvax!daveh PLINK: hazy BIX: hazy "That's me in the corner, that's me in the spotlight" -R.E.M.
FelineGrace@cup.portal.com (Dana B Bourgeois) (05/23/91)
Dave, thanks for all the comments on my questions about sustained SCSI transfer rates. First, I am talking about a SPARC 2 so I don't know if that is considered a low-cost workstation or not. You didn't mention it when listing a few models so I'd guess not. SUN prints numbers like 2-3 MB/s for it and claims it is SCSI 2. (oops, they claim 1.8 MB/s, 3 MB/s synchronous). Second, after I posted my qustions I received the latest Seagate catalog. They have some 5400 RPM SCSI-2 drives they rate at 28MB/s internal transfer and 10MB/s external transfer. Do you think they are saying the drive has the capability to transfer data+clock bits at 28MB/s and the SCSI protocol is holding it back to 10? (I ask Seagate and their sales people won't speculate.) I know that current disks (40-100 MByte quantums for example) are the bottle neck in data transfer with the Amiga line of machines. I am curious if disks are starting to come out that can transfer data faster than the SCSI interface. I tried to calculate disk rotation speed versus bits per track but that information isn't included for SCSI disks. If sustained transfer rates are increasing from 500KB/s to 2MB/s then there must be more bits per track since the fastest spinning drives are increasing RPM by only 50%. What's the real story in sustained transfer rates? Dana Bourgeois @ cup.portal.com
mscritsm@isis.cs.du.edu (Milton Scritsmier) (05/24/91)
> >Second, after I posted my qustions I received the latest Seagate catalog. >They have some 5400 RPM SCSI-2 drives they rate at 28MB/s internal >transfer and 10MB/s external transfer. Do you think they are saying >the drive has the capability to transfer data+clock bits at 28MB/s >and the SCSI protocol is holding it back to 10? (I ask Seagate and >their sales people won't speculate.) > Actually, that 28 MB/s figure is 28 megabits/sec. Megabits/sec is the standard measure these days for data transfer rates to the media. It does not take into account overhead such as the media format, but is a raw transfer rate number. Thus that 28 mbits/sec rate translates to 3.5 megabytes/sec. Thus even the current SCSI-1 synchronous transfer rate of 5 megabytes/sec is sufficient to keep up with the drive. >I know that current disks (40-100 MByte quantums for example) are the >bottle neck in data transfer with the Amiga line of machines. I am >curious if disks are starting to come out that can transfer data >faster than the SCSI interface. I tried to calculate disk rotation >speed versus bits per track but that information isn't included for >SCSI disks. If sustained transfer rates are increasing from 500KB/s >to 2MB/s then there must be more bits per track since the fastest >spinning drives are increasing RPM by only 50%. > >What's the real story in sustained transfer rates? There are more bits/track than before, especially since many drives now divide the media into zones where the number of bits per inch along any track is roughly constant. This has the effect of putting more bits per track on the outer part of the disk. (Without zones, the number of bits per track was constant across all tracks. This meant that the size of each bit cell was smaller nearer the inner part of the media than the outer part of the media since the circumference of a track is smaller near the inner part of the media than the outer part. As a result, the amount of data you could store in the drive was limited by how dense you could pack the bits near the inner part of the media, and some capacity was always wasted in the outer part of the media.) Also, you are correct in implying that the state of the art has improved in making each bit smaller on the media. But improving drive performance is like that old saw that a chain is no stronger than its weakest link. To go along with improvements in media recording densities, drive manufacturers also had to improve the DMA transfer rate of the drive interface by using buffers and faster electronics. Computer manufacturers had to improve their buses to handle the faster rates. There is a big difference between an XT bus and an EISA or Zorro bus. Currently, the raw performance of a single drive subsystem is still limited by the media transfer rate because of the mechanical nature of that process versus the electrical nature of everything else involved. JVC, however, supposedly has a 3 1/2 inch drive which can transfer data from the media at very high transfer rates, on the order of 15 megabytes/sec. If this is so, then even SCSI-2 would be a bottleneck.
daveh@cbmvax.commodore.com (Dave Haynie) (05/25/91)
In article <42601@cup.portal.com> FelineGrace@cup.portal.com (Dana B Bourgeois) writes: >First, I am talking about a SPARC 2 so I don't know if that is considered >a low-cost workstation or not. You didn't mention it when listing a few >models so I'd guess not. SUN prints numbers like 2-3 MB/s for it and >claims it is SCSI 2. (oops, they claim 1.8 MB/s, 3 MB/s synchronous). I haven't seen any numbers for a SparcStation 2. You have to be a little careful about what the vendor of a system says (of course you won't notice that I'd the "vendor" here), versus what a reviewer actually measures in a review. Unless they have an asterisk next to the numbers, and you read at the bottom of the page something like "...as measured with [insert favorite disk speed measurement tool here]". >Second, after I posted my qustions I received the latest Seagate catalog. >They have some 5400 RPM SCSI-2 drives they rate at 28MB/s internal >transfer and 10MB/s external transfer. Do you think they are saying >the drive has the capability to transfer data+clock bits at 28MB/s >and the SCSI protocol is holding it back to 10? (I ask Seagate and >their sales people won't speculate.) My guess is that they can grab data from their on-drive cache at 28MB/s, and send it over at the full SCSI-2/8-bit speed of 10MB/s. I don't believe for a second they get sustained 28MB/s or 10MB/s from the drive itself. >If sustained transfer rates are increasing from 500KB/s to 2MB/s then there >must be more bits per track since the fastest spinning drives are increasing >RPM by only 50%. The trick is, the fastest and most expensive drives are also big. So rather than a single or a pair of heads as in your lower density Quantums, they may have 6 or 8 heads, maybe even more (I don't know the typical number these days on a big drive). The faster rotation gives them a little better per head speed and a little less rotational seeking latency, but they really get speed with more heads. And the caches that most of them have. Regardless, I don't think you're going to get close to saturating SCSI off-peak without multiple drives, no matter who's making them these days. >Dana Bourgeois @ cup.portal.com -- Dave Haynie Commodore-Amiga (Amiga 3000) "The Crew That Never Rests" {uunet|pyramid|rutgers}!cbmvax!daveh PLINK: hazy BIX: hazy "That's me in the corner, that's me in the spotlight" -R.E.M.
jesup@cbmvax.commodore.com (Randell Jesup) (05/25/91)
In article <21894@cbmvax.commodore.com> daveh@cbmvax.commodore.com (Dave Haynie) writes: >In article <42601@cup.portal.com> FelineGrace@cup.portal.com (Dana B Bourgeois) writes: >>First, I am talking about a SPARC 2 so I don't know if that is considered >>a low-cost workstation or not. You didn't mention it when listing a few >>models so I'd guess not. SUN prints numbers like 2-3 MB/s for it and >>claims it is SCSI 2. (oops, they claim 1.8 MB/s, 3 MB/s synchronous). SCSI 2 doesn't really mean much. The "good stuff" in SCSI 2 is mainly the 10MB/s "Fast" synchronous mode. For a host adapter, SCSI 2 often merely means that you don't use any non-SCSI-1 commands that aren't in SCSI-2. (i.e. you use don't use anything "funny", and may use one or two new SCSI-2 commands on rare occasion.) We could probably call the 590/2091/3000 scsi a SCSI-2 controller by that definition, though we don't. SCSI-2 can also mean that it's a SCSI-1 controller with a SCSI-2 connector. BTW, 1.8 asynch, 3.0 synch is NOT fast for a modern SCSI controller. Really old ones do 1.5/NA, old ones do 3.0/5.0, and current ones do 5.0/5.0+, and SCSI-2 ones do ~5.0/10.0. (Taken from the NCR scsi chip family - the WD 33c93a chip is around 3-4.0/4.0 I think, the 33c93b is 4.0/10.0.) >>Second, after I posted my qustions I received the latest Seagate catalog. >>They have some 5400 RPM SCSI-2 drives they rate at 28MB/s internal >>transfer and 10MB/s external transfer. Do you think they are saying >>the drive has the capability to transfer data+clock bits at 28MB/s >>and the SCSI protocol is holding it back to 10? (I ask Seagate and >>their sales people won't speculate.) > >My guess is that they can grab data from their on-drive cache at 28MB/s, and >send it over at the full SCSI-2/8-bit speed of 10MB/s. I don't believe for a >second they get sustained 28MB/s or 10MB/s from the drive itself. 28MB/s includes clock bits, etc, etc. My 400+MB, 4400rpm 3.5" Seagate does 2MB/s sustained on a 3000 through the FS on the outer tracks, 1.5MB/s on the inner tracks. (Note that on the Amiga, max filesystem rates are essentially device rates.) It's a SCSI-2 drive (will do 10MB/s), but that's close to useless to me, since its sustained rate is 2MB/s max. The main putpose of the high xfer rate is for when the drive already has the sector you need in cache, and so you can run more drives on the same bus at their sustained rates. It's absolute max average is 2.1 MB/s at the interface. We've run one of these under Diskspeed at 90+% of maximum speed while running a Quantum 210S at 90+% of its maximum, at the same time (total xfer rate: ~2.8 MB/s) on an A3000. >The trick is, the fastest and most expensive drives are also big. So rather >than a single or a pair of heads as in your lower density Quantums, they may >have 6 or 8 heads, maybe even more (I don't know the typical number these days >on a big drive). The faster rotation gives them a little better per head >speed and a little less rotational seeking latency, but they really get speed >with more heads. They also get speed with more sectors/track and zone recording (at least on the outer tracks). Apparently they get up to 64 sectors/track. -- Randell Jesup, Jack-of-quite-a-few-trades, Commodore Engineering. {uunet|rutgers}!cbmvax!jesup, jesup@cbmvax.cbm.commodore.com BIX: rjesup Disclaimer: Nothing I say is anything other than my personal opinion. "No matter where you go, there you are." - Buckaroo Banzai
mscritsm@isis.cs.du.edu (Milton Scritsmier) (05/25/91)
In article <21894@cbmvax.commodore.com> daveh@cbmvax.commodore.com (Dave Haynie) writes: >The trick is, the fastest and most expensive drives are also big. So rather >than a single or a pair of heads as in your lower density Quantums, they may >have 6 or 8 heads, maybe even more (I don't know the typical number these days >on a big drive). The faster rotation gives them a little better per head >speed and a little less rotational seeking latency, but they really get speed >with more heads. And the caches that most of them have. Regardless, I don't >think you're going to get close to saturating SCSI off-peak without multiple >drives, no matter who's making them these days. Actually, the number of heads has very little to do with data transfer speeds. A track to track seek is about 2 or 3 ms, and this is much less than the average rotational latency of 6 to 8 ms that happens when the drive waits for the data to come under its head after it has seeked to the proper track. Thus having more heads to reduce the number of seeks doesn't buy you that much on a given command. Plus, most SCSI drives have 40 or more 512-byte sectors per track. That's at least 20K on each track, and I doubt that a lot of commands are that big. As for caches, there are times when it will help, but by and large the greatest factor in drive performance is such things as how fast the drive spins and how fast it can seek. For example, the old standard for rotational speed was 3600 rpm. Now we are seeing some drives at 4500 rpm or more. That's at 25% increase in speed, even if the data was recorded at the same density. Add the increases in recording density that are in fact taking place, and we are seeing increases from 10 megabits/sec a couple of years ago to 24 megabits/sec or more today.