• Please review our updated Terms and Rules here

questions about SCSI hard drive size and speeds.

oblivion

Veteran Member
Joined
Aug 28, 2010
Messages
1,003
Location
Apache Junction, AZ
lately I've added a SCSI hard drive to my 486. I actually had some diffidently in booting from it so I opted to just have it running as the D: drive along side of my IDE drive. this works fine for me.

but I have more questions about some odd things i've noticed. My IDE drive is a quantum fireball of some 20+gb and it is detected as 504mb which I know is a BIOS limitation. I though from my research that that limit when it came to SCSI and these old machines was 2gb but for some reason all the SCSI drives I try are detected at random sizes. they are all larger drives from 9 to 18gb and were not pre set up with partitions. I had then formatted and hooked them up. for instance the 18gb is detected as something like 750mb while the smaller 9gb drive is seen as 1.40gb

also according to speedsys the SCSI drives are by far slower then my IDE drive when I assumed they would be equal or faster.

the IDE as I said before is a quantum fireball 72000 rpm drive of 20+gb connected to the on board IDE controller its from around the year 2000. The SCSI drives are IBM and from around the same year. the IDE drives scores well over 1000 on the speedsys hdd test. close to 1500 whiles the 72000rpm SCSI drive scores about 400 and the 10000rpm does about 100 better at 500. I thought maybe because the detected size of the SCSI drives is a little bigger or because they are running through the VLB bus that that is making them be detected slower but almost a 1000 points slower?

its more of a curiosity question then anything as 1.4gb plus my 500mb drive is more then enough space for me and I don't really notice a speed issue when running either drive but I am rather curious.
 
if you look up the drive spec, there should be a field for media transfer rate. that is the real media transfer speed on and off the rotating media. the rest is usually on drive caching.
 
The transfer rate will be capped by the speed of the controller. A VLB SCSI controller for a 486 might well only support Fast SCSI or early Ultrawide so peaking at 10 or 20 MB/s. The drives are probably SCSI III U-160 or U-320 so at best 92% of the drive's performance is thrown away because the controller is so slow.

The big advantage of SCSI should be much lower CPU utilization during disk transfers.
 
I'm certain that you don't mean 72000 RPM--you'd be picking bits of disk drive out of your hair if that were the case. 7200RPM.

Narrow SCSI vs. wide SCSI matters, controller cache, read-ahead, etc. also matter. If you can boot from a different drive, you might get best performance by using the ASPI driver for the controller and then an ASPI Disk driver layered on that.

I'm running a 486 on an Adaptec 2840 with 8GB--and a 9GB IBM drive on a 386 with a DTC (mumble) controller, again with no issues. No special drivers involved.
 
Ha, yhea I ment 7200. I'll try and post more specific info after I get off work including perhaps some speedsys result screenshots. Its a 486. Even with most of the performance lost via the SCSI controller card the on board IDE controller is no newer and both hard drives are from roughly the same year so I'm still not sure why the IDE drive is so massivly faster.
 
The transfer rate will be capped by the speed of the controller. A VLB SCSI controller for a 486 might well only support Fast SCSI or early Ultrawide so peaking at 10 or 20 MB/s. The drives are probably SCSI III U-160 or U-320 so at best 92% of the drive's performance is thrown away because the controller is so slow.
Ahm, no. the media data rate is generally substantially slower than electronics. that 20 or 40 MB/s (Ultra SCSI/wide) you're quoting is all electronics. barring some very cleaver caching algorithm (or just very large on-drive cache) a drive in 486 era be pretty damn impressive if pulling 8~10 MB/s. I think I benched Quantum Viking II pulling 10MB/s off media around 1997, and that was purportedly an "enterprise class" scsi drive.
The big advantage of SCSI should be much lower CPU utilization during disk transfers.
that's true if the SCSI controller supports DMA. (not all SCSI cards from that era does that.) that advantage became less of an issue when IDE started supporting DMA.
Where SCSI really shined is when the host is running multi-threaded OS (Linux, Unix, WinNT etc) where the host can issue multiple cmds at once using cmd taq queuing. The drive firmware can intelligently sequence the cmds (since OS mostly uses simple queuing) for better performance. IDE on the other hand can process just one cmd at a time.
 
Last edited:
There were also more than a few ISA SCSI controllers with large amounts of RAM for caching. CSC FastCache is one--I think it could use up to 10MB.
 
There were also more than a few ISA SCSI controllers with large amounts of RAM for caching. CSC FastCache is one--I think it could use up to 10MB.

would an Isa controller with cache be faster then a vlb controller without? The vlb controller in using is a SiiG. So under DOS as far as speed is concerned an equel quality IDE and SCSI drive should be roughly equel in speed?
 
That's a tough one. VLB has a higher bus speed than ISA. In other words, no matter how much data is ready to push through a pipe, it's the size of the pipe that's going to matter. Caching helps even things out, but on bus-limited tasks, there's no question which will win.

On the other hand, 16-bit IDE will almost always beat out 8-bit narrow SCSI.
 
Last edited:
The correct answer to your question is, it depends ...

There are a few factors to consider:

  • Drive mechanics: How fast does it spin and how fast can it seek. On a sequential test the RPM is the important factor. On a random test the seek speed is the important factor.
  • How wide is the transfer bus to the host machine. Wider is better. 16 bit IDE is, well, 16 bits. Narrow SCSI is 8 bits while wide SCSI is 16 bits.
  • How smart is the drive. Given two drives with similar mechanics the firmware becomes the deciding factor. Not all firmware is created equal. Enterprise class drives (which used to be SCSI and are now SAS) generally have better/smarter firmware and faster electronics. Cache memory on the drive can be a big factor depending on the test.
  • How smart is the controller and software. The worlds fastest SCSI drive running on a crappy controller where the CPU does every byte transfer is going to be slower than a crappy IDE drive. Besides just having a good controller, the features have to be turned on. If your controller isn't using DMA and bus mastering, you are not in the game.
  • The advanced features of the drive are worthless if they are not used. SCSI drives generally supported tagged command queueing, which improves overall throughput but can hurt the latency of individual requests. A benchmark that sends one command at a time won't make use of that feature, but in the real world with a decent OS that feature will be used.


-Mike
 
okay, the motherboard is a UM 486V AIO motherboard. The on board IDE controller is detected as a ST506. the IDE hard drive is a 20.5gb quantum fireball plus AS. as for the SCSI drive I don't really want to open the PC up to look at it at the moment but is was from about the same time period, 7200RPM and IBM maybe a desk star. I went with it over the 10,000rpm drive I had because its whisper quiet and the other sounds like a jumbo jet firing up for lift off.

the SCSI controller is this
6fw6.jpg


here are the speedsys results

IDE drive
vpai.jpg


SCSI drive
ch8w.jpg
 
VL Bus SCSI should be cable of bus mastering with 32 bits per cycle being transferred.

Are there drivers loaded for the SCSI controller, or is it just doing stupid BIOS level compatibility? What about the IDE drive - is it using DMA or PIO?
 
VL Bus SCSI should be cable of bus mastering with 32 bits per cycle being transferred.

Are there drivers loaded for the SCSI controller, or is it just doing stupid BIOS level compatibility? What about the IDE drive - is it using DMA or PIO?

no drivers for the SCSI so just BIOS level stuff. as for the IDE drive.....i have no idea. on the speedsys screen is says PIO 4 UDMA5 above the hdd test but other then that i don't know.
 
Looking at the card, it seems as if it was designed to be an IDE controller first, with a SCSI controller added as an afterthought. I don't see much there.

On the other hand, look at the Adaptec AHA2840VL controller:

$(KGrHqVHJEwFERf0+IZsBRG4sIQM(w~~60_57.JPG


Still narrow SCSI; the only addition is the FDC connection, which involves the very top row of ICs near the floppy connector. Look at the number of packages on the rest of the card--for handling SCSI transfers.

My thought is that the SCSI part on your card was intended for low-demand devices such as CD-ROMs and scanners. Do you have the model number? That might cinch it.
 
His card is an IDE/SCSI/Floppy/I/O card but with just the SCSI section installed so there is a bunch of empty places for components. I have an Always card like that but it is full of chips.

VLB will always be faster then ISA just because of the bus bandwidth (especially 8 bit ISA).
 
I have a adaptec card same model as chuck(g) but Unfortunitly I think the chips are corrupt as it will not properly detect any drive attached. There's no model number on the SiiG card at all just that FCC id. I'd be willing to possibly grab a diffrent card if I can nab one cheap enough if you guys really think the issue may be a low end controller.
 
I have a adaptec card same model as chuck(g) but Unfortunitly I think the chips are corrupt as it will not properly detect any drive attached. There's no model number on the SiiG card at all just that FCC id. I'd be willing to possibly grab a diffrent card if I can nab one cheap enough if you guys really think the issue may be a low end controller.

Can your SCSI drive do 16-bit (68-position connector)? That would really improve things if you were to find a controller to match. If you're interested in a 2840VL, I can benchmark my drive as comparison. What are you using for a benchmark program?
 
Back
Top