Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!shadooby!accuvax.nwu.edu!tank!uxc!iuvax!rutgers!apple!oliveb!amiga!cbmvax!steveb
From: steveb@cbmvax.UUCP (Steve Beats)
Newsgroups: comp.sys.amiga
Subject: Re: DiskPerf results
Message-ID: <6720@cbmvax.UUCP>
Date: 1 May 89 13:54:11 GMT
References: <21149@srcsip.UUCP> <7965@killer.Dallas.TX.US>
Reply-To: steveb@cbmvax.UUCP (Steve Beats)
Organization: Commodore Technology, West Chester, PA
Lines: 41

In article <7965@killer.Dallas.TX.US>  writes:
>
>I suspect that the limiting factors are a) the file system, and b) the
>A2090A. I got about the same results from a 2090A and a Quantum 80S.
>The 2090A limits speed because it is limited to the old 1mb/sec
>asynchronous SCSI protocol (meaning that however hot your drive is,
>it's limited to 1mb/sec max throughput). However, someone else got
>basically the same numbers from a Quantum 80s and a Microbotics
>Hardframe (which has the new 5mb/sec synchronous SCSI, i.e., will take
>data as fast as the Quantum can dish it out, 4mb/sec). So I have to
>conclude that file system overhead has something to do with it,
>too....
>
>--
While it's true that the filing system will eat some of the performance from
a SCSI drive, it's not reasonable to expect a synchronous mode controller to
run at 5.0Mb/s.  Synchronous mode is relatively new (to controller vendors,
not the SCSI spec) and as such it can become quite a pain to support.  I
attempted to bring up a driver with full sync xfer support and found that a
LOT of drives simply crashed when given a sync xfer request.  For this reason
I had to stop initiating the message and just handle it when the target sent
the request first, (an extremely rare occurence).

Even if you get a drive that can support full speed synchronous transfers, you
are NOT going to get 5.0Mb/s transfer rates.  Most drives can pull data off
the media at 1.25Mb/s at 1:1 interleave.  That is the maximum throughput you
are going to get.  The only time that you can up this figure is if you have
multiple units on the controller, all running synchronous mode and all
supporting full disconnect reselect protocols.  In this scenario you do want
to burst data out of the drives internal cache buffer as fast as you can.
While you're bursting out one batch of data, another drive can be filling
it's buffer ready for when it reselects the initiator.

Even given a bunch of targets (say 4) all doing full speed sync transfers, you
will still be limited to the Amigas bus speed.  With a 16 bit DMA engine on
a stock a2000 this will come out to around 3.58Mb/s.  That's the ABSOLUTE
maximum data throughput you are going to get no matter how fast the drive
can supply data.  The fastest drive I have found to date is the CDC Wren V
That baby can sustain 1400K/s direct from the drive.  

	Steve