Return to BSD News archive
Newsgroups: comp.os.386bsd.misc
Path: sserve!newshost.anu.edu.au!harbinger.cc.monash.edu.au!msuinfo!agate!tfs.com!julian
From: julian@tfs.com (Julian Elischer)
Subject: Re: SCASI or IDE disk?
Message-ID: <CuLo9s.4pA@tfs.com>
Organization: TRW Financial Systems, Oakland, CA
References: <salem.136.2E48D0EF@hauk.hsr.no> <CuJL5C.z7M@yuma.acns.colostate.edu>
Date: Mon, 15 Aug 1994 23:37:04 GMT
Lines: 95
In article <CuJL5C.z7M@yuma.acns.colostate.edu>,
Sandy Lam <sandylam@lamar.ColoState.EDU> wrote:
> well, i currently am getting 15 megs/sec on my IDE controller.
> It's also DMA bus mastering. And the next level of IDE (ATA-3)
> will blow away this one in terms of performance and be 100%
> compatible. It's PCI btw, and relocatable so each IDE controller
> chip handles 4 devices and does not conflict with any others.
Well, I could just as well theat my VW goes really fast with a V8..
a Bus-mastering IDE controller eh? ok, so how do I get my standard
controller drivers to use this? Obviously I can't.
I understand that the disk interface in no way decides the bus interface,
but let's be realistic here..
1/ how do you get 15MBytes per second OFF a normal drive?
if you work out the rates of bits under the head, most good drives are
getting 90 revs/s * 50kB/track (80-100 sectors)
which is 4.5MBytes/second. you'de need multiple heads.. (some expensive
mainframe drives have this.. (not many IDE PC drives).)
of course you could be caching but we'll assume you know better than to
raise that one.... (I've seen drives with 120r/sec but that's only 6MB/sec)
(and that doesn't count stepping time)
2/ how do I put this card into a system and (without a special driver or BIOS
prom) get it to be useful. Obviously I need a driver for this board. Standard
IDE (ATA ok?) used programmed I/O so a bus-mastering interface would have
to be RADICALLY different.. I can't see this controller board
being a lot cheaper than a SCSI board, (it has equivalent complexity)
and how do I plug all my scsi devices (from the office and from junk sales
e.g. scsi scanners) into it?
I can see that by making a very intelligent design, and cutting out all
the delays, you can get IDE drives to be an efficient bulk storage
system with the speeds SCSI gives, (though I'd like to hear how you get
the rates you claim (I'll allow you might briefly BURST that speed
if you have been readingthe track into ram, but it's not sustainable from
a single head) (with today's PC cost hardware)
>
> also btw, Western Digitals suck rocks at high end performance.
> best i can get out of those puppies is 7 megs/sec.
explain please..
If this is sustained throughput, please let us in on the secret
>
> the downside? i'm developing for these chips, and they're not out
> yet. but soon they'll be integrated on MANY major brand name
> motherboards.
ah yes the "special interface"
We'll write special drivers to use them under Free/Net-BSD as soon as
they appear, and the specs become 'public' (they are going to do that right?)
but don't expect them to be wildly popular or even 'widely available'
the standard IDE interface will rule the IDE drive world for a while.
I think we can consider this a 'specialty product' along with SCSI,
and given the choice, I'd go for SCSI.. as I said, I have
SCSI devices already and I can borrow SCSI devices from the suns
at work etc. I think it sounds great, but ...
I guess the secret's in the BIOS drivers that will be able to use them,
If we can right drivers for these boards (hey you must be volunteering right?)
then they will be a useful addition to out system.
The will suffer none of the present IDE draw-backs (CPU hogging etc.).
But they hardly enter into the discussion of "what drive should I buy"
(SCSI or (standard)IDE)..
>
> not that this really helps solve that one dude's problem,
> but i just had to stick up for IDE. lots of ppl have been
> dogging it hard core saying it sucks performancewise and can't
> possibly ever be decent. btw, there are IDE CD ROMs and soon
> IDE ethernet controllers.
Hey there's nothing wrong with the IDE protocol as transmitted across the
ribbon cable.. it's the way it's "usually" controlled..
i.e. by the CPU directly, that's the problem. and if you are going to buy
a more expensive IDE controller, then why not get a SCSI controller
that's a whole lot more flexible and useful..
>
> SCSI's got a lot of hype going for it, but don't think IDE is
> going away any time soon.
no, but then remmeber that most IDE implimentation will continue
to eat CPU cycles as well. And I've never seen an IDE tape jukebox..
>
julian
p.s. I REALLY am interested in hearing more about what EXACTLY your figures
represent.. (remember no caching, just track buffering on the drive allowed) :-)
+----------------------------------+ ______ _ __
| __--_|\ Julian Elischer | \ U \/ / On assignment
| / \ julian@tfs.com +------>x USA \ in a very strange
| ( OZ ) 300 lakeside Dr. oakland CA. \___ ___ | country !
+- X_.---._/ USA+(510) 645-3137(wk) \_/ \\
v