Newsgroups: comp.sys.ibm.pc.hardware
Path: cantaloupe.srv.cs.cmu.edu!das-news.harvard.edu!noc.near.net!uunet!uunet.ca!torn!newshost.uwo.ca!valve.heart.rri.uwo.ca!wlsmith
From: wlsmith@valve.heart.rri.uwo.ca (Wayne Smith)
Subject: Re: IDE vs SCSI
Organization: The John P. Robarts Research Institute, London, Ontario
Date: Sat, 17 Apr 1993 17:48:27 GMT
Message-ID: <1993Apr17.174827.24434@julian.uwo.ca>
Sender: news@julian.uwo.ca (USENET News System)
Nntp-Posting-Host: valve.heart.rri.uwo.ca
Lines: 141

In article <C5LKEv.HpJ@feanor.xel.com> shaw@feanor.xel.com (Greg Shaw) writes:
>: Why don't you start with the spec-sheet of the ISA bus first?
>: You can quote SCSI specs till you're blue in the face, but if they
>: exceed the ISA bus capability, then what's the point?
>
>Who said ISA was necessary?  EISA or VLB are the only interfaces worth
>investing thousands of dollars (e.g. a new pc's worth of money ) in .

Then don't complain (maybe it wasn't you) that SCSI was so expensive on
PC's because all we've had until a year or two ago was the ISA bus.
(ie no one buys SCSI for ISA because ISA is slow)
Are you saying that SCSI on an ISA bus is not an automatic winner when
compared to IDE?

>You didn't read to carefully.  VLB-IDE uses the same connection mechanism
>as standard IDE.  If transfer rate is limited by IDE, whether it's
>interfaced to ISA, EISA or VLB matters not.

I get different transfer rates out of my IDE when I change my ISA bus speed.

>On mine, for one thing.  SCSI blows IDE out of the water, hands down.  If
>IDE has better throughput, why isn't it used on workstations and file
>servers?  

IDE is just a variant of the old IBM- MFM AT controller.  (at least that's
how it looks from a software point of view).  It was never meant to be
an all-encompassing protocal/standard to be implimented across different
platforms.

Is there any argument that 
IDE can (or can't) transfer data from the IDE drive at least as fast as the
drive is able to provide the data?  Are SCSI versions of IDE drives able
to deliver higher sustained transfer rates to their SCSI interface (because
of a higher RPM platter, different arrangement of heads, etc?)?

>: Given the original question (SCSI used only as a single hard drive
>: controller),  is it then necessary to get a SCSI drive that will do
>: at least 5, maybe 10 megs/sec for the SCSI choice to make any sence?
>: What does a 200-400 meg 5 megs/sec SCSI drive cost?
>
>No, that's the nice thing -- on a multitasking OS, SCSI can use both drives
>at once.  I've got unix loaded on one of my pcs (along with windogs) and the OS can only use one of the two IDE drives at one time.  It's pretty ugly.

If data is going from one drive to another, and if SCSI has the ability to
perform that transfer without the data having to go through the CPU or main
memory, then yes, that is the optimal way to do it.  As far as I know, IDE
can't do that.  But when the CPU wants data from both drives (data to be stored
in main memory) are you saying that SCSI can grab data from both drives 
at the same time *and* store/transfer that data to main memory also at the
same time?  Working off 1 IRQ and 1 DMA channel on an ISA (or whatever) bus?

>I just bought at Quantum 240 for my mac at home.  I paid $369 for it.  I
>haven't seen IDE drives cheaper.

A friend of mine just got a Maxtor 245 meg IDE drive for $320.  (that's 245
million bytes, or 234 mega-bytes).  With the basic $20 interface, he gets
close to 1 meg/sec transfer on his 286-20.  Does your figure include a few
hundred $$$ for SCSI drivers?

>No, actually, we're talking about SCSI being expensive simply because
>nobody did a common interface for the PC.  If they had a common (read:
>easily implemented) method of adding scsi to a PC (like as in a Sun or
>Mac), then you'd find SCSI the connection medium of choice.

So you're saying that SCSI would have been the default interface type,
considering that the vast majority of PC's don't have cd-rom drives or
tape backups or etc?  That most PC's only have (or had) 1 hard drive and
run DOS?  That SCSI hard drives cost a lot more than MFM or RLL drives
at the time?  (and how common were SCSI drives under 80 megs 4 to 10 years
ago?)  There's a lot more than the lack of a common interface card that
prevented SCSI from becoming the connection medium of choice.

>: I won't argue that the SCSI standard makes for a good, well implimented
>: data highway, but I still want to know why it intrinsically better
>: (than IDE, on an ISA bus) when it comes to multi-tasking OS's when
>: managing data from a single SCSI hard drive.
>
>On a single drive, SCSI is more expensive.

But on that point, is it faster?  This is what all this is about.  Do you
get more performance for the money.  For all the people that will only have
a single hard drive in their system (regardless of the OS) will the SCSI
choice really give them more performance than IDE?

>But, you bought your PC for
>expandibility, so, you'd want to add more drives or whatever.

True, but expandibility can also start on the bus, which means the option
is there for cd-rom drives or tape backups that run off their own cards.

>	1.  You can add many different types of devices and access them 
>	concurrently.

No argument.  This is always held up to the first time SCSI buyer as the
best reason.  But how many SCSI devices will the first time SCSI buyer
eventually acquire?  Again does it make sense to go SCSI for a single
hard drive system?

>	2.  A SCSI device works on many different machines (I have a mac
>	and a PC at home and moving hard drives between them is VERY nice
>	with SCSI -- hook them up and away they go)

With all the postings on the SCSI I or II specs, are you really sure that
PC and Apple SCSI hard drives are compatible?  And even if they are, 
is the data accessible from either machine (ie are there no formatting/
partitioning or file table differences?)  Is it really plug'n'play?

>	3.  SCSI devices work together better than IDE devices.  For
>	instance, recently, I added an older connor 100 meg IDE to a maxtor
>	212 meg IDE.  The connor *MUST* be setup as the slave.  It will
>	work no other way.  On SCSI, you set the address, check the
>	termination, plug it in, and away it goes.

So the C: drive on the connor becomes a logical D: drive to DOS.  Is this
really a problem?  

>	4.  I have a problem with IDE's mutual exclusion - I notice that
>	the time it takes to switch from accessing drive c: to drive d: is
>	quite long as compared to the time it takes to switch from drive c:
>	to d: on a SCSI system.  Under a multitasking OS, this is very
>	noticable, as many things can be going on at once.

After having two IDE drives in my system for temporary file transfers,
I have never seen any differences when switching between drives, nor
have I ever seen any differences when transfering files between drives or
to/from the same drive.

>One neat thing that I've noticed lately (a fringe benefit) has been the
>ability to add older (almost dead) drives as storage on a SCSI system with
>little problem -- we've got a bunch of almost dead 20 meg drives that I've
>added to my PC.  I've now got the interface full, but, it does allow me to
>have 4 20 meg drives, 1 240 meg drive, 1 tape drive, and 1 105 meg drive
>all on the same card.  

That is nice (as long as the power supply can keep up).  I do believe that
there is the possibility for up to 4 IDE drives on a PC.

>Simply put, SCSI is handier than IDE.  No mysterious jumpers to figure out.

But what about "mysterious" (and expensive) drivers to figure out?  At least
IDE doesn't require drivers that consume precious conventional (DOS) memory.
