[rescue] Sun Blade 100 Question - IDE

[Brian Hechinger]

On Sat, May 11, 2002 at 12:04:14AM -0500, Patrick Giagnocavo wrote:
> 
> You gotta be careful when you say those kind of things while defending
> SCSI.

greg already got this, but let me toss my $0.02 in.

> What about the way you can blow up stuff by mis-matching HVD and SE
> SCSI?  Why did they keep the same connector - they could have
> easily keyed it somehow, or put in a small CPLD to figure out if
> something wasn't right and avoid damage.

this is not the fault of the specification, this is the fault of the
implementation.  DIFFSENSE is part of the SCSI spec, when a DIFF controller
doesn't find DIFFSENSE it's supposed to shut down the channel.  why does
no-one implement this?  beats the hell out of me, wish i knew.

> What about the original 8-bit SCSI spec?  What about drive size
> limitations?  

hey, i think they did pretty good for a first try.  and it didn't take very
long for SCSI-2 to come out and solve that particular issue among others.

> Go ahead, plug in an earlier SCSI device to your U160
> device chain and watch performance go down...

what do you expect?  it's a closed electrical system.  you can only push it as
hard as the lowest common denominater can handle.  that's the way it works.  if
you are that concerned about performance, you shouldn't be plugging earlier
devices into U160 and expecting the world.

> In theory, SCSI is backwards compatible, but the price to convert
> between the various interfaces via cabling or tailgates often does not
> make sense from a cost standpoint.

no, i won't argue with you there.  wish there was a resonable explination as to
why all the different connectors were used.  it's a silly industry.  that's the
only explination i can come up with.

> IDE at least will never blow up stuff, is completely backwards
> compatible (aside from a few drive issues when putting an old Conner
> and old Seagate on the same channel), and the connector hasn't changed.

yeah, but if you connect two modern disks on the same channel you pay a
terrible penalty in performance.  which somewhat invalidates your argument
about plugging old devices into U160.  you don't even need to use an older
disk to destroy your performance. :)

> Connectors for SCSI include:

yeah, silly.  but then again, it's a silly industry.

> I am not flaming you, just pointing out some stuff that shows that
> SCSI isn't perfect either.  And many early SCSI devices are not
> exactly compliant.

just like any other industry (even outside the computer one) there are
specifications and there are implementations.  for some reason, all sectors
of all industries like to play the fucked up proprietary game, which usually
includes mucking up the implementation somehow.  this is barely the fault of
SCSI.  i'm only guessing that IDE sucked enough to start with that no one felt
it was nessesary to go outside of the specifications. :)

> Let's just say that current SCSI has advantages for multiple drives,
> for hot-swap (in terms of implementations, it is easier to find
> hotswap SCSI than hotswap IDE) and for some other stuff.

true, true.

> Let's say that IDE has advantages in terms of price/performance and

hardly.  the performance (as greg pointed out) blows.  unless you are doing
one small percentage of a particular type of access.

> generally higher storage capacity (higher capacities come out first in
> IDE format due to SCSI user's emphasis/preference for IOPs).

and then there is always the price issue.  i'm not looking for cheap U160,
even bottom of the barrel LVD would be good.  but make it cheap for christ's
sake.  oy!

but if you want to talk about technical points, i wish SCSI had died out to
make room for DSSI.  now *that* would have been cool.

-brian

*disclaimer* i'm not attacking you patrick, but please don't defend IDE. :)
-- 
"when it absolutely has to be there oh, sometime in the next few weeks"