[Info-vax] OT: news from the trenches (re: Solaris)

[Scott Dorsey]

 <johnson.eric at gmail.com> wrote:
>
>I don't see how you can get around that. Even a freshly crafted instruction set
>(this time done right! *pounds fist on the table*) will have the same kind of warts
>after 30 years of kicking ass and taking names. That's even assuming your better
>instruction set lasts that long. 

The problem with the x86 instruction set isn't that it's been around for a
long time and had a lot of cruft added to it... the problem is that it was
not really very good in the first place.  If anything, the attempts to add
stuff have improved it substantially.  But it's still just not very efficient
an instruction layout, compared with something like the Alpha.

Still, it is what everybody uses, so we have to use it too.  Life is like that.

>I'm sure even the designers of Itanium thought they had it right too.

Early on, they did.  But as they ran more and more simulations it became
clear that cache management was becoming a real problem, and as the design
evolved they just kept adding more and more cache to the thing in an attempt
to deal with the inherent problems.  By the time the first samples started
coming out, it was clear that it just wasn't going to hit the numbers they
were aiming for.

Which... more or less was the same way that iAPX 432 worked, except that
first of all simulation was not as well developed at the time so it took
a good bit longer before they realized it was going to be a pig, and 
secondly they dropped the design before it actually got to the sample stage.
--scott


-- 
"C'est un Nagra. C'est suisse, et tres, tres precis."