Thanks for the reply, I tend to agree (like most modern UNICES). I do
wonder why my kernel is over 7.5 Mg though?

SkyWriter wrote:
> Spidy wrote:
>
>
>>Hello, just curious. Is IRIX considered to be a monolithic OS, or does
>>it use a microkernel?
>
>
> i would be suprised to hear another thing other than monolithic (at least
> compared
> to microkernel, not monolithic in the literal sense since it's obviously
> modular).
>

In article <3f442ff6_1@news.iprimus.com.au>,
Spidy wrote:
:Thanks for the reply, I tend to agree (like most modern UNICES). I do
:wonder why my kernel is over 7.5 Mg though?

Depends on your version. Look at the size of the objects
in /var/sysgen/boot/ . The two biggest on the systems I looked
on were os.a and xfs.a .
--
"No one has the right to destroy another person's belief by
demanding empirical evidence." -- Ann Landers

Hi,

> :Thanks for the reply, I tend to agree (like most modern UNICES). I do
> :wonder why my kernel is over 7.5 Mg though?
> Depends on your version. Look at the size of the objects
> in /var/sysgen/boot/ . The two biggest on the systems I looked
> on were os.a and xfs.a .

I noticed the kernel on my Indy (R5@180) measures in at about 6MB.
This I thought was really quite a bit much. For a test, I made a
backup of it and stripped it. This wasn't so great as whilst the
machine boots you loose things like VINO support.

Is there any way in the base IRIX operating system of making

a) the kernel smaller
b) the kernel MIPS4

Perhaps something in /var/sysgen or some sort of switches ?

It actually a point that the whole of IRIX 6.5 is MIPS3 which is fair
enough, given the range of MIPS3 and 4 hardware it runs on, but how
much performance is actually being lost running IRIX 6.5 on MIPS4
hardware ? Especially given various peoples views on the benefits of
MIPSPro and the MIPS4 binaries.

SGI are raising the support platform to only MIPS4 based machines as
of 6.5.23, in so far as bugs-being-fixed etc but without actually
making an IRIX release[1] *for* MIPS4 how much of a disservice is
being done ?

[1] I'm assuming without prior knowledge that SGI wont make a MIPS4
only IRIX 6.5 because that would break software-certification and
force deprecate hardware support beyond IRIX 6.5 general support
remit.

Cheers,

Rob

In article ,
Rob wrote:
:Is there any way in the base IRIX operating system of making

:a) the kernel smaller

Smaller in what sense? Your reference to stripping the kernel
suggests to me that you are talking about the size of /unix .
Does it really matter what size /unix is, though, when one
considers that a lot of kernel activity is broken out to
DSO's? And does it matter whether /unix is 1 Mb or 4 Mb when
one considers that the kernel can malloc large gobs of memory
for kernel purposes?

Having a smaller /unix does not imply that the kernel would be
more efficient. Remember that one of the important tricks in
code optimization is to identify the patterns of code that are
going to be executed most often and make that code path as
tight as possible, copying out sections of the code for the
less-common paths. The code X; if (A) {Y}; Z; if (A) {W}
might be more efficient as if (A) {X; Y; Z; W} else {X; Z}
which is larger. Code optimization is full of time / space
tradeoffs.
--
100% of all human deaths occur within 100 miles of Earth.

In comp.sys.sgi.admin Walter Roberson wrote:
> In article ,
> Rob wrote:
> :It actually a point that the whole of IRIX 6.5 is MIPS3 which is fair
> :enough, given the range of MIPS3 and 4 hardware it runs on, but how
> :much performance is actually being lost running IRIX 6.5 on MIPS4
> :hardware ?

> IRIX 6.5 *is* -mips4 on Origin 2xxx, Origin 3xxx, and Octane (all
> of which are IRIX64 machines)

> IRIX 6.5 is -mips3 on R4x00 Challenge XL (an IRIX64 machine not
> able to run -mips4)

> IRIX 6.5 is -mips3 on an R12000 O2 (a machine able to run -mips4
> but which SGI has for marketing reasons constrained to 32 bit pointers.)

mips3 vs. mips4 has nothing to do with IRIX64 vs. IRIX. I don't understand
what point you are trying to make with the above list, since any CPU after
R4000 (which itsself is buggy) supports 64bit pointers. Mips4 on the
other hand just adds some new instructions like conditional moves, some
prefetch operations etc.) and isn't related at all. My O2 (R5k) runs
mips4 binaries just fine, but of course, there's no support for IRIX64.
64 bit pointers will just _decrease_ performance in many cases by wasting
memory bandwidth.

rainer


[...]

In article ,
Rainer Canavan wrote:
|In comp.sys.sgi.admin Walter Roberson wrote:
|> In article ,

|> IRIX 6.5 *is* -mips4 on Origin 2xxx, Origin 3xxx, and Octane (all
|> of which are IRIX64 machines)

|> IRIX 6.5 is -mips3 on an R12000 O2 (a machine able to run -mips4
|> but which SGI has for marketing reasons constrained to 32 bit pointers.)

:I don't understand
:what point you are trying to make with the above list, since any CPU after
:R4000 (which itsself is buggy) supports 64bit pointers. Mips4 on the
ther hand just adds some new instructions like conditional moves, some
refetch operations etc.) and isn't related at all. My O2 (R5k) runs
:mips4 binaries just fine, but of course, there's no support for IRIX64.

You missed the phrase "marketing reasons".

As far as I can see, nearly every mips4 capable machine *already* uses
-mips4 for IRIX 6.5's /unix . In case I wasn't sufficiently clear:

$ uname -aR
IRIX64 origin 6.5 6.5.17f 07121148 IP27
$ file /unix
/unix: ELF 64-bit MSB mips-4 executable (not stripped) MIPS - version 1

I don't have a Power Indigo^2 or R10K Indigo^2 to test on.

The one exception that I know of, the mips4 capable machine that has
a -mips3 /unix, is the O2. We already know that the O2 was restricted
not run IRIX64 for market-positioning reasons [so as to not undercut
the Octane], so we can readily hypothesize that the -mips3 /unix
on the O2 is mostly another market positioning matter. I suspect that
if you were to check an R5000 Indy that it would also be -mips3 /unix,
in which case the general rule would be "Any IRIX 6.5 kernel that
is not IRIX64, is compiled -mips3." It'd be interesting to check
an R10K Indigo^2.


You asked how much efficiency is being lost by virtue of IRIX 6.5 being
-mips3 insted of -mips4. The answer is that no efficiency at all
is being lost in any system that SGI markets as being high end: the
Octane, Octane2, and Fuel, and all the servers are -already- -mips4
for the kernel. Only their entry -mips4 machines have -mips3 kernels,
and it shouldn't surprise you if entry machines are not engineered
to be as efficient as possible.
--
Live it up, rip it up, why so lazy?
Give it out, dish it out, let's go crazy, yeah!
-- Supertramp (The USENET Song)

In article ,
roberson@ibd.nrc-cnrc.gc.ca (Walter Roberson) wrote:

: I suspect that
: if you were to check an R5000 Indy that it would also be -mips3 /unix,
: in which case the general rule would be "Any IRIX 6.5 kernel that
: is not IRIX64, is compiled -mips3." It'd be interesting to check
: an R10K Indigo^2.

R5k Indy: ELF N32 MSB mips-3
R10k Indigo2: ELF 64-bit MSB mips-4


Cheers - Tony 'Nicoya' Mantler

--
Tony 'Nicoya' Mantler -- Master of Code-fu -- nicoya@apia.dhs.org
-- http://www.apia.dhs.org/ -- http://nicoya.feline.pp.se/ --