why do we do it

[segv]

On Sun, 21 Aug 2005 02:32:35 +0200

I wonder how long time a full NetBSD build takes on the slowest hardware
able to do the build itself :-)

This also depends on the compiler, I've been reading that gcc 3.x are much
slower at compiling source files than the previous versions.

--
Posted automagically by a mail2news gateway at muc.de e.V.
Please direct questions, flames, donations, etc. to news-***@muc.de

[Nino Dehne]

I have an old DECStation 5000/133 that where I rebuild netbsd-current from
time to time.
build.sh started: Fri Aug 12 12:01:56 CEST 2005
build.sh ended: Sat Aug 20 19:16:05 CEST 2005

Hi,

I can understand doing it _once_ for the fun of it. But more than once?
I applaud your patience :)

Seriously, I run some slow platforms myself (AXPpci33, SPARCstation 2,
040/25 Amiga) but I only build their worlds on an i386 box that can do
it in some hours. Is there a real reason to not cross-build?

Regards,

ND

--
Posted automagically by a mail2news gateway at muc.de e.V.
Please direct questions, flames, donations, etc. to news-***@muc.de

[Johnny Billquist]

Post by Nino Dehne

I have an old DECStation 5000/133 that where I rebuild netbsd-current from
time to time.
build.sh started: Fri Aug 12 12:01:56 CEST 2005
build.sh ended: Sat Aug 20 19:16:05 CEST 2005

Hi,
I can understand doing it _once_ for the fun of it. But more than once?
I applaud your patience :)
Seriously, I run some slow platforms myself (AXPpci33, SPARCstation 2,
040/25 Amiga) but I only build their worlds on an i386 box that can do
it in some hours. Is there a real reason to not cross-build?

Bit rot? If not used, it will break in time. Such as the VAX port, which
actually can't run through build.sh, and haven't been able in several
years, and which don't show up when you do a cross-build.

Johnny

Johnny Billquist || "I'm on a bus
|| on a psychedelic trip
email: ***@update.uu.se || Reading murder books
pdp is alive! || tryin' to stay hip" - B. Idol

--
Posted automagically by a mail2news gateway at muc.de e.V.
Please direct questions, flames, donations, etc. to news-***@muc.de

[George Michaelson]

From memory, in 1982 the time to recompile 4.1BSD on a 780 was 2 days.

So we've not done too badly to move on up to a week in over 20 years.

-George

--
Posted automagically by a mail2news gateway at muc.de e.V.
Please direct questions, flames, donations, etc. to news-***@muc.de

[Johnny Billquist]

Post by George Michaelson
From memory, in 1982 the time to recompile 4.1BSD on a 780 was 2 days.
So we've not done too badly to move on up to a week in over 20 years.

Umm, given that it now takes slightly over a week to compile the whole
shebang on a VAX-8650, that would be about six or seven weeks on an
11/780...

4.3 Reno compiled blindingly fast in comparision on the same hardware.

Johnny

Johnny Billquist || "I'm on a bus
|| on a psychedelic trip
email: ***@update.uu.se || Reading murder books
pdp is alive! || tryin' to stay hip" - B. Idol

--
Posted automagically by a mail2news gateway at muc.de e.V.
Please direct questions, flames, donations, etc. to news-***@muc.de

[Sean Davis]

Post by segv
On Sun, 21 Aug 2005 02:32:35 +0200

I wonder how long time a full NetBSD build takes on the slowest
hardware able to do the build itself :-)

I suspect this would be a VAX 11/750: 0.65 MIPS.
More common is a MicroVAX II: 0.9 MIPS.
There is the even slower VAX 11/730: 0.3 MIPS, but IIRC it is not
supported by NetBSD.
I leave it to your homework to calculate estimated build times based on
the MIPS numbers. ;-)

I can also tell that on a VAX-8650, it takes a little over a week
nowadays.
In speed comparision, a VAX-8650 is about 6 VUPS (maybe a bit more), while
the 11/750 is the same 0.65 VUPS. That would leave us somewhere around 11
weeks. Now, an 11/750 can't have more than 14 megs of physical memory (or
something like it), which means it will probably be even worse.

Has anybody written anything to benchmark a machine in VUPS? I'd be
interested to know how many VUPS an Athlon gets, or an Opteron ;)

-Sean

--
Posted automagically by a mail2news gateway at muc.de e.V.
Please direct questions, flames, donations, etc. to news-***@muc.de

[George Michaelson]

the most significant measure is PhD's per CPU.

when we used DEC-10 class equipment, it was 50-60 logins per CPU, so
the PhD density outcome was quite high.

It dropped slightly in the Vax era.

Now, its usually multiple CPU's per PhD.

either the PhD is getting more complicated, or the CPUs are getting
dumber :-)

-george

--
Posted automagically by a mail2news gateway at muc.de e.V.
Please direct questions, flames, donations, etc. to news-***@muc.de

[Johnny Billquist]

Post by Sean Davis

Post by segv
On Sun, 21 Aug 2005 02:32:35 +0200

I wonder how long time a full NetBSD build takes on the slowest
hardware able to do the build itself :-)

I suspect this would be a VAX 11/750: 0.65 MIPS.
More common is a MicroVAX II: 0.9 MIPS.
There is the even slower VAX 11/730: 0.3 MIPS, but IIRC it is not
supported by NetBSD.
I leave it to your homework to calculate estimated build times based on
the MIPS numbers. ;-)

I can also tell that on a VAX-8650, it takes a little over a week
nowadays.
In speed comparision, a VAX-8650 is about 6 VUPS (maybe a bit more), while
the 11/750 is the same 0.65 VUPS. That would leave us somewhere around 11
weeks. Now, an 11/750 can't have more than 14 megs of physical memory (or
something like it), which means it will probably be even worse.

Has anybody written anything to benchmark a machine in VUPS? I'd be
interested to know how many VUPS an Athlon gets, or an Opteron ;)

Doubt there would be a point to it, since you don't have the same
instructions. VUPS is mostly meaningful to compare different VAXen, since
that will tell you how fast the same code would execute on different
machines. Binary code, that is.

The problems with the MIPS measurement is just that it used some C code
for reference, and when compilers got better at generating code, even a
VAX-11/780 became faster than one MIPS.

So when you depend on the result of a compiler, and use different
compilers, how do you tell the speed of the machine in relation to the
behavour of the compiler?

Johnny

Johnny Billquist || "I'm on a bus
|| on a psychedelic trip
email: ***@update.uu.se || Reading murder books
pdp is alive! || tryin' to stay hip" - B. Idol

--
Posted automagically by a mail2news gateway at muc.de e.V.
Please direct questions, flames, donations, etc. to news-***@muc.de

[Bill Studenmund]

Post by Johnny Billquist

Post by Sean Davis

Post by segv
On Sun, 21 Aug 2005 02:32:35 +0200
I leave it to your homework to calculate estimated build times based on
the MIPS numbers. ;-)

I can also tell that on a VAX-8650, it takes a little over a week
nowadays.
In speed comparision, a VAX-8650 is about 6 VUPS (maybe a bit more), while
the 11/750 is the same 0.65 VUPS. That would leave us somewhere around 11
weeks. Now, an 11/750 can't have more than 14 megs of physical memory (or
something like it), which means it will probably be even worse.

Has anybody written anything to benchmark a machine in VUPS? I'd be
interested to know how many VUPS an Athlon gets, or an Opteron ;)

Doubt there would be a point to it, since you don't have the same
instructions. VUPS is mostly meaningful to compare different VAXen, since
that will tell you how fast the same code would execute on different
machines. Binary code, that is.

Well, there is the concept of running the VAX simulator on those systems.
And given how CPU performance had grown over time, it could make for some
fast VAX systems. :-)

Take care,

Bill

[Johnny Billquist]

Post by Bill Studenmund

Post by Johnny Billquist

Post by Sean Davis

Post by segv
On Sun, 21 Aug 2005 02:32:35 +0200
I leave it to your homework to calculate estimated build times based on
the MIPS numbers. ;-)

I can also tell that on a VAX-8650, it takes a little over a week
nowadays.
In speed comparision, a VAX-8650 is about 6 VUPS (maybe a bit more), while
the 11/750 is the same 0.65 VUPS. That would leave us somewhere around 11
weeks. Now, an 11/750 can't have more than 14 megs of physical memory (or
something like it), which means it will probably be even worse.

Has anybody written anything to benchmark a machine in VUPS? I'd be
interested to know how many VUPS an Athlon gets, or an Opteron ;)

Doubt there would be a point to it, since you don't have the same
instructions. VUPS is mostly meaningful to compare different VAXen, since
that will tell you how fast the same code would execute on different
machines. Binary code, that is.

Well, there is the concept of running the VAX simulator on those systems.
And given how CPU performance had grown over time, it could make for some
fast VAX systems. :-)

True. And you should probably be able to rate the simlators VUPS.
A new compile of the simulator, with a better compiler (or other options)
would however give you a machine with a different VUPS rating. So it would
vary... :-)

Johnny

Johnny Billquist || "I'm on a bus
|| on a psychedelic trip
email: ***@update.uu.se || Reading murder books
pdp is alive! || tryin' to stay hip" - B. Idol

--
Posted automagically by a mail2news gateway at muc.de e.V.
Please direct questions, flames, donations, etc. to news-***@muc.de