From: "Albert D. Cahalan" <acaha...@cs.uml.edu>
Subject: performance & you-know-who
Date: 1999/05/09
Message-ID: <fa.fdkpgtv.1cggeb1@ifi.uio.no>#1/1
X-Deja-AN: 475752415
Original-Date: Sat, 8 May 1999 20:48:47 -0400 (EDT)
Sender: owner-linux-ker...@vger.rutgers.edu
Original-Message-Id: <199905090048.UAA09003@jupiter.cs.uml.edu>
To: linux-ker...@vger.rutgers.edu
X-Orcpt: rfc822;linux-kernel-outgoing-dig
Organization: Internet mailing list
Newsgroups: fa.linux.kernel
X-Loop: majord...@vger.rutgers.edu


There is a _new_ "Open Benchmark Invitation" on Mindcraft's site.
http://www.mindcraft.com/openbenchmark.html

This one is hard to refuse. PC Week has offered to supply the lab
and settle disputes over the rules. We got NT clients added, along
with an unlimited-patch test run and some random minor junk.

(we can kill atime, mke2fs with fancy options or use reiserfs, add the
Beowulf hack for multiple Ethernet cards, hack the scheduler, etc.)

Linux might really lose. This is like benchmarking knfsd against the
user-space NFS, with NT getting the kernel-mode advantage. Ouch.
NT will be keeping all files open during the entire http test,
binding processes and cards to processors, etc.

At least we get to spend a day tuning, so somebody will get a chance
to profile the kernel on very expensive hardware. If Linux does lose,
somebody might even get bothered enough to write ksmbd and khttpd.

To refuse is to admit failure, not even giving Linux a chance.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/

From: Andrea Arcangeli <and...@e-mind.com>
Subject: Re: performance & you-know-who
Date: 1999/05/09
Message-ID: <fa.hn3s2vv.4l2s9q@ifi.uio.no>#1/1
X-Deja-AN: 475752406
Original-Date: Sun, 9 May 1999 04:03:07 +0200 (CEST)
Sender: owner-linux-ker...@vger.rutgers.edu
Original-Message-ID: <Pine.LNX.4.05.9905090353100.878-100000@laser.random>
References: <fa.fdkpgtv.1cggeb1@ifi.uio.no>
To: "Albert D. Cahalan" <acaha...@cs.uml.edu>
X-Sender: and...@laser.random
X-Authentication-Warning: laser.random: andrea owned process doing -bs
Content-Type: TEXT/PLAIN; charset=US-ASCII
X-Orcpt: rfc822;linux-kernel-outgoing-dig
Organization: Internet mailing list
X-Public-Key-URL: http://e-mind.com/~andrea/aa.asc
MIME-Version: 1.0
Newsgroups: fa.linux.kernel
X-Loop: majord...@vger.rutgers.edu

On Sat, 8 May 1999, Albert D. Cahalan wrote:

>(we can kill atime, mke2fs with fancy options or use reiserfs, add the
>Beowulf hack for multiple Ethernet cards, hack the scheduler, etc.)

This mean that they must my tree to do the benchmark:

	ftp://e-mind.com/pub/andrea/kernel/2.2.7_andrea7.bz2

Killing atime and all other issues are really _minor_ issues compared to
my redesign of the flushtime handling, of the dirty buffer garbage
collection and on my new shrink_mmap design.

If the benchmark imply VM and I/O load, my tree will give a big boost (as
_worse_ a 10% improvement) and it's rock solid (I doesn't work by luck).

I also have a patch against pre-2.2.8-4 but I want to finish fixing the
new scheduler in the pre-patches before releasing it (I plan to finish
tomorrow).

How can I get in touch with people how is doing the benchmark?

Andrea Arcangeli


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/

Re: performance & you-know-who

> How can I get in touch with people how is doing the benchmark?

Maybe these:

Bob Young, President of Red Hat Software
Andrew Tridgell <tridge@samba.org> of the Samba team
Jeremy Allison <jallison@cthulhu.engr.sgi.com> of the Samba team
Alan Cox, who seems to be involved
John Taschek from PC Week at john_taschek@zd.com

The benchmark operation seems to be located in one of two places:

a. North Carolina's Research Triangle, which is perfect for Red Hat.
   I guess davem@redhat.com could end up with the job.

b. Foster City, California, 20 to 25 miles from Linus. Is he busy?
   This is SVLUG territory AFAIK. Hopefully rioting can be kept to
   a minimum. Jeremy Allison seems to be in the area.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/

From: Alan Cox <a...@lxorguk.ukuu.org.uk>
Subject: Re: performance & you-know-who
Date: 1999/05/09
Message-ID: <fa.j6k05fv.unc59h@ifi.uio.no>#1/1
X-Deja-AN: 475823069
Original-Date: Sun, 9 May 1999 14:59:57 +0100
Sender: owner-linux-ker...@vger.rutgers.edu
Original-Message-ID: <19990509145957.B20670@lxorguk.ukuu.org.uk>
References: <fa.feklguv.1dgsfb5@ifi.uio.no>
To: "Albert D. Cahalan" <acaha...@cs.uml.edu>, Andrea Arcangeli <and...@e-mind.com>
Original-References: <Pine.LNX.4.05.9905090353100.878-100...@laser.random> 
<199905090324.XAA31...@jupiter.cs.uml.edu>
Content-Type: text/plain; charset=us-ascii
X-Orcpt: rfc822;linux-kernel-outgoing-dig
Organization: Internet mailing list
Mime-Version: 1.0
Newsgroups: fa.linux.kernel
X-Loop: majord...@vger.rutgers.edu

> Alan Cox, who seems to be involved

I am not involved and I am not interested in being used to dignify Mindcraft's
last flailing attempts to stay in business at any cost.

Nor despite their claim have they addressed one of they key problems with
their open test : At the end of this, even if Mindcraft pull stupid games
the Linux people are obliged by the agreeemnt to issue a joint press
release

	", and which will be positive about ..."

I find it rather hard to believe there will be anything about Weiner to be
positive about.

Alan

-- 
With trembling hands he unfurled the ancient cracked parchment, this was
the place, it had to be. Uncertainly he began to mumble the chant "rdbms,
sql , third normal formal form, java,  table, scalable". Something moved..
From outside they heard a scream and a thud. The sales department had awoken


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/

Re: performance & you-know-who

>> Alan Cox, who seems to be involved
>
> I am not involved and I am not interested in being used to dignify
> Mindcraft's last flailing attempts to stay in business at any cost.

You were just flaming? I'd heard that you provided suggestions.

PC Week will be running this. I don't look forward to reports of Linux
people wimping out. I see only two weak excuses left, the Ami MegaRAID
card and unassisted Apache server.

It is stupid to simply refuse, and perhaps stupid to simply accept.
The only safe option is to create a list of demands that would make
the benchmark acceptable.

> the Linux people are obliged by the agreeemnt to issue a joint press
> release
> 
> 	", and which will be positive about ..."

That is quite silly. Consider these demands:

1. No more joint press release. They do theirs, we do a few dozen.
2. Zeus web server and/or (?) reverse squid proxy
3. Mylex or ICP-Vortex instead of Ami MegaRAID

Would that be enough for you? How about if VA Research Linux Solutions
was allowed to substitute an equal-cost Linux server? (good PR for them)

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/

From: Alan Cox <a...@lightning.swansea.linux.org.uk>
Subject: Re: performance & you-know-who
Date: 1999/05/10
Message-ID: <fa.mkj80rv.o4c80q@ifi.uio.no>#1/1
X-Deja-AN: 476028043
Original-Date: Mon, 10 May 1999 02:46:09 +0100
Sender: owner-linux-ker...@vger.rutgers.edu
Original-Message-ID: <19990510024609.A23839@lightning.swansea.linux.org.uk>
References: <fa.fdkbitv.1cg6fr5@ifi.uio.no>
To: "Albert D. Cahalan" <acaha...@cs.uml.edu>
Original-References: <19990509145957.B20...@lxorguk.ukuu.org.uk> 
<199905100042.UAA22...@jupiter.cs.uml.edu>
Content-Type: text/plain; charset=us-ascii
X-Orcpt: rfc822;linux-kernel-outgoing-dig
Organization: Internet mailing list
Mime-Version: 1.0
Newsgroups: fa.linux.kernel
X-Loop: majord...@vger.rutgers.edu

> You were just flaming? I'd heard that you provided suggestions.

I made a few observations, got no feedback and had my emails used to try
and make it look like we supported them

> 1. No more joint press release. They do theirs, we do a few dozen.
> 2. Zeus web server and/or (?) reverse squid proxy
> 3. Mylex or ICP-Vortex instead of Ami MegaRAID

That seems to be covered. No I'll wait for an independant body to
conduct an independant test on an independantly configured setup

> Would that be enough for you? How about if VA Research Linux Solutions
> was allowed to substitute an equal-cost Linux server? (good PR for them)

VA are expensive. If you want to play cost games you use someone else.
I'm busy until August, I have real projects to do paid for by real people.
I don't have time to go play games with Weiner

Alan

-- 
With trembling hands he unfurled the ancient cracked parchment, this was
the place, it had to be. Uncertainly he began to mumble the chant "rdbms,
sql , third normal formal form, java,  table, scalable". Something moved..
From outside they heard a scream and a thud. The sales department had awoken


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/

Re: performance & you-know-who

> Killing atime and all other issues are really _minor_ issues compared to
> my redesign of the flushtime handling, of the dirty buffer garbage
> collection and on my new shrink_mmap design.
> 
> If the benchmark imply VM and I/O load, my tree will give a big boost (as
> _worse_ a 10% improvement) and it's rock solid (I doesn't work by luck).

If there is much VM and disk activity, we lose. NT keeps the whole damn
data set in memory with file handles open the whole time.

So if your patch changes anything, we have worse problems.
(assuming you meant _disk_ I/O load, not network I/O load)

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/

From: torva...@transmeta.com (Linus Torvalds)
Subject: Re: performance & you-know-who
Date: 1999/05/09
Message-ID: <fa.jjtf2iv.1oneuj9@ifi.uio.no>#1/1
X-Deja-AN: 475752412
Original-Date: 9 May 1999 04:46:37 GMT
Sender: owner-linux-ker...@vger.rutgers.edu
Original-Message-ID: <7h33vd$ud$1@palladium.transmeta.com>
References: <fa.hn3s2vv.4l2s9q@ifi.uio.no>
To: linux-ker...@vger.rutgers.edu
Original-References: <199905090048.UAA09...@jupiter.cs.uml.edu> 
<Pine.LNX.4.05.9905090353100.878-100...@laser.random>
X-Authentication-Warning: palladium.transmeta.com: 
bin set sender to n...@transmeta.com using -f
X-Orcpt: rfc822;linux-kernel-outgoing-dig
Organization: Transmeta Corporation, Santa Clara, CA
Newsgroups: fa.linux.kernel
X-Loop: majord...@vger.rutgers.edu

In article <Pine.LNX.4.05.9905090353100.878-100...@laser.random>,
Andrea Arcangeli  <and...@e-mind.com> wrote:
>
>If the benchmark imply VM and I/O load, my tree will give a big boost (as
>_worse_ a 10% improvement) and it's rock solid (I doesn't work by luck).

According to what the mindcraft people reported about their runs
earlier, the disk light is not actually on very much during the test:
both Linux and NT keep the whole working set in memory. 

Of course, so far we haven't actually seen what is going on, so take
that with a pinch of salt. 

		Linus

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/

From: Matthew Kirkwood <weej...@ferret.lmh.ox.ac.uk>
Subject: Re: performance & you-know-who
Date: 1999/05/09
Message-ID: <fa.m48tjrv.2l23gs@ifi.uio.no>#1/1
X-Deja-AN: 475791781
Original-Date: Sun, 9 May 1999 11:10:28 +0100 (GMT)
Sender: owner-linux-ker...@vger.rutgers.edu
Original-Message-ID: <Pine.LNX.3.96.990509110733.25457A-100000@ferret.lmh.ox.ac.uk>
References: <fa.jjtf2iv.1oneuj9@ifi.uio.no>
To: Linus Torvalds <torva...@transmeta.com>
Content-Type: TEXT/PLAIN; charset=US-ASCII
X-Orcpt: rfc822;linux-kernel-outgoing-dig
Organization: Internet mailing list
MIME-Version: 1.0
Newsgroups: fa.linux.kernel
X-Loop: majord...@vger.rutgers.edu

On 9 May 1999, Linus Torvalds wrote:

> Andrea Arcangeli  <and...@e-mind.com> wrote:
> >If the benchmark imply VM and I/O load, my tree will give a big boost (as
> >_worse_ a 10% improvement) and it's rock solid (I doesn't work by luck).
>
> According to what the mindcraft people reported about their runs
> earlier, the disk light is not actually on very much during the test:
> both Linux and NT keep the whole working set in memory.

I may be missing something, but it seems to me that the "working set in
memory" thing basically makes a nonsense of their claim that this is an
"enterprise-class" setup.

Matthew.


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/

Why does Mindcraft insist on 4* 100BaseTX?

I've read the Mindcraft Open benchmark invitation,
and Mindcraft (ie Microsoft) require 4*100BaseTX.

It that a common installation, or do you usually
use Gigabyte ethernet?
I know that Windows NT can bind the 4 Interrupts to
the 4 CPU and improve the interrupt throuput,
but has anyone performed tests with Linux and this
hardware combination?

I'm sure Microsoft did these tests before they have
choosen the hardware.

Regards,
	Manfred

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/

Re: Why does Mindcraft insist on 4* 100BaseTX?

> I know that Windows NT can bind the 4 Interrupts to
> the 4 CPU and improve the interrupt throuput,
> but has anyone performed tests with Linux and this
> hardware combination?

Not only that they will have chosen the cards carefully to be
optimal for NT versus Linux - like they did with the raid
controller

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/

From: Linus Torvalds <torva...@transmeta.com>
Subject: Re: performance & you-know-who
Date: 1999/05/10
Message-ID: <fa.oeqbf8v.k4c185@ifi.uio.no>#1/1
X-Deja-AN: 476242847
Original-Date: Mon, 10 May 1999 09:59:06 -0700 (PDT)
Sender: owner-linux-ker...@vger.rutgers.edu
Original-Message-ID: <Pine.LNX.3.95.990510095740.6890D-100000@penguin.transmeta.com>
References: <fa.m48tjrv.2l23gs@ifi.uio.no>
To: Matthew Kirkwood <weej...@ferret.lmh.ox.ac.uk>
X-Authentication-Warning: penguin.transmeta.com: torvalds owned process doing -bs
Content-Type: TEXT/PLAIN; charset=US-ASCII
X-Orcpt: rfc822;linux-kernel-outgoing-dig
Organization: Internet mailing list
MIME-Version: 1.0
Newsgroups: fa.linux.kernel
X-Loop: majord...@vger.rutgers.edu



On Sun, 9 May 1999, Matthew Kirkwood wrote:
> On 9 May 1999, Linus Torvalds wrote:
> >
> > According to what the mindcraft people reported about their runs
> > earlier, the disk light is not actually on very much during the test:
> > both Linux and NT keep the whole working set in memory.
> 
> I may be missing something, but it seems to me that the "working set in
> memory" thing basically makes a nonsense of their claim that this is an
> "enterprise-class" setup.

It gets worse.

Their NT "enterprise-class setup" is apparently using a FAT filesystem and
only connecting to w95/w98 clients because that is where NT does best
(none of this silly security crap ;) 

		Linus


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/