From: st...@dpn.com (Steve Dixon)
Subject: Tune-up software?
Date: 1999/03/14
Message-ID: <36EC1F8C.B5A05CBE@dpn.com>#1/1
X-Deja-AN: 454964405
Approved: newsg...@newsgate.IAEhv.nl
Organization: IAE newsgate
Followup-To: poster
Reply-To: redhat-devel-l...@redhat.com
Newsgroups: linux.redhat.devel
my bosses are seriously looking to replace our SCO machines with linux
and ive been going over ther various things that we have on our SCO
machines. i was wondering if linux has such a thing like Tune-up Pro
from Olympus(www.olysoft.com)? it tunes certain memory and kernel
parameters on our SCO machines automatically. anyone no of something
for linux? i didnt know where else to ask, sorry.
Steve Dixon
--
To unsubscribe:
mail -s unsubscribe redhat-devel-list-requ...@redhat.com < /dev/null
From: kv...@ricochet.net (Kevin Vajk)
Subject: Re: Tune-up software?
Date: 1999/03/14
Message-ID: <Pine.LNX.4.04.9903141248490.2278-100000@barbelo.localdomain>#1/1
X-Deja-AN: 454967254
Approved: newsg...@newsgate.IAEhv.nl
References: <36EC1F8C.B5A05CBE@dpn.com>
Organization: IAE newsgate
Followup-To: poster
Reply-To: redhat-devel-l...@redhat.com
Newsgroups: linux.redhat.devel
On Sun, 14 Mar 1999, Steve Dixon wrote:
> my bosses are seriously looking to replace our SCO machines with linux
> and ive been going over ther various things that we have on our SCO
> machines. i was wondering if linux has such a thing like Tune-up Pro
> from Olympus(www.olysoft.com)? it tunes certain memory and kernel
> parameters on our SCO machines automatically. anyone no of something
> for linux? i didnt know where else to ask, sorry.
I don't know of any product like this, sorry.
I'm not familiar with SCO (my background is in HP-UX) but I think
you're gonna find that, in comparison, the process of building a
new kernel on Linux is fairly involved.
If I were you, I'd just skip the kernel tuning entirely. Linux
is extremely small and fast out of the box; I think you'll be
quite satisfied with an "untuned" Linux system.
- Kevin Vajk
<kv...@ricochet.net>
--
To unsubscribe:
mail -s unsubscribe redhat-devel-list-requ...@redhat.com < /dev/null
From: jmkel...@radix.net (James Michael Keller)
Subject: Re: Tune-up software?
Date: 1999/03/14
Message-ID: <36EC3625.71FB8503@radix.net>#1/1
X-Deja-AN: 454991176
Approved: newsg...@newsgate.IAEhv.nl
References: <Pine.LNX.4.04.9903141248490.2278-100000@barbelo.localdomain>
Organization: KellTech
Followup-To: poster
Reply-To: redhat-devel-l...@redhat.com
Newsgroups: linux.redhat.devel
I would dissagree. A lot of the "tuning" needed in various
Unix flavors isn't needed with linux. The bigest gain is as
someothers have mentioned - compiling for your chipt type (
He didn't mention what his SCO was running on so I'm not
able to comment on that )
Then only compiling those features needed for the
servers/workstations in question, or compling everything as
a module - with only the compiled in options needed to boot
( ie your boot partitions block device driver ) allowing the
admin to insert and remove modules for things when needed
without droping the system to load the new kernal.
For workstations and home use - I just compile in the
needed stuff, leave the rest out. In those cases I can
afford to use a 'make -j 15' in the Makefile and let my
system grind away for 15 minutes and compile a new kernal if
I needed something else.
Kevin Vajk wrote:
>
> On Sun, 14 Mar 1999, Steve Dixon wrote:
>
> > my bosses are seriously looking to replace our SCO machines with linux
> > and ive been going over ther various things that we have on our SCO
> > machines. i was wondering if linux has such a thing like Tune-up Pro
> > from Olympus(www.olysoft.com)? it tunes certain memory and kernel
> > parameters on our SCO machines automatically. anyone no of something
> > for linux? i didnt know where else to ask, sorry.
>
> I don't know of any product like this, sorry.
>
> I'm not familiar with SCO (my background is in HP-UX) but I think
> you're gonna find that, in comparison, the process of building a
> new kernel on Linux is fairly involved.
>
> If I were you, I'd just skip the kernel tuning entirely. Linux
> is extremely small and fast out of the box; I think you'll be
> quite satisfied with an "untuned" Linux system.
>
> - Kevin Vajk
> <kv...@ricochet.net>
--
===========================================================
James Michael Keller | jmkel...@radix.net
http://www.radix.net/~jmkeller
-----------------------------------------------------------
Contents (c)1999 James Michael Keller. All rights reserved
===========================================================
--
To unsubscribe:
mail -s unsubscribe redhat-devel-list-requ...@redhat.com < /dev/null
From: ch...@topdog.pas1.logicon.com (Chris Albertson)
Subject: Re: Tune-up software?
Date: 1999/03/15
Message-ID: <36ED5976.9B55A28@topdog.logicon.com>#1/1
X-Deja-AN: 455315380
Approved: newsg...@newsgate.IAEhv.nl
References: <36EC1F8C.B5A05CBE@dpn.com>
Organization: Logicon RDA
Followup-To: poster
Reply-To: redhat-devel-l...@redhat.com
Newsgroups: linux.redhat.devel
Steve Dixon wrote:
>
> i was wondering if linux has such a thing like Tune-up Pro
> from Olympus(www.olysoft.com)? it tunes certain memory and kernel
> parameters on our SCO machines automatically. anyone no of something
> for linux? i didnt know where else to ask, sorry.
Maybe a better question is "Do you need this under Linux?"
I would think not.
--
--Chris Albertson
ch...@topdog.logicon.com Voice: 626-351-0089 X127
Logicon RDA, Pasadena California Fax: 626-351-0699
--
To unsubscribe:
mail -s unsubscribe redhat-devel-list-requ...@redhat.com < /dev/null
From: j...@club-internet.fr
Subject: Re: Tune-up software?
Date: 1999/03/16
Message-ID: <19990316064352.1462.qmail@sidney.remcomp.fr>#1/1
X-Deja-AN: 455695398
Approved: newsg...@newsgate.IAEhv.nl
References: <m10MIyn-0007U1C@the-village.bc.nu>
Organization: IAE newsgate
Followup-To: poster
Reply-To: redhat-devel-l...@redhat.com
Newsgroups: linux.redhat.devel
> Resent-Cc: recipient list not shown: ;
> MBOX-Line: From redhat-devel-list-requ...@redhat.com Sun Mar 14 16:18:05 1999
> From: a...@lxorguk.ukuu.org.uk (Alan Cox)
> Date: Sun, 14 Mar 1999 22:03:29 +0000 (GMT)
> Content-Type: text
> Resent-From: redhat-devel-l...@redhat.com
> Reply-To: redhat-devel-l...@redhat.com
> X-Mailing-List: <redhat-devel-l...@redhat.com> archive/latest/1266
> X-Loop: redhat-devel-l...@redhat.com
> Precedence: list
> Resent-Sender: redhat-devel-list-requ...@redhat.com
> X-URL: http://www.redhat.com
> X-UIDL: c9cc2fc8b94614ae345df123497474a3
>
> > my bosses are seriously looking to replace our SCO machines with linux
> > and ive been going over ther various things that we have on our SCO
> > machines. i was wondering if linux has such a thing like Tune-up Pro
> > from Olympus(www.olysoft.com)? it tunes certain memory and kernel
> > parameters on our SCO machines automatically. anyone no of something
> > for linux? i didnt know where else to ask, sorry.
>
> Linux is as much as possible intended to be self tuning. The single big
> performance leap you will make is to build a kernel configuration matching
> your machine - especially CPU type.
>
Could you detail? I didn't find anything in 2.O's source having
significant on performance: impact of selective invalidation of TLBs
is insignificant except on purpose built examples.
And using the Byte test I measured effect of changing GCC parms to
about 2% when you go from 386-parms to those used by Pentiums. The
test platforms were a P75 and an Amd K6 300.
About tuning: it is a pity there isn't more info about the best
values for /proc/sysctl.
--
Jean Francois Martinez
Project Independence: Linux for the Masses
http://www.independence.seul.org
--
To unsubscribe:
mail -s unsubscribe redhat-devel-list-requ...@redhat.com < /dev/null
From: a...@lxorguk.ukuu.org.uk (Alan Cox)
Subject: Re: Tune-up software?
Date: 1999/03/16
Message-ID: <m10Myuk-0007U2C@the-village.bc.nu>#1/1
X-Deja-AN: 455698361
Approved: newsg...@newsgate.IAEhv.nl
References: <19990316064352.1462.qmail@sidney.remcomp.fr>
Organization: IAE newsgate
Followup-To: poster
Reply-To: redhat-devel-l...@redhat.com
Newsgroups: linux.redhat.devel
> Could you detail? I didn't find anything in 2.O's source having
> significant on performance: impact of selective invalidation of TLBs
> is insignificant except on purpose built examples.
Selective invalidate is one
Lack of a write protect from kernel space is the 2nd
The wp stuff makes a big difference to 2.2, but a lot less to 2.0
> And using the Byte test I measured effect of changing GCC parms to
> about 2% when you go from 386-parms to those used by Pentiums. The
> test platforms were a P75 and an Amd K6 300.
2% is not insubstantial. Thats half an hour every day of CPU time.
Alan
--
To unsubscribe:
mail -s unsubscribe redhat-devel-list-requ...@redhat.com < /dev/null
From: j...@club-internet.fr
Subject: Re: Tune-up software?
Date: 1999/03/18
Message-ID: <19990317215659.8709.qmail@sidney.remcomp.fr>#1/1
X-Deja-AN: 455915252
Approved: newsg...@newsgate.IAEhv.nl
References: <m10Myuk-0007U2C@the-village.bc.nu>
Organization: IAE newsgate
Followup-To: poster
Reply-To: redhat-devel-l...@redhat.com
Newsgroups: linux.redhat.devel
> Resent-Cc: recipient list not shown: ;
> MBOX-Line: From redhat-devel-list-requ...@redhat.com Tue Mar 16 13:05:36 1999
> From: a...@lxorguk.ukuu.org.uk (Alan Cox)
> Date: Tue, 16 Mar 1999 18:50:06 +0000 (GMT)
> Content-Type: text
> Resent-From: redhat-devel-l...@redhat.com
> Reply-To: redhat-devel-l...@redhat.com
> X-Mailing-List: <redhat-devel-l...@redhat.com> archive/latest/1288
> X-Loop: redhat-devel-l...@redhat.com
> Precedence: list
> Resent-Sender: redhat-devel-list-requ...@redhat.com
> X-URL: http://www.redhat.com
> X-UIDL: 80a580f9fd3c84464b446bdc51d57f9e
>
> > Could you detail? I didn't find anything in 2.O's source having
> > significant on performance: impact of selective invalidation of TLBs
> > is insignificant except on purpose built examples.
>
> Selective invalidate is one
> Lack of a write protect from kernel space is the 2nd
>
> The wp stuff makes a big difference to 2.2, but a lot less to 2.0
>
As far as I can remember in 2.0 the fact you select 386 or Pentium
when recompiling the kernel has no effect on wp: it is a runtime
optimization. In 2.0 there are only two parts of the code depending
on what CPU you selected at compile time: how you invert bytes and
selective invalidation.
About selective invalidation I took a worst (completely artificial)
case and assuming the process will run a tick and then be preempted
the loss of performance is about 3 in thousand on a K6-300.
I published a detailed analysis in February issue of LinuxGazette.
Goal was to dispel the kernel compiling myth who makes difficult to
convert non-geeks to Linuxm. I was tempted to submit the article to
you before publishing but I didn't dare.
> > And using the Byte test I measured effect of changing GCC parms to
> > about 2% when you go from 386-parms to those used by Pentiums. The
> > test platforms were a P75 and an Amd K6 300.
>
> 2% is not insubstantial. Thats half an hour every day of CPU time.
>
Assuming the box is doing nothing else on the whole day that running
kernel code generated by GCC. In real world it will be running user
code, or running assmbler parts of the kernel or waiting for devices
when no is process ready, at times it will do active loops and at
times it will just sit idle.
Also 2% of a century is two years but a two per cent speed increase
will only be noticeable on speed tests and computer races :-). The
user won't feel it.
--
Jean Francois Martinez
Project Independence: Linux for the Masses
http://www.independence.seul.org
--
To unsubscribe:
mail -s unsubscribe redhat-devel-list-requ...@redhat.com < /dev/null
From: a...@lxorguk.ukuu.org.uk (Alan Cox)
Subject: Re: Tune-up software?
Date: 1999/03/18
Message-ID: <m10NQjE-0007U1C@the-village.bc.nu>#1/1
X-Deja-AN: 455918370
Approved: newsg...@newsgate.IAEhv.nl
References: <19990317215659.8709.qmail@sidney.remcomp.fr>
Organization: IAE newsgate
Followup-To: poster
Reply-To: redhat-devel-l...@redhat.com
Newsgroups: linux.redhat.devel
> As far as I can remember in 2.0 the fact you select 386 or Pentium
> when recompiling the kernel has no effect on wp: it is a runtime
> optimization. In 2.0 there are only two parts of the code depending
> on what CPU you selected at compile time: how you invert bytes and
> selective invalidation.
That would explain why that one seemed to be noise level. I suspect
a 2.2 repeat would be fairly different. On 2.2 we use the hardware trap
invalid userspace pointers but on the 386 we have to do the old style
verify_area().
I agree its never a user problem though. Its the distributors job to get
that right
--
To unsubscribe:
mail -s unsubscribe redhat-devel-list-requ...@redhat.com < /dev/null