Path: nntp.gmd.de!newsserver.jvnc.net!howland.reston.ans.net!quagga.ru.ac.za!
ucthpx!itu1.sun.ac.za!arichfld
From: arich...@cs.sun.ac.za (Antony Richfield)
Newsgroups: comp.os.linux.misc,alt.folklore.computers
Subject: Linux' shortcomings in shootouts ...
Date: 19 Nov 1994 05:33:00 GMT
Organization: University of Stellenbosch
Lines: 9
Message-ID: <3ak2mc$spo@itu1.sun.ac.za>
NNTP-Posting-Host: 146.232.212.20
X-Newsreader: TIN [version 1.2 PL1]
I have heard that Linux performs miserably against various free BSD
forms running around ... how much of this is true/still valid/negligible
and so on ... any comment? insights? data? updates?
--
Antony Richfield at the University of Stellenbosch, South Africa
GAT d? H s+:- g+ p?+ !au a22 w+++ v?*(+) C+ U? P-- L 3 E- N++ K+(---) W-- M-- !V
-po+ Y+ t !5 !j R++ G'' !tv b++ D+ B? e+(*)>++++ u**(*) h*(-) f--(?)@ r-- !n !y*
-----------------------------------GEEK CODE 2.1-------------------------------
Path: nntp.gmd.de!xlink.net!howland.reston.ans.net!vixen.cso.uiuc.edu!
rasputin.ncsa.uiuc.edu!libor
From: li...@rasputin.ncsa.uiuc.edu (Libor)
Newsgroups: comp.os.linux.misc,alt.folklore.computers
Subject: Re: Linux' shortcomings in shootouts ...
Date: 20 Nov 1994 06:22:38 GMT
Organization: University of Illinois at Urbana
Lines: 22
Distribution: world
Message-ID: <3ampve$2hq@vixen.cso.uiuc.edu>
References: <3ak2mc$spo@itu1.sun.ac.za>
NNTP-Posting-Host: rasputin.ncsa.uiuc.edu
In article <3ak2mc$...@itu1.sun.ac.za>, arich...@cs.sun.ac.za
(Antony Richfield) writes:
|> I have heard that Linux performs miserably against various free BSD
|> forms running around ... how much of this is true/still valid/negligible
|> and so on ... any comment? insights? data? updates?
I know you asked about free BSD, but I haven't used it in a year. I do
know that BSDI (a commercial unix for Intel) does context switches
much faster the Linux, I saw a listing of the speed of context switches
for a large amount of machines. In some cases Linux took an order of
magnitude longer then some work stations. Times where in cycles, so
processor clocking had no effect on the results. However I saw this
about half a year ago, so Linux may have improved. From personal
experience, I would say that at least BSDI is noticably quicker then
Linux, but you get what you pay for. Don't get me wrong I think
Linux is great, aand I'm very happy it exists, but like most things
in the computer world there is room for improvment.
-Libor Anarchist, Atheist, Research Programmer
P.S. I saw the list on comp.arch might want to check if anyone there
has it.
Newsgroups: comp.os.linux.misc,alt.folklore.computers
Path: nntp.gmd.de!xlink.net!howland.reston.ans.net!pipex!uknet!info!iialan
From: iia...@iifeak.swan.ac.uk (Alan Cox)
Subject: Re: Linux' shortcomings in shootouts ...
Message-ID: <CzoJzA.8F5@info.swan.ac.uk>
Sender: n...@info.swan.ac.uk
Nntp-Posting-Host: iifeak.swan.ac.uk
Organization: Institute For Industrial Information Technology
References: <3ak2mc$spo@itu1.sun.ac.za> <3ampve$2hq@vixen.cso.uiuc.edu>
Date: Tue, 22 Nov 1994 17:45:09 GMT
Lines: 19
In article <3ampve$...@vixen.cso.uiuc.edu> li...@rasputin.ncsa.uiuc.edu
(Libor) writes:
>I know you asked about free BSD, but I haven't used it in a year. I do
>know that BSDI (a commercial unix for Intel) does context switches
>much faster the Linux, I saw a listing of the speed of context switches
>for a large amount of machines. In some cases Linux took an order of
>magnitude longer then some work stations. Times where in cycles, so
The actual context switch isnt too bad. The two problems are the scheduler
algorithm and the syscall setup time. People are working on cleaning these
up for 1.3.x. At the moment its one glaring area where most unices perform
better than Linux. Of course the QNX people will happily reel out their
benchmark to trash everyone 8)
Alan
--
..-----------,,----------------------------,,----------------------------,,
// Alan Cox // iia...@www.linux.org.uk // GW4PTS@GB7SWN.#45.GBR.EU //
``----------'`----------------------------'`----------------------------''
Path: nntp.gmd.de!dearn!blekul11!idefix.CS.kuleuven.ac.be!ub4b!EU.net!uunet!
heifetz.msen.com!zib-berlin.de!cs.tu-berlin.de!news
From: nic...@prz.tu-berlin.de (Juergen Nickelsen)
Newsgroups: comp.os.linux.misc,alt.folklore.computers
Subject: Re: Linux' shortcomings in shootouts ...
Date: 23 Nov 1994 18:39:17 GMT
Organization: Technical University of Berlin, Germany.
Lines: 23
Message-ID: <3b01sg$t43@news.cs.tu-berlin.de>
References: <3ak2mc$spo@itu1.sun.ac.za> <3ampve$2hq@vixen.cso.uiuc.edu>
<CzoJzA.8F5@info.swan.ac.uk>
Reply-To: nic...@cs.tu-berlin.de
NNTP-Posting-Host: toftum.prz.tu-berlin.de
Mime-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Transfer-Encoding: 8bit
In-reply-to: iialan@iifeak.swan.ac.uk's message of Tue, 22 Nov 1994 17:45:09 GMT
In article <CzoJzA....@info.swan.ac.uk> iia...@iifeak.swan.ac.uk (Alan
Cox) writes:
> The actual context switch isnt too bad. The two problems are the scheduler
> algorithm and the syscall setup time. People are working on cleaning these
> up for 1.3.x. At the moment its one glaring area where most unices perform
> better than Linux. Of course the QNX people will happily reel out their
> benchmark to trash everyone 8)
This is getting more and more technical and thus more and more
interesting. What I now *really* would like to see are the *figures*
for all this (time of context switch, scheduler overhead, syscall
setup time, perhaps more). Not opinions, but real numbers, processor
cycles (of the same processor, of course) preferred.
So what are these, for, say, BSDI, NetBSD, FreeBSD, Unixware, SCO,
Linix, QNX? Any *real* *data* is greatly appreciated.
[Hmm, this is perhaps not much of a folklore issue. But it may well be
one in a few years when we are all running plan9 or WindozeNT. :-) ]
--
Juergen Nickelsen
Newsgroups: comp.os.linux.misc,alt.folklore.computers
Path: nntp.gmd.de!xlink.net!howland.reston.ans.net!news.sprintlink.net!
sunserver.insinc.net!cuugnet!wooga!camz
From: c...@wooga.cuug.ab.ca (Martin Zimmerman)
Subject: Re: Linux' shortcomings in shootouts ...
Followup-To: comp.os.linux.misc,alt.folklore.computers
References: <3ak2mc$spo@itu1.sun.ac.za> <3ampve$2hq@vixen.cso.uiuc.edu>
<3b01sg$t43@news.cs.tu-berlin.de>
Organization: Camz Enterprises
Date: Thu, 24 Nov 1994 01:52:24 GMT
X-Newsreader: TIN [version 1.2 PL2]
Message-ID: <Czr17D.5LE@wooga.cuug.ab.ca>
Lines: 52
Juergen Nickelsen (nic...@prz.tu-berlin.de) wrote:
> In article <CzoJzA....@info.swan.ac.uk> iia...@iifeak.swan.ac.uk (Alan
> Cox) writes:
> > up for 1.3.x. At the moment its one glaring area where most unices perform
> > better than Linux. Of course the QNX people will happily reel out their
> > benchmark to trash everyone 8)
Actually, Linux does *quite* well for context switching, QNX still beats it
though :-) The benchmark I used was actually originally written for a SUN
workstation and is pretty portable across platforms. I have run the test
on DEC OSF/1 AXP machines, SUN Sparc boxes, HP 9000 boxes, as well as some
machines running QNX and linux. Here are the results:
sun (sparc20) 11.00 secs, 9.09/millisec 122 microsec/switch
dec 13.64 secs, 7.33/millisec 136 microsec/switch
hp 15.25 secs, 6.56/millisec 152 microsec/switch
cuug486 (linux) 16.29 secs, 6.14/millisec 163 microsec/switch
sun (sparc2) 23.96 secs, 4.17/millisec 240 microsec/switch
ibm 68.67 secs, 1.46/millisec 687 microsec/switch
qnx 4.21 [486/33] 9.56 secs, 10.46/millisec 96 microsec/switch
linux [486DX2/50] 13.87 secs, 7.21/millisec 139 microsec/switch
qnx 4.21 [386/40] 14.97 secs, 6.68/millisec 150 microsec/switch
The machines are a Sparc20 running solaris, a DEC 3300 AXP running
OSF/1, and HP 9000/415 running HP-UX, a 486/33 (I think) running Linux
(not sure which version), a Sparc2 running SunOS, and an *old* IBM
RS/6000 running AIX. These are machines that together make up the
network at CUUG.
More information on the context switch used as well as source are
available on my home page: http://www.cuug.ab.ca:8001/~zimmerm/ for
those of you interested. If you run the benchmark on your machine,
please send me the results.
> So what are these, for, say, BSDI, NetBSD, FreeBSD, Unixware, SCO,
> Linix, QNX? Any *real* *data* is greatly appreciated.
The scheduling algorithm *IS* important as well. QNX offers three
types of scheduling that the task can select from. I don't have the
details on the overhead of an interrupt and such, so hopefully Dan H
at QSSL will jump in and provide that info.
Cheers,
Camz.
--
Martin Zimmerman - Camz Enterprises - QNX Programming & Consulting
WWW: http://www.cuug.ab.ca:8001/~zimmerm/
email: c...@wooga.cuug.ab.ca Keeper of the QNX FAQ.
finger: zimm...@www.cuug.ab.ca Work in realtime - work in QNX.
Path: nntp.gmd.de!xlink.net!howland.reston.ans.net!quagga.ru.ac.za!
ucthpx!itu1.sun.ac.za!arichfld
From: arich...@cs.sun.ac.za (Antony Richfield)
Newsgroups: comp.os.linux.misc,alt.folklore.computers
Subject: Re: Linux' shortcomings in shootouts ...
Date: 24 Nov 1994 13:30:35 GMT
Organization: University of Stellenbosch
Lines: 48
Message-ID: <3b24hr$7os@itu1.sun.ac.za>
References: <3ak2mc$spo@itu1.sun.ac.za> <3ampve$2hq@vixen.cso.uiuc.edu>
<3b01sg$t43@news.cs.tu-berlin.de> <Czr17D.5LE@wooga.cuug.ab.ca>
NNTP-Posting-Host: 146.232.212.20
X-Newsreader: TIN [version 1.2 PL1]
Martin Zimmerman (c...@wooga.cuug.ab.ca) wrote:
: Juergen Nickelsen (nic...@prz.tu-berlin.de) wrote:
: > > better than Linux. Of course the QNX people will happily reel out their
: > > benchmark to trash everyone 8)
: Actually, Linux does *quite* well for context switching, QNX still beats it
: though :-) The benchmark I used was actually originally written for a SUN
: workstation and is pretty portable across platforms. I have run the test
: on DEC OSF/1 AXP machines, SUN Sparc boxes, HP 9000 boxes, as well as some
: machines running QNX and linux. Here are the results:
: sun (sparc20) 11.00 secs, 9.09/millisec 122 microsec/switch
: dec 13.64 secs, 7.33/millisec 136 microsec/switch
: hp 15.25 secs, 6.56/millisec 152 microsec/switch
: cuug486 (linux) 16.29 secs, 6.14/millisec 163 microsec/switch
: sun (sparc2) 23.96 secs, 4.17/millisec 240 microsec/switch
: ibm 68.67 secs, 1.46/millisec 687 microsec/switch
: qnx 4.21 [486/33] 9.56 secs, 10.46/millisec 96 microsec/switch
: linux [486DX2/50] 13.87 secs, 7.21/millisec 139 microsec/switch
: qnx 4.21 [386/40] 14.97 secs, 6.68/millisec 150 microsec/switch
: The machines are a Sparc20 running solaris, a DEC 3300 AXP running
: OSF/1, and HP 9000/415 running HP-UX, a 486/33 (I think) running Linux
: (not sure which version), a Sparc2 running SunOS, and an *old* IBM
: RS/6000 running AIX. These are machines that together make up the
: network at CUUG.
: More information on the context switch used as well as source are
: available on my home page: http://www.cuug.ab.ca:8001/~zimmerm/ for
: those of you interested. If you run the benchmark on your machine,
: please send me the results.
: > So what are these, for, say, BSDI, NetBSD, FreeBSD, Unixware, SCO,
: > Linix, QNX? Any *real* *data* is greatly appreciated.
: The scheduling algorithm *IS* important as well. QNX offers three
: types of scheduling that the task can select from. I don't have the
: details on the overhead of an interrupt and such, so hopefully Dan H
: at QSSL will jump in and provide that info.
Thank you very much, sir! this data I will be examining closely.
--
Antony Richfield at the University of Stellenbosch, South Africa
GAT d? H s+:- g+ p?+ !au a22 w+++ v?*(+) C+ U? P-- L 3 E- N++ K+(---) W-- M-- !V
-po+ Y+ t !5 !j R++ G'' !tv b++ D+ B? e+(*)>++++ u**(*) h*(-) f--(?)@ r-- !n !y*
-----------------------------------GEEK CODE 2.1-------------------------------
Path: nntp.gmd.de!xlink.net!howland.reston.ans.net!swrinde!sgiblab!
sgigate.sgi.com!fido.asd.sgi.com!slovax!lm
From: l...@slovax.engr.sgi.com (Larry McVoy)
Newsgroups: comp.os.linux.misc,alt.folklore.computers
Subject: Re: Linux' shortcomings in shootouts ...
Date: 26 Nov 1994 19:50:59 GMT
Organization: Silicon Graphics Inc., Mountain View, CA
Lines: 17
Message-ID: <3b83j3$e3v@fido.asd.sgi.com>
References: <3ak2mc$spo@itu1.sun.ac.za> <3ampve$2hq@vixen.cso.uiuc.edu>
<3b01sg$t43@news.cs.tu-berlin.de> <Czr17D.5LE@wooga.cuug.ab.ca>
Reply-To: l...@slovax.engr.sgi.com
NNTP-Posting-Host: slovax.engr.sgi.com
X-Newsreader: TIN [version 1.2 PL2]
I believe this is the code that I wrote. It would be nice if you didn't
use it or post numbers from it any more because it includes more
than just context switching time, it also includes the signal overhead.
I've written a new and improved one that is quite a bit more accurate -
still not perfect, but a lot closer.
You can get it in comp.benchmarks, I just posted something called lmbench
which includes the new ctx switching code.
: Actually, Linux does *quite* well for context switching, QNX still beats it
: though :-) The benchmark I used was actually originally written for a SUN
: workstation and is pretty portable across platforms.
--
---
Larry McVoy (415) 390-1804 l...@sgi.com
Path: nntp.gmd.de!xlink.net!howland.reston.ans.net!swrinde!sgiblab!
sgigate.sgi.com!enews.sgi.com!fido.asd.sgi.com!slovax!lm
From: l...@slovax.engr.sgi.com (Larry McVoy)
Newsgroups: comp.os.linux.misc,alt.folklore.computers
Subject: Re: Linux' shortcomings in shootouts ...
Followup-To: comp.os.linux.misc,alt.folklore.computers
Date: 26 Nov 1994 19:54:04 GMT
Organization: Silicon Graphics Inc., Mountain View, CA
Lines: 76
Message-ID: <3b83os$e3v@fido.asd.sgi.com>
References: <3ak2mc$spo@itu1.sun.ac.za> <3ampve$2hq@vixen.cso.uiuc.edu>
<3b01sg$t43@news.cs.tu-berlin.de>
Reply-To: l...@slovax.engr.sgi.com
NNTP-Posting-Host: slovax.engr.sgi.com
X-Newsreader: TIN [version 1.2 PL2]
Juergen Nickelsen (nic...@prz.tu-berlin.de) wrote:
: This is getting more and more technical and thus more and more
: interesting. What I now *really* would like to see are the *figures*
: for all this (time of context switch, scheduler overhead, syscall
: setup time, perhaps more). Not opinions, but real numbers, processor
: cycles (of the same processor, of course) preferred.
: So what are these, for, say, BSDI, NetBSD, FreeBSD, Unixware, SCO,
: Linix, QNX? Any *real* *data* is greatly appreciated.
See comp.benchmarks for source.
L M B E N C H 1 . 0 S U M M A R Y
------------------------------------
Processor, Processes - times in microseconds
--------------------------------------------
Host OS Mhz Null Null Simple /bin/sh Mmap 2-proc 8-proc
Syscall Process Process Process lat ctxsw ctxsw
--------- ------------- ---- ------- ------- ------- ------- ---- ------ ------
rs6000 AIX 2 62 23 2.0K 7.3K 23K 3817 20 32
seahorse HP-UX A.09.03 99 14 3.6K 10.1K 18K 116 25 29
snake HP-UX A.09.01 66 21 2.6K 5.7K 17K 156 47 55
IP22 IRIX 5.3 198 11 3.1K 8.0K 19K 260 40 38
pentium Linux 1.1.54 91 3 3.3K 15.4K 49K 33 66 94
alpha OSF1 V2.1 182 13 4.8K 16.1K 43K 172 25 42
ss20.50 SunOS 5.4 50 9 10.7K 57.5K 113K 130 54 85
ss20.61 SunOS 5.4 61 7 8.0K 45.8K 87K 104 37 52
*Local* Communication latencies in microseconds
-----------------------------------------------
Host OS Pipe UDP RPC/ TCP RPC/
UDP TCP
--------- ------------- ------- ------- ------- ------- -------
rs6000 AIX 2 143 385 820 498 1054
seahorse HP-UX A.09.03 193 244 832 262 812
snake HP-UX A.09.01 296 403 1195 367 1142
IP22 IRIX 5.3 131 313 671 278 641
pentium Linux 1.1.54 157 658 1030 1164 1591
alpha OSF1 V2.1 185 404 718 428 851
ss20.50 SunOS 5.4 194 590 935 560 1196
ss20.61 SunOS 5.4 150 414 622 335 784
*Local* Communication bandwidths in megabytes/second
----------------------------------------------------
Host OS Pipe TCP File Mmap Bcopy Bcopy Mem Mem
reread reread (libc) (hand) read write
--------- ------------- ---- ---- ------ ------ ------ ------ ---- -----
rs6000 AIX 2 34 6.0 76.1 63.0 81 120 99 169
seahorse HP-UX A.09.03 38 35.2 44.7 32.1 25 31 49 52
snake HP-UX A.09.01 19 17.8 34.4 22.3 22 24 45 39
IP22 IRIX 5.3 34 22.1 32.3 43.7 32 31 69 66
pentium Linux 1.1.54 13 2.4 9.8 4.7 18 18 48 32
alpha OSF1 V2.1 32 12.1 39.4 22.7 39 41 76 78
ss20.50 SunOS 5.4 11 11.0 22.9 30.0 26 31 80 62
ss20.61 SunOS 5.4 24 19.5 31.0 30.7 23 24 59 40
Memory latencies in nanoseconds
(WARNING - may not be correct, check graphs)
--------------------------------------------
Host OS Mhz L1 $ L2 $ Main mem TLB Guesses
--------- ------------- --- ---- ---- -------- --- -------
rs6000 AIX 2 61 15 229 247 776 No L2 cache?
seahorse HP-UX A.09.03 98 10 10 393 481 No L1 cache?
snake HP-UX A.09.01 65 15 15 378 1051 No L1 cache?
IP22 IRIX 5.3 197 10 76 1018 1129
pentium Linux 1.1.54 90 11 294 439 1254
alpha OSF1 V2.1 182 10 56 321 452
ss20.50 SunOS 5.4 49 20 284 291 575 No L2 cache?
ss20.61 SunOS 5.4 60 16 115 816 961
--
---
Larry McVoy (415) 390-1804 l...@sgi.com
Newsgroups: comp.os.linux.misc,alt.folklore.computers
Path: nntp.gmd.de!xlink.net!howland.reston.ans.net!Germany.EU.net!EU.net!
uknet!info!iialan
From: iia...@iifeak.swan.ac.uk (Alan Cox)
Subject: Re: Linux' shortcomings in shootouts ...
Message-ID: <Czz9oC.Gw4@info.swan.ac.uk>
Sender: n...@info.swan.ac.uk
Nntp-Posting-Host: iifeak.swan.ac.uk
Organization: Institute For Industrial Information Technology
References: <3ampve$2hq@vixen.cso.uiuc.edu> <CzoJzA.8F5@info.swan.ac.uk>
<3b01sg$t43@news.cs.tu-berlin.de>
Date: Mon, 28 Nov 1994 12:36:12 GMT
Lines: 22
In article <3b01sg$...@news.cs.tu-berlin.de> nic...@cs.tu-berlin.de writes:
>This is getting more and more technical and thus more and more
>interesting. What I now *really* would like to see are the *figures*
>for all this (time of context switch, scheduler overhead, syscall
>setup time, perhaps more). Not opinions, but real numbers, processor
>cycles (of the same processor, of course) preferred.
I've got nothing more useful than the byte benchmarks syscall test
[watch this its basically one syscall in a loop and some vendors optimise
for the benchmark!] the byte pipe test (context switching a pipe) and
multiple scripts test (this one really shows the weaknesses in the Linux
scheduler).
Anyone with some really good benchmark recommendations - I too will be
interested in this and if we can start a context switch/scheduling/syscall
overhead fight between the competitors we all benefit afterwards ;)
Alan
--
..-----------,,----------------------------,,----------------------------,,
// Alan Cox // iia...@www.linux.org.uk // GW4PTS@GB7SWN.#45.GBR.EU //
``----------'`----------------------------'`----------------------------''