From: Volker Dittmar <ditt...@internet.de>
Subject: Linux vs. NT: Benchmarks
Date: 1999/01/31
Message-ID: <36B49497.E3478308@internet.de>#1/1
X-Deja-AN: 439067493
Content-Transfer-Encoding: 7bit
X-Accept-Language: de
Content-Type: text/plain; charset=us-ascii
Organization: Posted via the Nacamar Network
Mime-Version: 1.0
Newsgroups: comp.os.ms-windows.nt.advocacy
Just take a look at some benchmarks which compare Linux and NT as
file- and webservers at:
http://www.zdnet.com/sr/stories/issue/0,4537,387506,00.html
Yes, the benchmark has been done by Ziff-Davies. They've been trying
to hold Linux down for quite a few years, so you can't accuse them
for being unfriendly to M$ or NT.
Just guess who wins these benchmarks?
It is a rhetorical question: of course, Linux wins in every case,
as Linux fans told you before (some people, like Boris, had told
you its a lie, now you can see for yourself. I'm just thinking to
remove Boris from my killfile to see how he will react...).
It shouldn't surprise you to see what a big difference Linux makes,
and how much faster it is. At 32 clients, Linux + Samba has an
advantage of 250% (as a file server). As a web server, Linux is
up to 50% faster. Only with very few users NT is as fast as Linux.
I've just told you... and don't forget to compare the price... hehe
I hope one day they will redo this benchmark with kernel 2.2.x. Linux
has improved its performance, esp. on SMP-machines, but even for now
it kicks NT's butt.
Ciao,
Volker Dittmar
From: bob_gib...@my-dejanews.com
Subject: Re: Linux vs. NT: Benchmarks
Date: 1999/02/09
Message-ID: <79p850$vc7$1@nnrp1.dejanews.com>#1/1
X-Deja-AN: 442330632
References: <36B49497.E3478308@internet.de>
X-Http-Proxy: 1.1 x3.dejanews.com:80 (Squid/1.1.22) for client 193.69.73.2
Organization: Deja News - The Leader in Internet Discussion
X-Article-Creation-Date: Tue Feb 09 12:02:43 1999 GMT
Newsgroups: comp.os.ms-windows.nt.advocacy
X-Http-User-Agent: Mozilla/4.0 (compatible; MSIE 4.01; Windows NT)
Hey, I like Linux:> but I work in an invironment/location where you better
know NT or forget about working in the computer industry. Also, just
remember that speed isn't everything - how about security? Security is more
important than just speed these days. Also, Linux is just to complicated for
most users and has a long way to go to reach the desktop - a long
way............ Sorry Linux:<
In article <36B49497.E3478...@internet.de>,
Volker Dittmar <ditt...@internet.de> wrote:
> Just take a look at some benchmarks which compare Linux and NT as
> file- and webservers at:
>
> http://www.zdnet.com/sr/stories/issue/0,4537,387506,00.html
>
> Yes, the benchmark has been done by Ziff-Davies. They've been trying
> to hold Linux down for quite a few years, so you can't accuse them
> for being unfriendly to M$ or NT.
>
> Just guess who wins these benchmarks?
>
> It is a rhetorical question: of course, Linux wins in every case,
> as Linux fans told you before (some people, like Boris, had told
> you its a lie, now you can see for yourself. I'm just thinking to
> remove Boris from my killfile to see how he will react...).
>
> It shouldn't surprise you to see what a big difference Linux makes,
> and how much faster it is. At 32 clients, Linux + Samba has an
> advantage of 250% (as a file server). As a web server, Linux is
> up to 50% faster. Only with very few users NT is as fast as Linux.
>
> I've just told you... and don't forget to compare the price... hehe
>
> I hope one day they will redo this benchmark with kernel 2.2.x. Linux
> has improved its performance, esp. on SMP-machines, but even for now
> it kicks NT's butt.
>
> Ciao,
> Volker Dittmar
>
-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/ Search, Read, Discuss, or Start Your Own
From: Volker Dittmar <v...@feri.de>
Subject: Re: Linux vs. NT: Benchmarks
Date: 1999/02/09
Message-ID: <36C05A8B.A3E38D43@feri.de>#1/1
X-Deja-AN: 442437656
Content-Transfer-Encoding: 7bit
References: <36B49497.E3478308@internet.de> <79p850$vc7$1@nnrp1.dejanews.com>
X-Accept-Language: de
Content-Type: text/plain; charset=us-ascii
Organization: FERI Systems GmbH
Mime-Version: 1.0
Newsgroups: comp.os.ms-windows.nt.advocacy
bob_gib...@my-dejanews.com wrote:
>
> Hey, I like Linux:> but I work in an invironment/location where you better
> know NT or forget about working in the computer industry. Also, just
> remember that speed isn't everything - how about security? Security is more
> important than just speed these days. Also, Linux is just to complicated for
> most users and has a long way to go to reach the desktop - a long
> way............ Sorry Linux:<
We're speaking about database and file servers - not about desktops!
Most
people with NT knowledge will confuse this completely. Its like saying:
"I can't use this machine as a database server, because it can't run
Excel..." - a complete nonsense. And as a server, sometimes NT is easier
to administrate, sometimes its Linux (remotely, it is always Linux).
Speaking about security: this one of the VERY weak spots of NT. Because
the methods of security are proprietary and hidden, not many know HOW
secure NT really is, but according to some people in the know it is
*VERY* poor. With Linux, the opposite is true: the safety is great,
because of the peer review of the source code. This is safety on SOURCE
CODE level. Some threats have been removed because someone has seen a
potential danger just by looking into the code. With NT, this is impos-
sible. (If you want to nkow more about this, see the discussion of Linux
vs. NT at http://www.idg.net - its quite amazing)
And don't tell me NT 4.0 has C2 level security - this, simply, isn't
true. The C2 certificate has been given to NT 3.50, and it has NOT
BEEN reviewed with NT 4.0. M$ tried to confuse people about this.
(BTW: the C2 security was given WHEN and ONLY when the machine had
no floppy drive and no network connection... very secure indeed, esp.
if used as a server...).
The discussion is amazing: I tell the people, Linux is faster than NT...
I'm called a liar... I prove it... and everyone keeps telling me "forget
about speed, security is it..." and NOW I tell you that Linux is more
secure... again you will tell me that I lie... I will prove it to you...
now what will you tell me next?
Come on, there is no way out: Linux is the successor of Windows NT.
In the internet (a trend setter, anyway) it already has won against
NT. The corporate server market will be next, and then it will start
to conquer the desktop - not this year, but next year. One after the
other, step for step. In year 2005 it has overrun NT (Windows 2000).
The "long way to the desktop": just another two years until it is
taken seriously. Even M$ shares this opinion.
Ciao,
Volker Dittmar
From: "Boris" <borisspa...@pleasemovil.com>
Subject: Re: Linux vs. NT: Benchmarks
Date: 1999/02/09
Message-ID: <36c0ab34$0$16680@nntp1.ba.best.com>#1/1
X-Deja-AN: 442561589
References: <36B49497.E3478308@internet.de> <79p850$vc7$1@nnrp1.dejanews.com>
<36C05A8B.A3E38D43@feri.de> <36c07187.1188208@news.gv2.com>
<36C09C99.6A48CF99@internet.de>
X-Mimeole: Produced By Microsoft MimeOLE V4.72.3155.0
X-Trace: nntp1.ba.best.com 918596404 16680 209.24.240.6
Reply-To: "Boris" <XXXbo...@movil.comXXX>
Newsgroups: comp.os.ms-windows.nt.advocacy
Ignore this troll.
Boris
From: Mark Robinson <Plasm...@iname.com>
Subject: Re: Linux vs. NT: Benchmarks
Date: 1999/02/10
Message-ID: <36C0DDD3.186A99BD@iname.com>#1/1
X-Deja-AN: 442626536
Content-Transfer-Encoding: 7bit
References: <36B49497.E3478308@internet.de> <79p850$vc7$1@nnrp1.dejanews.com>
<36C05A8B.A3E38D43@feri.de> <36c07187.1188208@news.gv2.com>
<36C09C99.6A48CF99@internet.de> <36c0ab34$0$16680@nntp1.ba.best.com>
X-Accept-Language: en
Content-Type: text/plain; charset=us-ascii
X-Complaints-To: abuse@home.net
X-Trace: news.rdc1.bc.wave.home.com 918609343 24.65.12.67 (Tue, 09 Feb 1999 17:15:43 PDT)
Organization: @Home Network Canada
MIME-Version: 1.0
NNTP-Posting-Date: Tue, 09 Feb 1999 17:15:43 PDT
Newsgroups: comp.os.ms-windows.nt.advocacy
Boris wrote:
> Ignore this troll.
>
> Boris
Boris you are the troll, Volker is not.
From: "Boris" <borisspa...@pleasemovil.com>
Subject: Re: Linux vs. NT: Benchmarks
Date: 1999/02/10
Message-ID: <36c140e0$0$16688@nntp1.ba.best.com>#1/1
X-Deja-AN: 442724656
References: <36B49497.E3478308@internet.de> <79p850$vc7$1@nnrp1.dejanews.com>
<36C05A8B.A3E38D43@feri.de> <36c07187.1188208@news.gv2.com>
<36C09C99.6A48CF99@internet.de> <36c0ab34$0$16680@nntp1.ba.best.com>
<36C0DDD3.186A99BD@iname.com>
X-MimeOLE: Produced By Microsoft MimeOLE V4.72.3155.0
X-Trace: nntp1.ba.best.com 918634720 16688 209.24.242.229
Reply-To: "Boris" <XXXbo...@movil.comXXX>
Newsgroups: comp.os.linux.advocacy,comp.os.ms-windows.nt.advocacy
>Boris you are the troll, Volker is not.
Oh yeah? You are too dumb to figure out what's going on? I'm lazy but this time I'll chew
it for you.
1. This zdnet test. http://www.zdnet.com/sr/stories/issue/0,4537,387506,00.html . How come
they only tested upto 64 MB memory NT configurations? Production NT servers with that
little memory will definitely experience RAM starvation. All NT tests at www.tpc.org or
www.specbench.org mention systems with at least 512MB RAM (more often 2 to 4GB. And they
didn't bother to tune NT. It's very easy: via control panel you can select whether your
server will mostly serve file, run applications,or manage user domain accounts.
Sure NT has larger memory footprint than Linux. But who cares: memory is cheap nowdays.
2. Linux is based on outdated inferior architecture: monolithic kernel, no threads as
opposed to NT which is employs micro-kernel architecture and kernel threads.
3. Security: NT is WAY more secure than Linux which has very basic security. Should I chew
this one too? OK. NT has NTFS file system which allows you to specify access to individual
files and directories for users, groups, etc. Unlike Unix basic root,owner,group,world
permissions model NT allows to specify arbitrary user and group names in file permissions.
In addition, you can turn on auditing for NTFS files: attempts to access those files will
be recorded in OS Security Log. OS Security Log can contain ~ 100 different types of
events including: file access, registry access, logon, logoff, process creation, system
time change, user account creation and deletion, priviledge level change in user accounts,
etc. All NT user and kernel objects are protected by access control lists similar to NTFS
files: shared memory segments, named pipes, mailboxes, etc. Page file space is cleaned
when reused by different processes to ensure that confidential information doesn't leak
between processes. Lots of extra features: like account lockout after several unsuccessful
login attemps, enforcing strong passwords - are available. SP4 contains Security Console
which allows administrator configure most security-related system settings; after saving
config info Sec. Console can be brought up at any time to compare current system security
settings with config file. For example, if strong passwords feature was enabled during
system config and disabled later Sec. Console will highlight this difference from original
configuration. Administrator can just synch system settings with original config.
4. I can understand that some people prefer Unix because they are used to it and don't
want or are not smart enough to use NT efficiently. But that so-called Linux community
produces tremendous amounts of stink combined with very moderate technical achievements.
Boris
From: Shashi M Sharma <shas...@charvak.kla-tencor.com>
Subject: Re: Linux vs. NT: Benchmarks
Date: 1999/02/10
Message-ID: <86aeymbm9m.fsf@charvak.kla-tencor.com>#1/1
X-Deja-AN: 442748365
References: <36B49497.E3478308@internet.de> <79p850$vc7$1@nnrp1.dejanews.com>
<36C05A8B.A3E38D43@feri.de> <36c07187.1188208@news.gv2.com>
<36C09C99.6A48CF99@internet.de> <36c0ab34$0$16680@nntp1.ba.best.com>
<36C0DDD3.186A99BD@iname.com> <36c140e0$0$16688@nntp1.ba.best.com>
Organization: Another Netscape News Server User
Newsgroups: comp.os.linux.advocacy,comp.os.ms-windows.nt.advocacy
"Boris" <borisspa...@pleasemovil.com> writes:
[snipped]
> 2. Linux is based on outdated inferior architecture:
>monolithic kernel, no threads as
> opposed to NT which is employs
>micro-kernel architecture and kernel threads.
>
Linux has kernel threads. In fact almost all modern UNIX supports
kernel threads.
> 3. Security: NT is WAY more secure than Linux which has very basic
> security. Should I chew this one too? OK. NT has NTFS file system
> which allows you to specify access to individual files and
> directories for users, groups, etc. Unlike Unix basic
> root,owner,group,world permissions model NT allows to specify
> arbitrary user and group names in file permissions.
NT may be theoretically secure but its not as secure as Solaris. NT
security is limited to DACL while Solaris provides Mandatory ACL. All
of the rest stuff is available on Solaris.
[snipped]
> n. Administrator
can just synch system settings with original config.
>
> 4. I can understand that some people prefer Unix because they are
> used to it and don't want or are not smart enough to use NT
> efficiently. But that so-called Linux community produces tremendous
> amounts of stink combined with very moderate technical achievements.
>
> Boris
Whatever.
From: "Boris" <borisspa...@pleasemovil.com>
Subject: Re: Linux vs. NT: Benchmarks
Date: 1999/02/10
Message-ID: <36c16ccf$0$16678@nntp1.ba.best.com>#1/1
X-Deja-AN: 442765278
References: <36B49497.E3478308@internet.de> <79p850$vc7$1@nnrp1.dejanews.com>
<36C05A8B.A3E38D43@feri.de> <36c07187.1188208@news.gv2.com>
<36C09C99.6A48CF99@internet.de> <36c0ab34$0$16680@nntp1.ba.best.com>
<36C0DDD3.186A99BD@iname.com> <36c140e0$0$16688@nntp1.ba.best.com>
<86aeymbm9m.fsf@charvak.kla-tencor.com>
X-MimeOLE: Produced By Microsoft MimeOLE V4.72.3155.0
X-Trace: nntp1.ba.best.com 918645967 16678 209.24.240.89
Reply-To: "Boris" <XXXbo...@movil.comXXX>
Newsgroups: comp.os.linux.advocacy,comp.os.ms-windows.nt.advocacy
>Linux has kernel threads. In fact almost all modern UNIX supports
>kernel threads.
I think you are mistaken. What system call do you use to create a new thread in Linux?
Boris
From: Bernd Paysan <bernd.pay...@gmx.de>
Subject: Re: Linux vs. NT: Benchmarks
Date: 1999/02/10
Message-ID: <79s8gk$klt$1@nnrp1.dejanews.com>#1/1
X-Deja-AN: 442815862
References: <36B49497.E3478308@internet.de> <79p850$vc7$1@nnrp1.dejanews.com>
<36C05A8B.A3E38D43@feri.de> <36c07187.1188208@news.gv2.com>
<36C09C99.6A48CF99@internet.de> <36c0ab34$0$16680@nntp1.ba.best.com>
<36C0DDD3.186A99BD@iname.com> <36c140e0$0$16688@nntp1.ba.best.com>
<86aeymbm9m.fsf@charvak.kla-tencor.com> <36c16ccf$0$16678@nntp1.ba.best.com>
X-Http-Proxy: 1.0 x1.dejanews.com:80 (Squid/1.1.22) for client 194.139.17.7
Organization: Deja News - The Leader in Internet Discussion
X-Article-Creation-Date: Wed Feb 10 15:27:27 1999 GMT
Newsgroups: comp.os.linux.advocacy,comp.os.ms-windows.nt.advocacy
X-Http-User-Agent: Mozilla/4.5 [en] (X11; I; SunOS 5.6 sun4c)
In article <36c16ccf$0$16...@nntp1.ba.best.com>,
"Boris" <XXXbo...@movil.comXXX> wrote:
>
> >Linux has kernel threads. In fact almost all modern UNIX supports
> >kernel threads.
> I think you are mistaken. What system call do you use to create a new thread
in Linux?
clone(2)
It has been in Linux since 1.3.xx, and is supported by linuxthreads and glibc
for years now.
But user threads and kernel threads are something completely different. Linux
had kernel threads (read: multithreading inside the kernel) since the very
beginning. What was added with clone was kernel support for user threads. User
threads, using a lightweight user-space threading method was supported with
linuxthreads even before clone(2) was added to the kernel.
You seem to talk about Linux 1.2. We are talking more about Linux 2.2 here.
Bernd Paysan
"Late answers are wrong answers!"
http://www.jwdt.com/~paysan/
-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/ Search, Read, Discuss, or Start Your Own
From: "Boris" <borisspa...@pleasemovil.com>
Subject: Re: Linux vs. NT: Benchmarks
Date: 1999/02/10
Message-ID: <36c1ff22$0$16681@nntp1.ba.best.com>#1/1
X-Deja-AN: 442970598
References: <36B49497.E3478308@internet.de> <79p850$vc7$1@nnrp1.dejanews.com>
<36C05A8B.A3E38D43@feri.de> <36c07187.1188208@news.gv2.com>
<36C09C99.6A48CF99@internet.de> <36c0ab34$0$16680@nntp1.ba.best.com>
<36C0DDD3.186A99BD@iname.com> <36c140e0$0$16688@nntp1.ba.best.com>
<86aeymbm9m.fsf@charvak.kla-tencor.com> <36c16ccf$0$16678@nntp1.ba.best.com>
<79s8gk$klt$1@nnrp1.dejanews.com>
X-MimeOLE: Produced By Microsoft MimeOLE V4.72.3155.0
X-Trace: nntp1.ba.best.com 918683426 16681 209.24.240.18
Reply-To: "Boris" <XXXbo...@movil.comXXX>
Newsgroups: comp.os.linux.advocacy,comp.os.ms-windows.nt.advocacy
It must be new development which I wasn't aware of. Anyway, NT had threads since it's
inception. What I mean by "kernel threads" is that each process has 1 or more threads
which are scheduled independently of each other by OS (unlike co-routines or smthg);
different threads can execute on different CPUs simultaneously. Writing multi-threaded
applications is natural way of programming on NT. Unlike Unix user threads which were just
co-routines traditionally (except for some platforms where they were mapped to "kernel
threads") NT threads were always *real* threads. We have some extremely powerful
thread-related system calls in NT (eg. IO completion ports).
My understanding was that most Linux apps create child processes in situations where NT
apps use multiple threads.
Even though Linux got support for multi-threaded apps finally (as you just said), I'm sure
most apps don't use those new features yet. It will probably take a while to upgrade.
Boris
Bernd Paysan wrote in message <79s8gk$kl...@nnrp1.dejanews.com>...
>In article <36c16ccf$0$16...@nntp1.ba.best.com>,
> "Boris" <XXXbo...@movil.comXXX> wrote:
>>
>> >Linux has kernel threads. In fact almost all modern UNIX supports
>> >kernel threads.
>> I think you are mistaken. What system call do you use to create a new thread
>in Linux?
>
>clone(2)
>
>It has been in Linux since 1.3.xx, and is supported by linuxthreads and glibc
>for years now.
>
>But user threads and kernel threads are something completely different. Linux
>had kernel threads (read: multithreading inside the kernel) since the very
>beginning. What was added with clone was kernel support for user threads. User
>threads, using a lightweight user-space threading method was supported with
>linuxthreads even before clone(2) was added to the kernel.
>
>You seem to talk about Linux 1.2. We are talking more about Linux 2.2 here.
>
>Bernd Paysan
>"Late answers are wrong answers!"
>http://www.jwdt.com/~paysan/
>
>-----------== Posted via Deja News, The Discussion Network ==----------
>http://www.dejanews.com/ Search, Read, Discuss, or Start Your Own
From: hou...@natlab.research.philips.com (Houben S.H.M.J.)
Subject: Re: Linux vs. NT: Benchmarks
Date: 1999/02/11
Message-ID: <houben.918732126@crow>#1/1
X-Deja-AN: 443203120
X-Nntp-Posting-Host: crow
Sender: n...@natlab.research.philips.com (USEnet Subsystem)
References: <36B49497.E3478308@internet.de> <79p850$vc7$1@nnrp1.dejanews.com>
<36C05A8B.A3E38D43@feri.de> <36c07187.1188208@news.gv2.com>
<36C09C99.6A48CF99@internet.de> <36c0ab34$0$16680@nntp1.ba.best.com>
<36C0DDD3.186A99BD@iname.com> <36c140e0$0$16688@nntp1.ba.best.com>
<86aeymbm9m.fsf@charvak.kla-tencor.com> <36c16ccf$0$16678@nntp1.ba.best.com>
<79s8gk$klt$1@nnrp1.dejanews.com> <36c1ff22$0$16681@nntp1.ba.best.com>
Organization: Philips Research Laboratories Eindhoven, Netherlands
Newsgroups: comp.os.linux.advocacy,comp.os.ms-windows.nt.advocacy
In <36c1ff22$0$16...@nntp1.ba.best.com> "Boris" <borisspa...@pleasemovil.com> writes:
>It must be new development which I wasn't aware of.
I believe Linux has had threads since 2.0.
>Anyway, NT had threads since it's
>inception.
Maybe NT has had threads longer than Linux. I don't know.
Anyhow, Linux certainly has threads now, and has had it for quite some time.
>What I mean by "kernel threads" is that each process has 1 or more threads
>which are scheduled independently of each other by OS (unlike co-routines or smthg);
>different threads can execute on different CPUs simultaneously.
That's exactly the kind of threads Linux has. They are scheduled just like
processes are.
>Writing multi-threaded
>applications is natural way of programming on NT.
...and using fork() is a natural way to do programming on unix.
>Unlike Unix user threads which were just
>co-routines traditionally (except for some platforms where they were mapped to "kernel
>threads")
Originally, unix didn't have "real" threads, that's certainly true.
But the original unix is more than 25 years old!
They actually managed to add *some* functionality in that time. ;-)
POSIX now defines a standard thread library (pthreads) which is
available on Linux. So there is now a standard unix way to do threads.
> NT threads were always *real* threads. We have some extremely powerful
>thread-related system calls in NT
>(eg. IO completion ports).
Don't know what these are, so cannot comment on them.
>My understanding was that most Linux apps create child processes in situations where NT
>apps use multiple threads.
That's correct, AFAIK.
>Even though Linux got support for multi-threaded apps finally (as you just said), I'm sure
>most apps don't use those new features yet. It will probably take a while to upgrade.
As I said before, Linux has had threads for quite some time now.
Let me elaborate on the thread vs. child process issues.
In NT, there is an essential difference between threads and processes;
creating new processes is rather expensive, because a completely
"fresh" memory image has to be created. Threads are the cheap alternative.
So it is logically that many applications depend on those light-weight threads.
However, in Linux (as in any unix), we have fork(). fork() creates a new
copy of the process, but *not* by copying the whole memory image at once.
The memory is only copied one one of the two new processes tries to write
to it. If they do not write to some memory page, then the memory remains
shared between the processes. This makes fork() much cheaper than the
NT system call for spawning new processes.
Besides, processes that have been fork()ed
can still share their read-only data and their file descriptors.
They can communicate with each other through SysV semaphores, message queues,
and shared memory.
This is sufficient for most server-type applications.
(Were the individual processes operate more or less independently.)
Since processes are still quite isolated from each other, this makes
multiprocessing much easier. IIRC, Solaris can efficiently run web servers
on SMP boxes with > 100 processors. Granted, Linux is still lacking in this
respect, but it is already better than NT.
Now, if you need really closely-intertwined threads/processes, then
threads are the way to go. And they are available. It is just that, for many
applications, processes are just more efficient.
Greetings,
Stephan
--
S.H.M.J. Houben
E-mail: hou...@natlab.research.philips.com
From: "Boris" <borisspa...@pleasemovil.com>
Subject: Re: Linux vs. NT: Benchmarks
Date: 1999/02/15
Message-ID: <36c92714$0$210@nntp1.ba.best.com>#1/1
X-Deja-AN: 444868399
References: <36B49497.E3478308@internet.de> <79p850$vc7$1@nnrp1.dejanews.com>
<36C05A8B.A3E38D43@feri.de> <36c07187.1188208@news.gv2.com>
<36C09C99.6A48CF99@internet.de> <36c0ab34$0$16680@nntp1.ba.best.com>
<36C0DDD3.186A99BD@iname.com> <36c140e0$0$16688@nntp1.ba.best.com>
<86aeymbm9m.fsf@charvak.kla-tencor.com> <36c16ccf$0$16678@nntp1.ba.best.com>
<79s8gk$klt$1@nnrp1.dejanews.com> <36c1ff22$0$16681@nntp1.ba.best.com>
<houben.918732126@crow>
X-MimeOLE: Produced By Microsoft MimeOLE V4.72.3155.0
X-Trace: nntp1.ba.best.com 919152404 210 209.24.241.38
Reply-To: "Boris" <XXXbo...@movil.comXXX>
Newsgroups: comp.os.linux.advocacy,comp.os.ms-windows.nt.advocacy
>However, in Linux (as in any unix), we have fork(). fork() creates a new
>copy of the process, but *not* by copying the whole memory image at once.
>The memory is only copied one one of the two new processes tries to write
>to it. If they do not write to some memory page, then the memory remains
>shared between the processes. This makes fork() much cheaper than the
>NT system call for spawning new processes.
NT has copy-on-write. This is the way VC debugger implements breakpoints.
NT spawning a new process isn't similar to fork. If you compare NT CreateProcess() with
Unix fork+exec you'll see that they are doing the same thing: loading executable,
loading/mapping DLLs. NT CreateProcess() has no business duplicating memory image of
parent process. So Unix fork implementation using copy-on-write doesn't provide any
advantages over CreateProcess() in this case.
Boris
From: jer...@netcom.com (Jeremy Allison)
Subject: Re: Linux vs. NT: Benchmarks
Date: 1999/02/16
Message-ID: <jeremyF79GI1.nM@netcom.com>#1/1
X-Deja-AN: 445026173
Sender: jer...@netcom13.netcom.com
References: <36B49497.E3478308@internet.de> <79p850$vc7$1@nnrp1.dejanews.com>
<36C05A8B.A3E38D43@feri.de> <36c07187.1188208@news.gv2.com>
<36C09C99.6A48CF99@internet.de> <36c0ab34$0$16680@nntp1.ba.best.com>
<36C0DDD3.186A99BD@iname.com> <36c140e0$0$16688@nntp1.ba.best.com>
<86aeymbm9m.fsf@charvak.kla-tencor.com> <36c16ccf$0$16678@nntp1.ba.best.com>
<79s8gk$klt$1@nnrp1.dejanews.com> <36c1ff22$0$16681@nntp1.ba.best.com>
<houben.918732126@crow> <36c92714$0$210@nntp1.ba.best.com>
Organization: ICGNetcom
Newsgroups: comp.os.linux.advocacy,comp.os.ms-windows.nt.advocacy
"Boris" <borisspa...@pleasemovil.com> writes:
>>NT system call for spawning new processes.
>NT has copy-on-write. This is the way VC debugger implements breakpoints.
>NT spawning a new process isn't similar to fork. If you compare NT CreateProcess() with
>Unix fork+exec you'll see that they are doing the same thing: loading executable,
>loading/mapping DLLs. NT CreateProcess() has no business duplicating memory image of
>parent process. So Unix fork implementation using copy-on-write doesn't provide any
>advantages over CreateProcess() in this case.
But Boris, what about when you want to do a fork() *AND
NO FOLLOWING EXEC* !
How does CreateProcess help you here ?
How can the Win32 API in NT even do this ?
(BTW: It *is* possible, I workd on the amazing code written
by Steve Chamberlin at Cygnus in Cygwin32 that does a fork()
for NT. Unfortunately the only way to do this in user space with
no kernel help is to implement it *without* copy on
write - thus it's horribly slow :-( ).
Also the NT POSIX subsystem can do this with kernel
help - unfortunately you can't use this from within
a Win32 app.....
Jeremy Allison,
Samba Team.
From: hou...@natlab.research.philips.com (Houben S.H.M.J.)
Subject: Re: Linux vs. NT: Benchmarks
Date: 1999/02/17
Message-ID: <houben.919255796@jackdaw>#1/1
X-Deja-AN: 445309616
X-Nntp-Posting-Host: jackdaw
Sender: n...@natlab.research.philips.com (USEnet Subsystem)
References: <36B49497.E3478308@internet.de> <79p850$vc7$1@nnrp1.dejanews.com>
<36C05A8B.A3E38D43@feri.de> <36c07187.1188208@news.gv2.com>
<36C09C99.6A48CF99@internet.de> <36c0ab34$0$16680@nntp1.ba.best.com>
<36C0DDD3.186A99BD@iname.com> <36c140e0$0$16688@nntp1.ba.best.com>
<86aeymbm9m.fsf@charvak.kla-tencor.com> <36c16ccf$0$16678@nntp1.ba.best.com>
<79s8gk$klt$1@nnrp1.dejanews.com> <36c1ff22$0$16681@nntp1.ba.best.com>
<houben.918732126@crow> <36c92714$0$210@nntp1.ba.best.com> <jeremyF79GI1.nM@netcom.com>
Organization: Philips Research Laboratories Eindhoven, Netherlands
Newsgroups: comp.os.linux.advocacy,comp.os.ms-windows.nt.advocacy
In <jeremyF79GI1...@netcom.com> jer...@netcom.com (Jeremy Allison) writes:
>Also the NT POSIX subsystem can do this with kernel
>help - unfortunately you can't use this from within
>a Win32 app.....
I've heard this before, and find it somewhat strange.
Is this just because MS wanted to discourage people
from using the POSIX API, or is there some real (i.e. technical)
reason for this?
Greetings,
Stephan
--
S.H.M.J. Houben
E-mail: hou...@natlab.research.philips.com
From: jer...@netcom.com (Jeremy Allison)
Subject: Re: Linux vs. NT: Benchmarks
Date: 1999/02/17
Message-ID: <jeremyF7B7Fw.1n7@netcom.com>#1/1
X-Deja-AN: 445383990
Sender: jer...@netcom13.netcom.com
References: <36B49497.E3478308@internet.de> <79p850$vc7$1@nnrp1.dejanews.com>
<36C05A8B.A3E38D43@feri.de> <36c07187.1188208@news.gv2.com>
<36C09C99.6A48CF99@internet.de> <36c0ab34$0$16680@nntp1.ba.best.com>
<36C0DDD3.186A99BD@iname.com> <36c140e0$0$16688@nntp1.ba.best.com>
<86aeymbm9m.fsf@charvak.kla-tencor.com> <36c16ccf$0$16678@nntp1.ba.best.com>
<79s8gk$klt$1@nnrp1.dejanews.com> <36c1ff22$0$16681@nntp1.ba.best.com>
<houben.918732126@crow> <36c92714$0$210@nntp1.ba.best.com>
<jeremyF79GI1.nM@netcom.com> <houben.919255796@jackdaw>
Organization: ICGNetcom
Newsgroups: comp.os.linux.advocacy,comp.os.ms-windows.nt.advocacy
hou...@natlab.research.philips.com (Houben S.H.M.J.) writes:
>In <jeremyF79GI1...@netcom.com> jer...@netcom.com (Jeremy Allison) writes:
>>Also the NT POSIX subsystem can do this with kernel
>>help - unfortunately you can't use this from within
>>a Win32 app.....
>I've heard this before, and find it somewhat strange.
>Is this just because MS wanted to discourage people
>from using the POSIX API, or is there some real (i.e. technical)
>reason for this?
No there is no technical reason for not exposing
this in the Win32 API. It's political. MS don't
want you to write POSIX apps. MS doesn't even
want you to have Win32 features that POSIX
apps could use. POSIX apps are *portable* (shudder :-).
Thus, if you decide to change OS vendors, you can
(theoretically) re-compile and move to another
platform (ie. from Solaris to Linux for example).
This doesn't fit well at all with MS's "Windows everywhere"
philosophy.
Having said that, POSIX complience is a checkbox
for certain US government purchasing requirements.
These agencies cannot legally purchase an OS without
a level of POSIX complience (note that these agencies
don't have to *use* the POSIX system, it just must
be there in order to get the theorectical
portability of apps such a system supplies). MS wants
to see NT into these agencies (that's where a lot of
the money is - look at the US Navy's NT decisions for
example) - thus NT must have a POSIX system.
Note it doesn't have to be *useable* and only
has to fit the minimum level of POSIX complience
to pass the purchasing requirement documents.
Just my cynical 2cents-worth :-).
Jeremy Allison.