Technology and Trends
 USENET Archives
  
From: "Boris" <borisspa...@pleasemovil.com>
Subject: Phase #2 of Mindcraft tests. Linux in deep s***.
Date: 1999/06/25
Message-ID: <37743005$0$219@nntp1.ba.best.com>#1/1
X-Deja-AN: 493989976
X-Priority: 3
X-MimeOLE: Produced By Microsoft MimeOLE V5.00.2014.211
X-Trace: nntp1.ba.best.com 930361349 219 209.24.240.229
X-MSMail-Priority: Normal
Reply-To: "Boris" <XXXbo...@movil.comXXX>
Newsgroups: comp.os.linux.advocacy,comp.os.ms-windows.nt.advocacy

Gee, Jeremy Allison. I new that he's full of s*** right away.
Here are results of open Linux vs. NT tests arbitrated by PC labs. 
Mindcraft and Linux representatives were present during those tests.
NT beats Linux by WIDE margin, especially on higher end hardware.
http://www.zdnet.com/pcweek/stories/news/0,4153,1015266,00.html

Boris

From: jer...@netcom.com (Jeremy Allison)
Subject: Re: Phase #2 of Mindcraft tests. Linux in deep s***.
Date: 1999/06/27
Message-ID: <7l3q40$rq7@dfw-ixnews16.ix.netcom.com>#1/1
X-Deja-AN: 494279294
References: <37743005$0$219@nntp1.ba.best.com>
Organization: Netcom
X-NETCOM-Date: Sat Jun 26  7:09:36 PM CDT 1999
Newsgroups: comp.os.linux.advocacy,comp.os.ms-windows.nt.advocacy
NNTP-Posting-User: jeremy

"Boris" <borisspa...@pleasemovil.com> writes:

>Gee, Jeremy Allison. I new that he's full of s*** right away.
>Here are results of open Linux vs. NT tests arbitrated by PC labs. 
>Mindcraft and Linux representatives were present during those tests.
>NT beats Linux by WIDE margin, especially on higher end hardware.
>http://www.zdnet.com/pcweek/stories/news/0,4153,1015266,00.html

Well thanks for thinking it's all my fault Boris, I'm flattered :-).

Just one point I'd like to make though, before you go off
on your victory dance :-).

The problem revealed in this benchmark was the same one
that the previous PC Week benchmark told us - that the
Linux kernel TCP stack needs more multi-threading work for
MP systems. However, the Samba architecture is still 
provably a fast one. The reason I know this can be found
on this web page :

http://www.zdnet.com/pcweek/stories/jumps/0,4270,401974,00.html

in this comment about Samba running on Solaris x86 :

"In this configuration, the powerful capabilities of Solaris 7's
networking kernel were unleashed--to the tune of 360Mbps on
NetBench."

Now correct me if I'm wrong, but this (using Samba)
is a faster result (on the same PC Week network with
the same clients) than the second benchmark got (338Mbps)
using NT with a similar 4 processor Xeon box.

Now the Solaris result was after removing the disk subsystem
bottleneck on Solaris, so isn't really a fair comparison.
However, it does show that given a good SMP aware TCP
stack, Samba will perform at the same or better level than
an in-kernel SMB implementation.

The Linux TCP stack needs work, and that is indeed being
done. As for the other numbers that RedHat and Penguin
got, I'm suprised they are so low, but not being there
this time it's difficult for me to comment on the specifics
there.

Cheers,

	Jeremy Allison,
	Samba Team.

From: s...@choenet.com.remove.this.com (Sang K. Choe)
Subject: Re: Phase #2 of Mindcraft tests. Linux in deep s***.
Date: 1999/06/27
Message-ID: <37758ae7.391207140@news>#1/1
X-Deja-AN: 494376805
Content-Transfer-Encoding: 7bit
References: <37743005$0$219@nntp1.ba.best.com> <7l3q40$rq7@dfw-ixnews16.ix.netcom.com>
X-Original-NNTP-Posting-Host: 192.168.100.107
Content-Type: text/plain; charset=us-ascii
X-Complaints-To: newsabuse@remarQ.com
X-Trace: 930475814.812.77 XL4TBL80I701BD166C qube-01.us-ca.remarq.com
Organization: Posted via RemarQ, http://www.remarQ.com - The Internet's Discussion Network
MIME-Version: 1.0
NNTP-Posting-Date: Sun, 27 Jun 1999 09:30:14 GMT
Newsgroups: comp.os.linux.advocacy,comp.os.ms-windows.nt.advocacy

On 27 Jun 1999 00:09:36 GMT, jer...@netcom.com (Jeremy Allison) wrote:

>"In this configuration, the powerful capabilities of Solaris 7's
>networking kernel were unleashed--to the tune of 360Mbps on
>NetBench."
>
>Now correct me if I'm wrong, but this (using Samba)
>is a faster result (on the same PC Week network with
>the same clients) than the second benchmark got (338Mbps)
>using NT with a similar 4 processor Xeon box.
>
>Now the Solaris result was after removing the disk subsystem
>bottleneck on Solaris, so isn't really a fair comparison.
>However, it does show that given a good SMP aware TCP
>stack, Samba will perform at the same or better level than
>an in-kernel SMB implementation.

No, what this shows is that under Solaris 7, if the disk subsystem is
removed from the equation as a bottleneck, the performance goes from
206Mbps to 350Mbps.  In otherwords, the result that should be used to
compare Solaris 7 + Samba to NT would be 206Mbps to 338Mbps--a
difference of about 132Mbps.  Hardly what I would call the same nor a
better level than "an in-kernel SMB implementation".

Of course, if you insist on using the RAM disk version of the Solaris
benchmark, you should do the same with NT (yes a RAM disk is possible
under NT) to eliminate NT's disksubsystem from the equation.

From: jer...@netcom.com (Jeremy Allison)
Subject: Re: Phase #2 of Mindcraft tests. Linux in deep s***.
Date: 1999/06/27
Message-ID: <7l5ouo$1aq@dfw-ixnews17.ix.netcom.com>#1/1
X-Deja-AN: 494488069
References: <37743005$0$219@nntp1.ba.best.com> <7l3q40$rq7@dfw-ixnews16.ix.netcom.com> 
<37758ae7.391207140@news>
Organization: Netcom
X-NETCOM-Date: Sun Jun 27  1:02:00 PM CDT 1999
Newsgroups: comp.os.linux.advocacy,comp.os.ms-windows.nt.advocacy
NNTP-Posting-User: jeremy

s...@choenet.com.remove.this.com (Sang K. Choe) writes:

>Of course, if you insist on using the RAM disk version of the Solaris
>benchmark, you should do the same with NT (yes a RAM disk is possible
>under NT) to eliminate NT's disksubsystem from the equation.

Of course. What I am really interested in is a Solaris+Samba
NetBench run without the old ufs filesystem, with something
like Veritas running on Solaris. That would give more info.
Only on x86 hardware of course. When run on proprietary
hardware, using a proprietary UNIX (IRIX for instance) Samba
will happily scale up as you add processors.

The ramdisk result does show however, that the architectural
design of Samba is sound (it's the same design as used in the
Vantive customer service application server, which I also
had a hand in, and that application scales to many thousands
of users).

Regards,

	Jeremy Allison,	
	Samba Team.