samba - Feb 1999 - Benchmark results.

Hi all,

	For people who are looking for some objective
numbers to help recommend Samba to their employers (I
know there are some of you on this list :-) you might
want to look at the following couple of articles.

The first one is in Smart Reseller (a USA trade press
magazine) at :

http://www.zdnet.com/sr/stories/issue/0,4537,2196106,00.html

titled : "The Best Windows File Server: Linux!".

It covers Samba on Linux vs. NT on a single processor
system (and confirms the rumours I've heard that Linux
+ Samba outperform NT with more than 12 users).

The second becnhmark I must confess I helped with.
This was done at PC Week (another USA trade press
magazine) and was in a large server configuration.

This was a VA Research 4 x 450MHz Xeon processor
machine, with 18GB of storage in a RAID 5 configuration
and 2GB of memory.

This article may be found at :

http://www.zdnet.com/pcweek/stories/news/0,4153,387766,00.html

titled: "Linux: Enterprise-ready".

There are no comparitive NT numbers in this benchmark.

The reason for that is that NT on the multiprocessor box performed
so poorly in the test (a factor of 5 slower than Samba) that
both PC Week and myself were convinced that something was wrong
with the NT tuning.

The problem was that NT refused to use more that 300MB
of memory for file cache (despite tweaking the only 
public setting that NT allows to modify this). Even
searchin all Microsoft technical references and Web
pages on NT tuning did not allow us to find anything
to force NT to use more memory for disk cache.

On the Linux side I used to documentation in
/usr/src/linux/Documentation/vm 

to set the disk cache size to 80% of all available
memory.

Hope these numbers are interesting to people - use
them as you will :-) :-).

Regards,

	Jeremy Allison,
	Samba Team.

--------------------------------------------------------
Buying an operating system without source is like buying
a self-assembly Space Shuttle with no instructions.
--------------------------------------------------------

Jeremy Allison wrote in samba SAMBA digest number 1966:

	> 	For people who are looking for some objective
	> numbers to help recommend Samba to their employers (I
	> know there are some of you on this list :-) you might
	> want to look at the following couple of articles.
	> 
	> The first one is in Smart Reseller (a USA trade press
	> magazine) at :
	> 
	> http://www.zdnet.com/sr/stories/issue/0,4537,2196106,00.html
	>
	> titled : "The Best Windows File Server: Linux!".
	>
	> It covers Samba on Linux vs. NT on a single processor
	> system (and confirms the rumours I've heard that Linux
	> + Samba outperform NT with more than 12 users).
	 
For those interested in what the test was run on:
>From the accompanying sidebar at:
	http://www.zdnet.com/sr/stories/issue/0,4537,387506,00.html
How We Tested
Our test platform was a pair of identically configured and outfitted 266MHz
PCs. Each had 64MB of memory and a 4GB IDE disk drive. An Intel Corp.
EtherExpress Pro 100B network interface card connected the server to our
client network, which is driven by a pair of Synoptics (now Nortel Networks)
28115 Fast Ethernet switches. All tests ran at a network speed of 100Mbps.
In each instance, we performed a fresh install of the operating system. We
let the operating system decide the default disk configuration in each case.
In other words, we didn't try to tweak any settings. (Our guess is that
results would have been even more embarrassing if we had.)

We measured the throughput of up to 30 clients for each server. The clients
were
a mix of 166MHz, 200MHz and 233MHz machines running Windows 95. Each of the
clients ran the WebBench 2.0 static test workload to measure http server
performance and NetBench 5.01 for Server Message Block (SMB), or file
service, testing. Both benchmarks are available free from www.zdbop.com.

We configured each of the Linux boxes to run Samba (the SMB server) and
Apache
(the Web server)--but that's it. We weren't running DNS or Sendmail.
Since
they were being evaluated as servers, the Linux boxes were not running the X
display system. The NT box was running Internet Information Server 4.0 under
NT 4.0 with Service Pack 4. Again, no additional services were running.

According to ZDLabs' results (see test charts), each of the commercial Linux
releases ate NT's lunch. Our tests also revealed that Apache for OpenLinux
is superior to Apache for Red Hat and SuSE. Moreover, Samba for Red Hat
scales better than its counterparts.

Dave Lawson
Unix System Admin
Menasco Aerospace
Oakville, Ontario