Samba beats Windows (vnunet) [LWN.net]

Vnunet covers tests of Samba vs. Windows. "Samba 3's scalability is as impressive as its performance. While Windows performance scales up well initially, it then drops off quickly as more clients access the server. In contrast, Samba 3 offered excellent throughput up to the limit of our test, conducted on a low-spec Intel server."

to post comments

Samba beats Windows (vnunet)

Posted Oct 13, 2003 22:31 UTC (Mon) by einstein (subscriber, #2052) [Link] (4 responses)

I haven't seen any figures - I'd be very curious to see the absolute numbers. For instance, how do the linux results achieved on this test compare to the linux results from the microsoft-sponsored test? That would be very instructive...

Samba beats Windows (vnunet)

Posted Oct 14, 2003 1:17 UTC (Tue) by gdt (subscriber, #6284) [Link] (1 responses)

...how do the linux results achieved on this test compare to the linux results from the microsoft-sponsored test?

I'm assuming that you are referring to the Mindcraft tests of 1999. These are now too old to be of much interest as both Windows and Linux have moved on. For example, the Mindcraft tests used the Linux 2.2 kernel.

Let's look at the world through the eyes of an IT manager.

It's Windows' movement which is of interest to IT managers as it leads to a strategic headache. Software support for their Windows NT 4.0 file servers is coming to an end. The choices are to upgrade to Windows Server 2003, which is expensive and to operate well requires implementing Active Directory; or to cross-grade to Linux, which can be cheap and which does not require a change to authentication mechanisms.

Samba's performance is a tactical consideration informing the strategic choice. The important point for the IT manager is that Samba does not perform poorly, no matter who's benchmark you believe. So the capacity planning numbers don't change too much between Windows NT and Samba.

The capacity planning numbers for Windows 2003 Server are different. Going with Windows 2003 Server implies a hardware upgrade for most sites. So in going with Samba on Linux IT managers gain a further choice: cross-grade to Samba on Linux and retain the same hardware; or cross-grade to Samba on Linux on new hardware and do some server consolidation.

The IT Week performance figures indicate that Samba's performance is >4x Windows 2003 Server. A well-choosen hardware upgrade could replace ten or more existing file servers. Unlike a Windows-based server consolidation those existing servers need not be trashed, they can run Linux and do less intensive tasks such as print serving, DNS and DHCP.

Samba beats Windows (vnunet)

Posted Oct 14, 2003 13:52 UTC (Tue) by erat (guest, #21) [Link]

The choices are to upgrade to Windows Server 2003, which is expensive and to operate well requires implementing Active Directory; or to cross-grade to Linux, which can be cheap and which does not require a change to authentication mechanisms.

You hit on a subject (and all too quickly brushed past it) that I'm sure will tint IT folks' evaluations of WS2K3 and Linux/Samba: authentication, or more accurately, management of identity data.

There was a time when machines running NT did little more than serve files, share printers, and perhaps run web servers (the latter still blows my mind. Why would anyone do such a thing??). Doing side-by-side comparisons of Windows servers and Linux servers was easier because the functionality expected of each was sort of limited. However, with identity management, evaluations become a bit more complicated.

I've used WS2K3 and have done a fair amount of work with Active Directory (along with the cross-domain and cross-forest goofiness that goes along with it). As much as I hate to say it, I think it's a decent system. Setting up AD is fairly straightforward. And with apps like Vintela Authentication or with "experimental" configurations of LDAP/DNS/Samba/pam_smb you should be able to use identity data from AD with Linux. I won't get into an MS sales pitch because I'm sure they don't need/deserve the advertising, but I did want to point out that the identity management stuff in WS2K3 may actually be a selling point, not a hinderance.

Onward...

The new version of Samba has some capabilities of dealing with identity management (winbind?), and I believe it works with AD. I haven't looked into this; I'm merely parroting information that I heard elsewhere. A truly fair comparison between WS2K3 and Linux/Samba would need to include a comparison of how identity management works with AD and winbind, or even how Linux/Samba can be used to completely replace AD.

As for speed... Compare how Linux/Samba stacks up against a WAN with WS2K3 servers that utilize cross-forest trusts (including identity firewalls), global catalog sync'ing across the WAN, etc. I'm not saying one will outshine the other because I haven't done any such benchmarking. I do think that kind of comparison is more valuable than one that simply places Linux/Samba on top just because it does a few things quickly.

A few more details, plus links to older benchmarks

Posted Oct 14, 2003 2:53 UTC (Tue) by dank (guest, #1865) [Link] (1 responses)

I have a bit more info at http://www.kegel.com/nt-linux-benchmarks.html#file

A few more details, plus links to older benchmarks

Posted Oct 14, 2003 9:01 UTC (Tue) by torsten (guest, #4137) [Link]

http://www.kegel.com/nt-linux-benchmarks.html#file

Windows advantage?

Posted Oct 14, 2003 7:28 UTC (Tue) by bojan (subscriber, #14302) [Link] (1 responses)

It would be interesting to find out from MS people what is now going to be the "Windows advantage". They have ads all over TV in Oz about Windows 2003 Server saving businesses hundreds of thousands of dollars. This piece on Vnunet sure doesn't look like something that would support those ads. My experience also tells me that Linux machines running Samba usually don't have to be touched for months, which would suggest a very low TCO. What are we going to hear this time around?

M$ == "market-speak" ?

Posted Oct 14, 2003 11:26 UTC (Tue) by Duncan (guest, #6647) [Link]

I believe they can (will) take two approachs (well, say two and a half..), continuing a
pattern that's becoming well defined and quite predictable.

First, they can continue claiming the hundreds of thousands of dollars saved off of
traditional Unix, which is where they are making the comparisons for those claims
now, after all, and valid they indeed are (tho they of course neglect to mention the
even LOWER cost Linux alternative, but that's what marketing is SUPPOSED to
do, accent the positive and ignore the negative unless forced to respond by a direct
question on it, so there you have it).

Second, and this is something we're seeing more of lately, they make the sometimes
legit claim that Linux admins are rarer and may be rather more costly. (Again, they
focus on the positive and ignore the negative, that fewer of said admins are normally
needed, but again, that's marketing, telling the truth, but not the WHOLE truth,
and CERTAINLY not "nothing BUT the truth," especially when it's MS
marketing..)

Third (or second and a half), as an offshoot of two, they sponsor "corner-case" case
studies where given narrow enough definitions and a well enough defined and
limited corner case, they may actually come out on top. Thus, for instance, one such
study covered five-year cost instead of the three- or even two-year upgrade cycle
MS prefers when comparing the "savings" of Licensing 6.0, such that the initial cost
of the MS licenses were spread out over a longer period to make them less onerous
as compared to the lower or free initial licensing cost of the Linux solution, while at
the same time emphasizing the alleged higher cost of Linux admins, to pull MS
ahead under a very tightly controlled and narrowly defined corner case scenario.

Duncan

Samba beats Windows (vnunet)

Posted Oct 14, 2003 8:58 UTC (Tue) by ayeomans (guest, #1848) [Link] (1 responses)

Here is a more technical version of the article. Sorry, no graph or numbers.

Graph now available

Posted Oct 14, 2003 14:47 UTC (Tue) by ITWeek (guest, #16003) [Link]

You can see a graph of IT Week's test results here.