Samba Beats Windows IT Week Labs Test Results

jmhowitt writes "Tests by IT Week Labs show the latest version of the open-source Samba file and print server software is 2.5 times faster than Windows Server 2003 in the same role. The news comes as many firms are grappling with the consequences of Microsoft ending support for NT4, coupled with uncertainty about when Microsoft will next update Windows. The performance difference between Windows Server 2003 and Samba 3 has increased dramatically compared with Samba 2 and Windows 2000 Server."

Comments

The best thing about Samba...

[_Hellfire_]

Apart from how bloody quick it is is the fact that you can log every transaction. This is immensely useful in a mission critical environment when you have to figure our exactly why one person in particular out of the entire network is having trouble. Check your Samba logs and 99.9% of the time your answer will be there.

As a system administrator I appreciate having that level of scrutiny on any network I take care of.

Re:The best thing about Samba...

[jschrod]

OK, flamebait, and not funny.

I have worked with several large organizations that use Samba to serve files to 10,000s of workstations. It works much better than W2K servers, not to speak of NT servers. Samba on a Sun HA cluster (e.g., F15K systems) is an appropriate choice for file sharing that is considered critical.

And please note that I don't say this because I'm anti-Microsoft. In fact, I'm the CEO of a company that is a MS partner, and I am very satisfied with MS' support for our company. (It's much better than that of most other proprietary vendors, though not in the league of Oracle and their ilk.) I regularily plan and deploy heterogenous infrastructures for very large installations (i.e., > 50,000 users) and can back up my opinion with real-world experience from several places. What are your credentials?

Re:Best choice for the job?

[Make]

you mean, there is a kernel implementation of the NFS daemon. There is also one which runs purley in userspace, but on Linux, it is not used very much anymore nowadays.

There is smbclient in userspace (which is similar to an FTP client), but if you want to mount an SMB share into the linux VFS, you need the kernel module - like you need the NFS kernel module if you want to mount an NFS filesystem.

The numbers....

[Bill_Mische]

There are a lot of people asking about the numbers.

Well, I get the print edition of IT Week and the numbers are there on page 19, in the form of a lovely little graph. The (print) article says they used a HP ProLiant BL10 eCLass Server (900MHz PIII, 40Gb ATA, 512Mb Ram) and goes in to a little detail about the benchmarking software used.

I couldn't see a copy of the article on their website but you can download an electronic copy, in some god-forsaken windows only ebook format, from www.itweek.co.uk/ebook.

Re:Excuse me...

[Frobnicator]

A 2.5 x speed advantage can only mean one thing: misconfigured Windows machines.

You might be interested in this article describing the implementations of SMB and Samba, as well as the documentation list at samba.org. Additionally, Microsoft has had many issues with the LMB, DMB, and backup browser implementations, partly due to historical network decisions and old networking bugs and highly visible on volitle networks. Some of the difficulties with SMB include methods of caching the data and cache consistancy, the inconsistant rate of refereshes done within the network, (on windows) the auto-detection and auto-creation of LMB and LMB-backup nodes, and the presumption of the underlying network.

I'll focus on the network cache consistancy problem since that's the one I've had problems with. I don't know about the general speed issue (what speed are you referring to? throughput? Resource availablility? Master Browser updates? connection speed and concurrency under a heavy user load? ) I have experienced all kinds of problems with a highly volitile network, with programmers running multiple OS's inside of virtual machines. These virtual OS's need to be frequently restarted, meaning the network is constantly gaining and droping objects.

A prime example of Microsoft's bad cache coherency problem is that if an object is deleted or removed from the network, the information can take over an hour to propogate through the entire network. The worst case isn't nearly as bad in the pure-Samba implementation, but the difficulty remains. This failure means that newly added resources aren't immediately visible on the network, or recently removed resources take a long time to be removed, and show up as errors when you try to access them. Or the object can be visible on some machines, but not available on others.

When there is a high level of volitility on the network (machines being frequently rebooted or shut down, network re-wiring, etc.) this can really plague any SMB or CIFS network, but is especially hard on Windows boxes, and more so the older your Windows implementation. Problems are exacerbated if either the LMB or LMB-backup system is the one going back up and down, because the Windows boxes will respond less-quickly to the problem; this results in further instability for the SMB network, since critical nodes are not available, propogate incorrect data, and take longer to reconfigure.

As you mentioned, the Samba boxes are faster than the Windows boxes, but not as big of a difference as you experience. You said you have "a LAN full of Win2000/XP boxes", which probably means they are on most or all of the time. Is it unreasonable to assume that the author has a more volitle network, or is otherwise more prone to speed impairment issues?

frob

Re:Nice advertising

[AstroDrabb]

Sadly the MS Empire does not allow you to release benchmark stats for their products. You agree to this when you use their products through their EULA. I am sure IT Labs doesn't want to get crap over it. Then agian, maybe they will find a way to post the numbers.