Some (okay, all) of my server machines are getting a bit old and creaky and I'm looking at some much needed hardware upgrades at our small (very small) business. I run Slackware on almost everything, including X terminal workstations, except for one PC DOS and two Win 98 boxes that run legacy industrial apps and keep their data in sync via a slackware Samba/NFS/print server.

I've been thinking about dual core/dual processor machines, particularly as an upgrade to the PIII XDMCP server that all the X terminals connect to, but also on the Samba server.

A friend loaned/gave me an old 200 MHz Pentium Pro dual processor box with 256 MB and both IDE and scsi to experiment with and I set it up with Slackware 11 and the generic 2.6.17 smp kernel from "extra." Seems to work fine, but the thing is, it's old too and I really don't have much to compare it with. By my casual observations, I'd say this dual processor box seems to multi-task with KDE and desktop apps as well or better than a much faster PII with the same amount of ram, but that could be the result of cache, or scsi, or my imagination, or whatever. I wouldn't know how to run, or interpret, any sort of benchmarking test if my life depended on it...I'm the administrator only because I can't run away very quickly.

So, the question being, have any of you really had an opportunity to compare smp performance versus single core/single processor performance in a multi-user environment running typical desktop apps like openoffice in KDE, and if so, did you conclude that this sort of use warrants the additional expense of SMP, or did the results persuade you that a faster, single processor machine, was the way to go?

As I said, this is a very small business, and number crunching/database stuff is minimal.

Funny...

Well, to the best of my knowledge, said program has to take advantage of using 2 procs to begin with. But, I'm not overly knowledgeable on this subject. The only real performance gain I see on my core 2 duo machine is when I'm compiling something with MAKEOPTS set. But then, that's a function of GCC that allows for the use of a multi core system...

But, I think even a newer single proc box would outshine what your running now. Maybe you should rather focus on an actual MHZ speed increase rather than multiple procs? Just a thought. Atleast that wouldn't break the bank and you would definately notice some performance gains.

Sort of what I was wondering...an application like openoffice would probably have to be built to take advantage of dual processors...BUT...

Just as a "for instance"...suppose I have four or five users logged in on the XDMCP/application server. Most of them are running, say, Firefox and openoffice, at the same time...would there be a division of duties between the two processors, so that even though the application might only utilize one processor, different sessions might be operating on different processors for an overall performance gain?

Well, I have a desktop and a laptop. They both have 1GB RAM, and P4 processors. The desktop has a 3.06 GHz HT (SMP), and the laptop has 2.80 GHz. Other than that they are quite similar (except graphics cards). The desktop is much faster than the laptop in everything. Now, I know a P4 doesn't really have 2 cores, but it does have 2 virtual processors. This helps most in multitasking and I estimate results in a 20-25% performance increase ...