Network Benchmarking

[maximal]

Hello,

I need some help figuring out how to benchmark network performance in a medium size network (250 hosts, 3 servers). If anyone has any suggestions on benchmarking applications and/or methodologies in popular use for this task please advise. I know that motherboard built-in Gbit NICs are not all they are cracked up to be, but even major server OEMs (DELL, HP) use them, so i've been leaning towards installing multi-port Gbit PCI-X or PCIe cards and disabling built-in ones altogether. I've been looking at Inter Pro series Gbit NICs and they seem to look good on paper. This of course will not pass muster with IT brass unless I have some solid benchmarking numbers to back my conjecture up, hence the cry for help. And for good measure there were some performance targets that were thrown at me such as: >=25Mbit/s full-duplex bandwidth per host per server.

A bit of background:

Company is going through a total hardware IT infrastructure overhaul, IT recourses are starched pretty thin (explains why a DB admin, me, was put in charge of network performance validation).

 

[Lemon law]

I tried to stay off this thread due to the Peter principle and my level of incompetence, but after 24 hours, I think you deserve some sort of answer.

And I very much wonder if the specific benchmarks you use matter all that much because you are asking a company to make a huge dollar investments in a very big network.
And then are wondering if changing network cards from on board to third part will help.

If I were you, I would be asking for a small R&D budget, buying a few pieces of equipment, and then trying to quantify if the performance improvement on the one server and the one computer that got the better NIC improves through put or does not seem to justify that 250 computer investment. That way you can try quite a few alternate ideas without investing bundle in equipment that may not be worth the money or do the job down the road.

But I agree, you need the benchmarking programs better network experts than I on this forum should be suggesting.

 

[ScottMac]

What kind of benchmarks are you looking to obtain?

Bandwidth studies? Server performance?

Are you going to present this as "Well, we'll PROBABLY see XXXX in this range"....based on MARKETING?? :Q

As LL mentions, you need to set up some kind of pilot with a series of tests that can be projected into a larger domain. At least then you can say 'Based on actual tests we saw XXXX performance; by expanding the {franastatic co-incidence variations} we can safely assume that expansion within these limits will maintain similar performance"

Guessing don't cut it. It doesn't matter how starched your recources are .

When the budget is thin, you can't afford to guess; there'll be no more money to correct bad guesses, then you're (well, your users are) really screwed.