Re: [tlug] switch for small cluster

[Curt Sampson <cjs@example.com>]

On 2008-06-04 10:00 +0200 (Wed), Michal Hajek wrote:

> Since the switches will be used in a cluster, we do not need any
> software enhacements whatsoever. We just need pure speed. And in this
> case I would say that the cheep switches would do the same service. 
> 
> Could you please comment on this idea? 

Well, you've got a pretty massive price gap between cheap switches
and "good" ones. You can get a cheap 32-port gigabit switch for, I'm
guessing, a few hundred bucks (<50,000 yen). You want a good Cisco
32-port gigabit switch, you'll pay (I'm guessing again) 200,000 yen at a
minimum.

So your issue here is an economic one: is it worth it to know you're
probably (not certainly) getting pretty good speed without worrying
about it? Or are you willing to spend some time and effort to figure out
what you're getting for your money?

If the latter, buy the cheapest switch you can find, and benchmark that,
and then you'll have the ability to benchmark others.

If cost is no object, just buy a known-reasonably-good expensive one.

If you're not sure, well, take the cheap way out, and buy a cheapish one
(but maybe not a no-name brand), and wait to see who complains. (Keeping
in mind that if there are complaints, you're back to this same decision,
but without this option.)

Stephen can tell you more about this I'm sure, but it turns out that
this is a basic issue with buying data-centre gear: it's often more
far more expensive to figure out how well something is working than it
is just to (depending on your situation) to buy stuff that (might be)
guaranteed to work well or to buy stuff that's cheap and hope.

cjs
-- 
Curt Sampson       <cjs@example.com>        +81 90 7737 2974   
Mobile sites and software consulting: http://www.starling-software.com