So yesterday my boss asked me if I thought it would be a good idea if we got some new hardware for an upcoming project (he feels the project is of a size where the additional expense will be somewhat justifiable). 'Absolutely' was my surprised reply as I'm just a wee codemonkey and not very high on the ol' budget totem pole. So, this morning he informs me that he's putting in for 2 Sun Fire 280R's. I was really disappointed as I was hoping for one of these babies :-)
That's what led me to the conclusion that any software development shop needs to buy hardware early and often. By the time we eventually got some auxilliary Linux servers for building and testing, we had wasted 4-6 months in discussion and procurement (until the VP demanded we start using more Linux boxes!), which trickled over into schedule creep because we couldn't build and test quickly enough. That also led to long lead times to find, isolate, and fix bugs (and test bugfixes).
Re:New hardware
vek on 2002-02-26T21:02:52
I once had a boss who insisted we develop on Solaris...
Right, you should see our server room, it's like a advertisement for Sun. If it ain't Solaris then don't bother seems to be the motto.
The problem was that our development server was quite overloaded..., that Pentium II consistently outperformed our Solaris box...
We have the same issue here. We've got a couple of old(er) Enterprise 250's as our dev boxes. My Linux desktop outperforms them by a fair margin. The powers that be are brinwashed into using Sun hardware for everything here!Re:New hardware
ziggy on 2002-02-26T21:15:34
It's not so much that Solaris is slow, it's more that any computer you buy this week will outperform anything that's been installed and running for a year. That was the point I was trying to make to my management lo those many jobs ago. And if your (Solaris) systems are over 18 months old, then lord help you, because you've probably missed at least one doubling of Moore's law with all of that uptime.;-) The approach I used was a focus on that observation, prove that it was true with a moderately old x86 Linux box, truer with a recent vintage x86 Linux box, and then show the price comparisions; something under $10,000 was within the discretionary budget, while upgrading the E450 (without really making it faster cost) was $250,000 that was already spent (and still underperforming).
It also helped that we had a series of "senior developers" running Linux on their desktops writing software in Perl and C that just ran on either Linux or Solaris (or AIX or Tru64).
In my circumstances, we were developing a database engine and webapps using that database engine. Management already knew about CPU bound vs. disk bound vs. memory bound application constraints, and it was clear that what we needed most was lots of cheap disk with a reasonable amount of memory and a reasonably modern CPU. Cutting corners actually helped us because not enough disk was more of a factor than slow disk (even if those cut corners led to an overall faster performing box).
