You guys go through servers and hardware like wildfire through brush. Do yourselves a favor, take that money, save it, and then spend it hiring someone to do professional code for your site. You've been over-taxing your resources for years and you know it. You've known it even since before 'project Ferrox' was first boasted years ago. Throwing hardware at the problem again is just another patch job. You have a developer on your team who once told me he could reprogram the entire site in a week, and here it's been years and one missed expectation of an upgrade after another. Seriously, save your cash and put it into a professional or accredited programmer. If you can't afford that, then at the very least you should be hiring a professional consultant to give real advice. You're way past overdue for a major software upgrade. Both the users of FA and your colo will appreciate this.
Hello Myr, I see you are still bitter over whatever bit your bum back then and still live it up to this day.
Software hasn't been the problem of FA for a very, very long time. The code and part of the the database has been optimized up to the point where there are no optimizations left to throw at it - all done with only one purpose - to make the website work on the hardware that is giving us the performance problems.
The server software has been finetuned to handle lots of connections effortlessly.
However, if you would have done a little bit of research you would have known by now that FA currently faces one hardware problem alone - the nvidia based gigabit network on the motherboard. It has been spoken off many times.
FreeBSD 6.x does not the stable drivers for that chipset( and neither does 7.x), so the network traffic takes a high toll of the total performance of the machine as packet processing is being done on the CPU as it seems. As a rough estimate, the CPU on the server makes ~40 thousand context switches a second, when as a normal working server does around 5-6.
This degrades total server performance significantly, up to the point where there there is simply no horsepower enough to handle the load, no matter how optimized your code is.
These network drivers are also unstable and sometimes cause kernel panics.
Add to that the problems with colocation facilities, and you have an explanation why FA is slow and goes offline at times.
And we are not throwing hardware at the ineffective code to take care of the performance problem.
That's funny, because I remember a lot more than that going wrong, particularly with ram problems and incompatible parts ordered numerous times. You got upset behind the scenes several times about those things, and rightfully so.
It's really amazing what a single pair of servers can do to power a state-wide university and how reliable they can be when they've got well-coded software on them, even when there's millions of daily visitors and hundreds of thousands of users with accounts. You guys really need to stop endlessly neglecting software.
EDIT: It also seems you have a problem of letting go of the past. What happened was years ago. It's time to move on.
I am sorry, but from the looks of things, you are the one who can not let go of the past, as you are the one who is digging in the archives of history.
Nobody stood still during all this time, and we have moved forward a long way. It it unfortunate that you do not see it.
What's the new server likely to be?
Two quad core Xeon L5420 2.5Ghz 50W processors on an Asus Z7S mobo would be the start of a good server.
Yes, something like that if we can afford it.
But as I have said in my previous reply to post on this thread, the final solution really depends on the amount of money we can raise. Given enough, we would probably end up getting two servers, one for www and other software needs, and another for a file server.
The former would boast something like 2xQuad core Xeons with 2GB of RAM, and RAID1 of 80Gig SATA drives, and the latter would probably have a dual core CPU, 16Gig of RAM for filesystem cache, and a RAID5 of 1TB drives for storage. This is the ideal world solution right now, unfortunately, it is too costly for us at the moment.
Time will tell however.
$70 going in from me. 
