Tale of Two Server Rooms
Recently I've had a chance to go onsite at two different clients pursuing the extremes of scale up and scale out architecture. The first client visit was to a highly air conditioned office building basement containing several thousand server boxes. Racks were stretched far and wide and I needed map coordinates to get to the computer I was there to work on. It is one of four such facilities that this company has (one being just for disaster recovery). I sat there underground on a fine summmer day in a wool sweater wearing earplugs to cut the fan noise for seven hours. The box I was installing software on will be used for about two weeks this year to vett the software my team developed. After that, it will pretty much just sit there consuming power and generating heat until the next release. In all, it was a fine example of the Wintel scale-out philosophy taken to its logical extreme, and it seemed like there should be an easier way.
Client visit number two was to a university. It was a mentoring gig, and I had to go onsite to help with an XML parsing problem. We sat in the clients office, ssh'd into the debian box they were using for production and ran some perl scripts looking for the bug. After we found the bug, and I was getting ready to leave, the client mentioned that the debian box wasn't actually a box, it was a VM running on their IBM mainframe. He had watched the tech do a network install of debian - the OS was slurped down over the net and no CD swapping was required. It's not very often I'm running code on a mainframe, and I was floored by the ease and transparency of the process. The university runs all their ERP software on that one machine - supporting 250,000 part time students, 50,000 full time, and several office buildings full of administrators.
Windley has more on mainframes and centralization.
http://www.windley.com/2003/05/22.html#a634
8:51:15 AM 
|
|
