- This topic is empty.
-
Posts
-
That is a very interesting read.
The one benefit to this is that if you use the same hardware on all systems in the cluster, then you could very easily take an image of one machine and use that image to dump down on the others. This makes building and maintaining the farm a lot easier than building each system individually. Not only that, but you get consistency in the build, even if its wrong. If something does go wrong, you're pretty much guaranteed to have it go wrong on all the boxes, not just one.
With the amount of data transfer they're talking about here, I definitely agree with the gigabit minimum. If you can afford it, gigabit fiber is actually a better way to go, as fiber is obviously faster than ethernet/copper.
Hot swappable RAID configurations are always good when you're talking about having a system that needs to be operational almost 100% of the time, especially during a data crunch period. Assuming one of the disks does go bad, you can remove the one drive, put in the new one, and the OS will probably automatically begin stripping the drive. The only downside to this is that you waste CPU cycles working on stripping the data. But, if I had to choose between CPU cycles and system downtime, I think I'd rather use the extra CPU cycles.
Oh and by the way, did I mention…*drool*
Felt like looking at some 3D related stuff today, as it has been a long time. Well, I was browsing around HighEnd3d.com (a very good resource for 3D news, articles, and tutorials), and came across this article on the principles of planning and setting up a Renderfarm.
http://www.highend3d.com/artic…..ms-27.html
I've only read the first page so far (class just started), but I'm finding it a very good read. Makes me almost wish I had to money just to design a system like that – for rendering, number crunching, data handling, or whatever. I would love to be able to run something like that.
Anyway, just an interesting read, so I thought I'd share it.
- You must be logged in to reply to this topic.