At 18:49 7/11/2005, Brian Weeden, wrote: >I need a sanity check on something that I have been asked to comment >on at work involving upgrading one of our computer resources. The >problem is I haven't done any hands-on distributing computing or >linux/unix installs myself. > >Right now we have a system with 4 multi-CPU SGI servers (12 CPUs per) >with a couple terabytes of RAID storage. That forms the network. >There is also a SGI Oxygen server sitting at each workstation to do >graphics and a few minor applications. We mainly use the network to >hold a large amount of data stored in text file format. It is our >archive and historical reference. We have apps that are not very >graphically intensive but may want to add more in the future. We do >have some apps that are CPU intensive (strings of calculations) and >will be adding more. The real problem is we are tied to the Sun >software/hardware that cost us a ton of money in the first place and >we can't upgrade it without completely replacing it. We are also in >the process of completely changing our mission so we need to be >flexible for the future. > >What I want to propose is a non-proprietary solution that uses COTS >hardware/software, is expandable, and with good uptime. For the >hardware I was thinking of using rackmount blade servers. Then attach >a mess of RAID storage. If we need more power, we add more blades. If >we need more storage, we add more drives to the array. > >I figure we can load just about any flavor of Linux we need. Right >now we are using Irix (because everything is SGI) and I don't think it >would be that hard to recompile our software to work with Linux. What >I am envisioning is the OS sitting on top of all the blade server >hardware and we simply add the individual apps we need as we have the >need. > >At the workstations we can use basic Linux boxes. The workstations >don"t have to be super performers, just good enough to be able to >display around 100megs of data and plot it, maybe some simple 3D >functions. I want most of the calculations and data management do be >done on the server with viewing the data done on the workstations. > >Would you have to use a special distro/install of Unix/Linux to >utilize the blade servers? Would you have to recompile each time you >add another blade, or could it just automatically use and distribute >the resources you give the OS? > >Anything I am being really stupid about?
Here is some cluster/distributed computing info about FreeBSD: http://daleco.biz/cluster.txt http://dcwww.camp.dtu.dk/cluster-howto.html http://lists.freebsd.org/mailman/listinfo/freebsd-cluster http://freebsd.isc.org/ports/parallel.html http://www.Google.com/search?q=Computing+Clusters+Leo+Laporte+Roman+Loyola http://www.Google.com/search?q=Computing+Clusters+site%3Ag4tv.com Start Here to Find It Fast! -> http://www.US-Webmasters.com/best-start-page/ $8.77 Domain Names -> http://domains.us-webmasters.com/
