I R a Lynnucks guy
Oct. 12th, 2005 02:32 pm![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
revchrisYesterday afternoon and this morning were spent in the incredibly noisy, headache-generating server room building a linux computational cluster.
Why? Because we could.
One of my profs is a computational modeler and is teaching a class on the subject. She asked me if I could come up with a few machines to build a cluster on so that she could make her class write software to run on it. At first glance, this should have been easy, as I figured we could just drop ROCKS on them and be done with it. ROCKS, unfortunately, requires 16GB drives and I have 10GBs. The Powers That Be don't want to hear about it, because it's running RedHat9, which they don't like because it's unsupported. The whole cluster, however, can be installed with directions that fit on 34 PowerPoint slides (which were emailed to me from our spies at USC), and a little fiddling to get the head node to talk on the internal network to the cluster and on the external network to be able to download the Fortran compiler and MPICH. THen I turned everything over to one of the Prof's grad students to let them actually do the software and user account setups.
Realistically, unless you have no budget (like we did for this project) and really want a cluster, you could get faster processing out of a single p4-3.4 machine than you'll get out of our cluster (4x p3-800/512MB/10GB), but this way, it only cost me about a day of work, and I learned a little more about Linux in the process.
Now we're waiting for the machines for the big cluster to show up (100x dual-Opteron246/40GB/2GB), and hoping the remodeling for that lab will be done in time. The lab is designed to eventually support 200 machines, and has power and cooling designed around a theoretical high-end of 400W per machine. Monday one of the engineers on that project called me to find out why we weren't heating that room (it has two 40kW 8500CFM air handlers for cooling, but no outside air or heat). It's going to be a very windy lab - 650 ft⊃ with 8-foot ceilings at 17,000cfm means an air change every 18 seconds.
Why? Because we could.
One of my profs is a computational modeler and is teaching a class on the subject. She asked me if I could come up with a few machines to build a cluster on so that she could make her class write software to run on it. At first glance, this should have been easy, as I figured we could just drop ROCKS on them and be done with it. ROCKS, unfortunately, requires 16GB drives and I have 10GBs. The Powers That Be don't want to hear about it, because it's running RedHat9, which they don't like because it's unsupported. The whole cluster, however, can be installed with directions that fit on 34 PowerPoint slides (which were emailed to me from our spies at USC), and a little fiddling to get the head node to talk on the internal network to the cluster and on the external network to be able to download the Fortran compiler and MPICH. THen I turned everything over to one of the Prof's grad students to let them actually do the software and user account setups.
Realistically, unless you have no budget (like we did for this project) and really want a cluster, you could get faster processing out of a single p4-3.4 machine than you'll get out of our cluster (4x p3-800/512MB/10GB), but this way, it only cost me about a day of work, and I learned a little more about Linux in the process.
Now we're waiting for the machines for the big cluster to show up (100x dual-Opteron246/40GB/2GB), and hoping the remodeling for that lab will be done in time. The lab is designed to eventually support 200 machines, and has power and cooling designed around a theoretical high-end of 400W per machine. Monday one of the engineers on that project called me to find out why we weren't heating that room (it has two 40kW 8500CFM air handlers for cooling, but no outside air or heat). It's going to be a very windy lab - 650 ft⊃ with 8-foot ceilings at 17,000cfm means an air change every 18 seconds.
