If Google can keep things in Beta, why can't we? So.. beware... Things might break. Please join the mailing list and report any glitches that you come across.
It was recently decided that having a split personality is not the best thing to have in a cluster as two different resource managers start to compete while not being aware of each other. That is why we have separated our cluster into a Compute cluster and a Hadoop cluster. Read on for more about the two.
The compute cluster comes in handy whenever you need a lot of cores to get your job done. It is just like looking at the CPU and memory load of the other servers and then deciding which one to use for your job, only the job schedule will take care of looking at the CPU load for you and schedule the resources on a first come first serve basis (at least for the time being, queue priorities may change in the future).
- 1 head node: iln1
- 2 development nodes: ild1, ild2
- 28 compute nodes:
- 896 CPU cores
- 1792 GB RAM
- Torque resource manager
- MAUI job scheduler
- CentOS 6.3
- 2 head nodes
- 2 development nodes
- 36 compute nodes:
- 1152 cores
- 2.25 TB RAM
- Each node:
- 2x AMD Opteron 6276 (Interlagos) @2.3GHz - 3.2GHz, 16 cores/CPU, AMD64, VT
- 64 GB RAM
- 2 TB local HDD
There is a mailing list for all those interested in what is currently happening with the cluster and the configuration of the cluster:
List address: firstname.lastname@example.org
Manage your subscription: https://mailman.stanford.edu/mailman/listinfo/ilcluster
We decided that it is not good to have a split personality that is why we now have a set of nodes dedicated to the compute cluster and another set of nodes dedicated to a hadoop cluster.
- TORQUE resource manager, MAUI job scheduler
- Nodes: iln01-iln28
- Submission node: ilhead1
- Apache Hadoop
- Nodes: iln29-iln36
- Submission node: iln29