The UCSC/HHMI Kilokluster project
|
|
|
The purpose of the UCSC/HHMI Kilokluster project is to provide high throughput
computational services for bioinformatics applications. With over one thousand
processors operating in parallel, the KiloKluster might be one of the most
powerful computational clusters used for bioinformatics in the world.
Because of the massive growth that UCSC has been experiencing, space has been
a primary issue in the planning of this project. Thanks to the space saving
designs of Rackable Systems, the
KiloKluster occupies only nine racks and is less than one hundred square feet
in size.
The KiloKluster consists of 504 compute nodes. Each compute node contains two
Pentium III processors running at 866Mhz, a gigabyte of memory, and a forty
gigabyte hard drive.
The compute nodes are divided into eight sub-clusters, each with its own
switching apparatus and server. The compute nodes are networked using fast
ethernet, and the sub-clusters are connected to each other and to a main
server through a gigabit ethernet switch.
Sun's open source Gridware software
is used to assign tasks to the compute nodes. We chose Gridware primarily
because of its open source (thanks Sun!), and its ease of use compared to other
open source schedulers.
Technical questions may be directed to
cluster-admin@cse.ucsc.edu.
| |
|
|