The Integration of Scalable Systems Software with the OSCAR Clustering Toolkit

The Scientific Discovery Through Advanced Computing (SciDAC) project has implemented a Scalable Systems Software (SSS) center. The SSS center is charged with building a scalable and robust operating environment for high end computing systems. Part of the approach is building a general solution for multiple platforms, to include both clusters and individual machines. This starts a trend toward standardization of systems within computing centers, which saves money in the long run by reducing operator training and lowering software development and support costs. The development in SciDAC: SSS is focused on components and not on the underlying operating system. The first target operating system for the individual components to run on is Linux on x86 based hardware. The vehicle chosen for delivery of the software is OSCAR, which supplies both the installation mechanism and the support for the underlying operating system. OSCAR, or Open Source Cluster Application Resources, is a clustering toolkit that has been widely used for installing/configuring smaller clusters of up to 256 nodes. The biggest advantage of OSCAR is that it works on top of specific Linux distributions. OSCAR makes an effort to package commonly available clustering software (an example is MPI/PVM) and provide sane default configurations for this software. This is done so that beginning system administrators can install clusters with minimal coaching and help. One of the biggest challenges for the SSS-OSCAR effort was converting the SSS software into OSCAR packages. An OSCAR package is a combination of RPMs and scripts, and there is a published API for doing this. Some history of both projects, a description of the OSCAR packaging API, and the new packages that are available to OSCAR users from the SSS project is available in this paper. Also included is a summation of the experience of many new package writers using the OSCAR API, and some insights into the strengths and weaknesses of choosing OSCAR for the deployment vehicle for the SSS suite.