Openomics: innovation drives value

In 2001, our ISV Engineering group at Sun --f.k.a. Market Development Engineering-- released the first version of the Hardware Activity Reporter (HAR) tool for Solaris 8 and up. Starting with Solaris 8, Sun had begun to deliver public interfaces for the SPARC and x86 hardware performance counters --libcpc, to access CPU counters and libpctx, to track a process--, leading to a new generation of performance monitoring tools: the Solaris cpustat and cputrack commands, hardware-counter profiling in Sun Studio's Performance Analyzer, SPOT and HAR, to name a few.

HAR differs from other tools in the fact that it combines the low-level counts into higher-level metrics more useful to application programmers. Application programmers are typically interested in the following metrics: CPI, FLOPS, MIPS, address bus percentage utilization, cache miss rates, branch and branch miss rates, and stall rates. These metrics help in assessing the fair usage of available processing units, locating bottlenecks and guiding tuning efforts, when needed. Check out this SDN article on how to use HAR to identify performance bottlenecks and quantify code-tuning improvements through a couple of case studies in the fields of CFD and MCAD.

Since the release of Solaris 10, HAR 1.x had become somewhat obsolete as we stopped adding support for new processors after UltraSPARC IV and Pentium P6. Today, thanks to the work of Amir Javanshir, we are happy to announce the release 2.0 of HAR, featuring…
[Read More]

Daimler AG, one of the world’s largest auto manufacturers and the maker of the luxury Mercedes cars, had been running CD-adapco's STAR-CD application for Computational Fluid Dynamics (CFD) to simulate large models of multi-million cells on a cluster of Sun Fire X4100 servers, for some time. Interestingly, Sun was chosen for its ability to innovate in the commodity computing (Lintel) space; while all vendors now feature the same chips from AMD and/or Intel, Sun differentiates itself through a well-thought and high-quality packaging that delivers higher RAS and lower power consumption --down to 56% reduction compared to competitive servers at the time of launch.

"The server products were particularly convincing, especially because of their well-designed heat management. Ultimately, our demands in this regard are very high because of the density of the computers in the server rooms. Favorable heat management provides a long hardware service life."
Dr. Volker Schwarz, Team Leader, Aerodynamics and Aeroacoustics Department, Daimler AG

The computation part of STAR-CD scaled very well across the 100's of processors in the Daimler cluster. Parallel computing makes it possible to routinely work on a 100M-cell mesh, thus allowing engineers to simulate entire structures. As a result, the start-up and closing phases of a simulation run, where models are decomposed into input files for each cluster node to read, could eventually dominate the simulation time, and Daimler experienced a bottleneck beyond 100 processors with NFS, which is not exactly suited for highly parallel I/O.

Enters Lustre. The Lustre file system employs object-based storage to scale to 10000's of nodes…

[Read More]