The E-Mail Synopsys User's Group (ESNUG) is a no-holds-barred user forum in which EDA tools and vendors often are skewered. Recently, however, a networking tool called Load Sharing Facility (LSF) from Platform Computing (www.platform.com) has been getting a lot of attention-and some rave reviews. LSF, which enables load sharing by distributing jobs to available CPUs in heterogeneous networks, appears to be of far more interest to engineers than the perennial Windows-vs.-Unix debate.
LSF could, in fact, help make the platform issue a moot point because it automates load balancing for mixed networks of PCs and Unix workstations. By helping users set up "compute farms" using the most cost-effective hardware, it may create a nice opening for Linux.
"Overall, I don't think I can say enough good things about LSF-but don't tell them that; they'll just want to raise their prices," wrote one ESNUG correspondent.
"I have had very positive results with LSF and praise it very highly, particularly for ASIC regression testing," said another. This engineer noted, however, that LSF is an "extremely configurable" tool that requires close cooperation between system administrators and users.
When one engineer reported he was disappointed with LSF because of problems with job submission, several others suggested the software was probably configured incorrectly.
ESNUG moderator John Cooley recently devoted an entire posting to a white paper from Blackstone EDA, the Boston distributor for Platform Computing. "LSF has become a very hot topic in the EDA world . . . a lot of chip designers I've known for years are independently singing the praises of LSF to me," Cooley wrote.
The white paper, "Optimizing Your Compute Farm with LSF," shows how LSF lets users set up a "network-centric" environment in which all CPU resources are controlled by one batch scheduler. Users submit jobs to the system and LSF identifies the available hosts and automatically dispatches jobs to them.
What can we learn from this interest in LSF?
The big concern among chip designers is not so much silicon issues like electromigration, or even the availability of intellectual property. It's "how can I get enough compute power to design chips that are so darn big?" Anything that helps designers make better use of existing or future CPU resources will be a winner and may be more appealing than trying to shoehorn chip designers into a new OS they may not want.