Linux has still a long way to go

Linux has still a long way to go

World Domination for Linux in clusters? Where? OK, some universities (cash-stripped as usual) deploy them to give more students and researchers the opportunity to play with them. Great for them, but they are most often not the serious users. Some companies look seriously for Linux clusters since they see the possibility for cost savings for low- to mid-range demands. (E.g., I'm currently involved in benchmarking Linux clusters for crash simulation.)

But, from my hands-on experience with Linux clusters, Sun Cluster (both 2.x and SC3), IBM'S HACMP and POE, Veritas clusters, and (last, but not least) VMS clusters -- believe me: Linux has still quite some way to go until it comes in the region where the major player work today. That holds for all RAS areas (reliability, availability, and serviceability), and for performance.

Linux clusters make headway, and that's a good thing; but they are focused on MPP-style problems which we don't always have and which are hard to program for the application developpers.

In the end, it boils down to simple decisions: If you need your LS-DYNA job (that's crash simulation) or your NASTRAN computation finished the next morning, since it saves hundreds of expensive engineering hours, then you'll go with an IBM Regatta and not with a Linux Cluster. If you have a bit more time, you save the money in hardware and fight with the Linux cluster software.