What are the advantages of moving to a Linux HPC cluster?
We're trying to consolidate servers. During the dot-com boom, every department in my company bought its own servers, so we have departments running Solaris/SPARC, Mac, WinTel, HP-UX and more. We've got 1,800 in-house users and about 5,000 users who access many of our servers and applications over the Web. We want to bring as many servers as possible into our main datacenter. What are the advantages of moving to a Linux HPC cluster (probably with Dell servers) instead of going with a mainframe? Are there other architectures that we should look at?
Just about every new project today revolves around server consolidation, so I'm not surprised to see question like this. Obviously there are many ways to consolidate servers today. You mention Linux clustering and possibly going with a mainframe. I'm not sure how much you would really consolidate by going with an Intel platform. More than likely, you'd be throwing out 300 Unix servers for 600 PC Servers. If Linux is that important in your plans, then Linux on an IBM zSeries (OS/390) might be the solution for you. Here, you can get the benefit of some of the best of both worlds. Being more of a Unix person myself, I would not advocate going mainframe unless you have the people already in place to support such an environment.
What else can you do? Why not look at the IBM Server consolidation solution, Regatta. It looks and feels like a mainframe, but it is all Unix (AIX) and/or Linux (SuSE). Using their logical partition strategy, you can put up to 32 partitions in one Regatta that could replace over 100 stand-alone servers.
For details on the pSeries servers at IBM, see this URL:
This was first published in September 2003