Tuesday, January 04, 2005

fyi Linux continues inroads into server market


Pointer to article:

Kobielus kommentary:
My head did a quick compute on these statements from the article:

• "The growth of blade servers is also an important contributing factor [in Linux’s adoption]."
• “[H]alf of all blade servers shipped today run Linux.”
• “Initial users of blade servers were those who embrace new technology first and tended to be in the high-performance computing arena, which is a stronghold of Linux.”
• “The increased scalability of the Linux kernel Version 2.6 has resulted in more multiprocessor Linux server shipments, the report said, and dual-processor systems are now the predominant configuration, followed by uniprocessors and four-processor systems.”

It’s clear to me, from my research, that blade servers are becoming the primary proving ground for grid-computing architectures in enterprises. Forget these globe-spanning grids that search for signs of intelligent life in the universe and so forth—“they’re out there” (OK, dumb pun intended), but they’re research tools rather than corporate infrastructure. Likewise, don’t think that data-center grids will obsolete traditional server clusters any time soon: though, clearly, server clusters are becoming more dynamic in their real-time task allocation among available CPUs, storage resources, and so forth.

No, focus instead on grids that are entirely contained within bladed computing chassis, yoking the many blades (each its own “stand-alone” server) into virtual supercomputer. The popular conception of grids involves massive “resource aggregation” among many physically separate machines over the wide or local area. But the increasing reality of grid deployments will involve more flexible “resource partitioning”—both through “hard blades” (i.e., separate slip-in server boards) and “soft blades” (i.e., virtual machines that slice up resources on each hard blade into logically distinct computing nodes).

By concentrating computing resources into single chassis, resource-partitioning intra-chassis grids are inherently more manageable than the scattered, distributed, resource-aggregation grids. Expect more intra-chassis grids to take root in data centers everywhere. Expect most of them to incorporate hard-bladed and soft-bladed Linux environments (anywhere from dozens to hundreds to thousands of Linux instances per chassis). Expect them to usurp more enterprise high-performance computing applications. Expect those grids to continue scaling and screaming faster and faster, through “scale-in” (intra-chassis Linux partitioning) and “scale-out” (extra-chassis Linux aggregation).

It's undeniable. Blades are growing ever sharper.