Seven Barriers to Server Consolidation
We set out to find what is holding back server consolidation, in the hope that identifying the problems will spur a stronger drive to overcome them -- and produce a leaner, meaner data center in the end.
Whenever the conversation in IT circles drifts toward low-power servers and the new class of chips that run them, I'm reminded of an early scene in "Joe Versus the Volcano."
Joe (Tom Hanks) walks into his dead-end job where Mr. Waturi (the brilliant Dan Hedaya) is having an endless telephone argument with another middle manager: "I know he can get the job, but can he do the job? (Pause) I know he can get the job ..." In the end, we never find out if "he" actually gets the job, let alone succeeds or fails, but the point is that low-power servers present the same dilemma to IT management.
No doubt, the technology is sound. With today's multicore designs, servers can pack more low-power cores into a single chassis, both increasing performance and lowering consumption compared to current designs. So, yes, they can get the job. The question for many is whether these low-power designs truly provide the most bang for the buck -- performance/watt in industry parlance. Can they do the job?
Take Dell's new PowerEdge C5000 micro servers, for example. These are the company's third-generation, low-power machines sporting AMD's Phenom II and Athlon II chips, along with Intel's Xeon E3-1200 processors. While the initial C5000s were targeted at such high-density, low-cost applications like cloud services and Web hosting, these latest machines are taking a bead on more mainstream enterprise applications, handling complex front-end workloads and even virtual environments.
The problem with this, says IT guru David Chernicoff, is that once you expand low-power's traditional role as a purpose-built platform for singular functions into a general service processing engine, the performance end of the equation begins to suffer dramatically. If you are planning to increase the virtual footprint in your data center (and who isn't?), wouldn't it make more sense to deploy fewer, more powerful servers than legions of micro systems? Ask any CIO whether it makes sense to scale out virtual infrastructure with more blades or scale up (consolidate) onto bigger platforms and you'll get your answer.
As I mentioned, though, it's not like low-power systems have no use whatsoever. As we continue to see a steady stream of releases from SeaMicro, Calexda and others, it is clear that demand for Atom-, ARM- and Phenom-based machines will only increase among cloud providers and other Web-based users. As PC Magazine's Michael Miller points out, highly parallel applications like Memcached and Hadoop are tailor-made for multinode platforms. But CAD/CAM, database or other apps that require out-of-order processing and branch prediction? Not so much.
So if our friend Mr. Waturi had been an IT executive discussing the merits of low-power server technology, the conversation might have been mercifully short. Yes, they can get the job. Yes, they can do the job, provided it's the right job in the right environment. Still, are they a worthwhile investment?
I'm not arguing that with you.
I'm not arguing that with you.
I'm not arguing that with you!