It seems like it has been ages since my last blog entry. Little things like a new project, a vacation, some holidays all seemed to pull me into new and interesting things, most not involving a computer. And while I thought about blogging from the beach in Maui, there always seemed to be a little umbrella drink between me and my Blackberry (although now I really understand why many refer to them as "crackberries"). But, at last the new year rolls around and I'm working myself back into my normal routine. Slowly. With help from Starbucks. Luckily, a wise few chose to extend their holidays through the end of the week so I could return to my labors of love slowly, gently, and with a little time to blog.
In our last entry, I was sitting alternately in meetings about Green Computing, about large, google-sized farms of computers, about complexities of management, about strategies for simplifying management of large farms, about costs of machine configuration, about the human cost for managing complex environments, about the costs for securing, installing, updating, configuring, reconfiguring, administering, relocating, cooling, powering, virtualizing, migrating, reconfiguring again, replacing, modernizing, consolidating, and, oh, then my head exploded and I wrote a blog entry.
Now, all these things don't seem to be isolated events, but most of these problems seem to be showing up at nearly every major data center. A few data centers managers are very proud that they have installed computers 5 times faster than their old ones, and thus reduced costs, or they have installed the latest high powered cooling systems, or installed the largest data center powered directly from the Hoover Dam and thus, they project, they have managed to avert another data center crisis for a year or two.
Of course, most improvements have come at the cost of the environment, or perhaps the data centers fix (which doubled the power consumption for the company) doesn't show that they increased their annual power bill by more than the cost of all of the upgrades. The most common solution seems to be to add more of your favorite dual processor Intel or AMD based system, some with a whopping two or four processors, each supported by several fans, various power supplies, several co-processors, a variety of PCI cards, oh, and gigabytes and gigabytes of memory per machine (memory gobbles up what, now, up to about 20% of your power in some configurations?) Yet the workloads continue to double or triple in size every year. That is nearly an order of magnitude in compute power every three to four years. On average, that increase in workload seems to exceed Moore's Law, meaning that simple upgrades of processors and pure technology replacement is never going to keep pace with the growing rate of compute/storage/network consumption.
Vendors are working hard to address parts of the problem, as I alluded to in the last article. Yes, processors are getting faster and memories are getting more dense. Cooling doors for cabinets are taking out some heat. Virtualization is starting to improve efficiency, but at the cost of increased complexity for systems administrators. All of these improvements combined do not seem to be keeping up with the explosive rate of growth in terms of consumption of compute cycles.
I think we are due (overdue, in fact) for a paradigm shift.
I believe we have many technologies at hand which have been poised for a while to provide some relief from this fast growing cycle. There is even a term for this paradigm shift that several large companies are starting to use. While the meaning of the term is still fairly open and ambiguous, I believe it pulls together some basic principles hold some promise for alleviating many of today's pain points in computing. Specifically, this paradigm shift is being called "Cloud Computing" and provides a framework which, over time, can help alleviate not only the rapid growth of workloads, but also be a focal point for energy management, for abstraction of workloads from physical hardware, and ultimately, for the overall simplification of IT.
Labels: cloud-computing cloudcomputing