It seems that every few years, IT settles down somewhat. While processors get ever faster, storage gets ever bigger, and bandwidth grows exponentially, the workloads these technologies support seem to stabilize to a certain degree. Not long ago, Windows Active Directory domain controllers required their own dedicated server for even small shops, while now they're almost an afterthought in terms of performance. Where Microsoft Exchange used to be the 800-pound gorilla in most organizations, it can now generally be relegated to a VM, or at least a middle-tier server. Even big apps like SAP can suddenly seem small in the face of hugely powerful, modern multicore CPUs.
With the huge gains in CPU performance, storage densities, and network bandwidth seen over the past two years, traditional pain points simply aren't nearly as painful anymore. So what happens now?
[ Intel's new Westmere-EP Xeon is a worthy successor to the amazing Nehalem. See "InfoWorld review: Intel's Westmere struts its stuff." ]
If history is any guide, what happens now is a surge of new software. Unfortunately, much of that will be corner-cutting bloat, but at least some of it will be innovative and useful code that actually takes advantage of the hardware performance gains.
For instance, take the AES-NI instructions introduced in the new Intel Westmere-EP chips. In my performance testing, AES-NI produced a 400 percent performance boost in AES encryption tasks, reducing the time required to encrypt an 851MB file from 13.5 seconds to 3 seconds. That's simply huge. That means that whole-disk encryption tools leveraging the new instructions can realistically encrypt large volumes without a significant performance penalty. Encrypting swap can become commonplace. Conducting high-volume, highly secure digital transactions will no longer require dedicated crypto offloads or boatloads of CPU time. This just might be the key that locks all of our doors.
On the other hand, the ever-increasing core count of modern CPUs renders previously difficult software-only solutions quite attractive. Software RAID seemed like a significant performance hit not that long ago, requiring enough CPU cycles to threaten the core functions of the server. Now, it's not that big of a deal. Heck, why not dedicate a core just for that? You have at least three or five more.
