In another blog entry, I interviewed Bloglines founder Mark Fletcher. As a result of AskJeeve's acquisition of Bloglines, Fletcher is now the general manager/vice president of "Bloglines at AskJeeves" (a mouthful that needs to be fixed). Given that Bloglines is basically a software utility, I asked Fletcher the same question that I've started to ask the providers of other software utilities (e.g.: salesforce.com). The idea is to do an informal survey that reality checks the idea of an on-demand hardware utility with an audience already predisposed to the idea of utility computing. I asked him whether, in addressing the scaling challenges that a software utility like Bloglines must address, the idea of a $1/CPU/hour compute utility (along the lines that Sun is going) is interesting to him at all in terms of a way to cost-efficiently scale his business.
Said Fletcher, "I've seen Sun's ad. They run a billboard on 101 (in the Bay Area). We take a different philosophy. We look for the cheapest Intel hardware we can find and run Linux on it. In building this, we assume that we're going to have failures and we design the 'system' to tolerate them. It's much the same philosophy that Google takes. You get so much bang for the buck that it's a no-brainer. Also, they're focusing on a different market. We're not CPU bound. We're storage bound."
As it turns out, Sun is making storage available on a $1/gigabyte/month utility basis as well. (Perhaps an awareness problem for Sun?) But its storage utility is probably best suited to customers of the current CPU utility initiative and that's largely restricted to high performance computing (HPC) applications (that are well suited to grids). This application restriction, as best as I can tell from Sun president/COO Jonathan Schwartz's various pitches for utility computing, is supposed to be temporary. Which is why, in a recent column, I took a more strategic look at the traction that Sun's $1/CPU/hour message is getting.
Tactically speaking, the HPC market, where specialized CPU-bound applications like Monte Carlo simulations can be spread across a grid of low-cost systems,