No, we’re not talking about the aircraft carrier.

Instead, this is about the potential for Blue Gene/P [ domino.research.ibm.com ] to be used as a massive high performance web server. Okay, perhaps not just massive, but…. alright, my vocabulary probably doesn’t contain a term that can sufficiently describe it.

This was featured on slashdot a few days back and tianbo [ nordschleife.metaforix.net ] sent me an email reminder to take a look at it.

So here’s the paper: Project Kittyhawk [ domino.research.ibm.com ]

I think the most interesting part of this is that instead of just running the good ol’ scientific stuff like Computational Fluid Dynamics (CFD), this project looks into more “practical” or rather, “saleable” aspects of high performance computing in the form of web services or more generically, on-demand computing power.

However, such a system may present a chicken-and-egg problem in the real world. Pragmatically, interest in such a system revolves around the benefits of the potential economies of scale. But achieving such a scale would first require sufficient client demand.

Today, web companies who require this sort of infrastructure already have their own in-house systems that have (almost always) been grown from scratch. More likely than not, it will be the new startups who will benefit from Kittyhawk. Considering the fact that a commercial offering of Kittyhawk will require someone who can afford it, it’ll probably be offered by IBM themselves. So what will happen next? I wonder.

Related posts:

1. Community DNS

1. frankchn Says:
   February 10th, 2008 at 11:36 pm

   I am skeptical about the commercial viability of the project.

   The people who run the five nines or six nines availability and high load systems, as you said, have home-grown systems tailored to their specific needs. Even organizations with limited budgets (e.g. the Wikimedia Fdn) can use open source technologies and commodity hardware to set-up and run what is arguably a very dynamic site, so I am not really sure who the Kittyhawk project is targeted at.

2. bangky Says:
   February 11th, 2008 at 7:30 pm

   Hi,

   True enough, you’ve got a valid point there about how (relatively) small organizations can actually rely on low-cost methods to get their systems up and running.

   However, do note that commodity hardware may have certain limitations. Apart from the potential cooling and power issues, blade servers tend to squeeze more performance into a denser form-factor that usually comes with custom tools to facilitate more effective infrastructure management. In the long term, it probably provides a cost-performance ratio that commodity hardware is unlikely to match.

   But the (literally) million dollar question is – what about Kittyhawk? I really can’t say for sure but possibly it could be something along the lines of what Amazon’s Elastic Compute Cloud (EC2) is providing right now. Blue Gene probably has more than enough capacity to host an entire virtual computing environment for a multitude of users.