James Hamilton had a blog post yesterday about "Embarrasingly Distributed Cloud Services". It points to a paper that he and some others at Microsoft wrote to present at the ACM Hotnets 2008 Workshop in Calgary. The paper is pretty amazing and besides some pretty deep formulas and theory for networks a number of good points are made on mega data centers versus newer, geo-diverse, distributed or containerized designs. It's a paper about trends in the data center industry and what models will work best for particular types of processing and applications. To me, it was incredibly insightful and you can tell that the authors have not only done their research, but have been seen what works and what doesn't, as well as have the vision for what the future holds. I (obviously) recommend reading the paper (here), but also have a few comments/observations of my own:
1. Products -- maybe I've had too much caffiene today, but I can see a mix of software products that are ideal for a predictive and scalable deployment across the new, geo-diverse containerized data center designs. 3Tera continues to be a hot company with their Applogic software and recent global cloud computing expansion. I thought of them several times when reading the paper. I also thought about another cool company (that 2 of my friends work at) called Cariden. I don't know if my description of what they do would do it justice -- let's just call it network simulation and traffic planning/engineering software.
2. I haven't done a plug for my almost year-old white paper in while, so what better opportunity. If these geo-diverse, smaller data centers are built in closer proximity to the end user, then site selection becomes an issue that many have not thought of before (because they never ventured outside of their region of the country). So -- check out my Data Center Site Selection white paper.
3. It might be a stretch to relate this one to the paper -- but it was something I ran across recently that could come into play with containers and an increased need for sensors and environment data. I was browsing around the Dust Networks website a few days back, via a link from the IP for Smart Objects website. Dust Networks makes embedded wireless sensor networking for monitoring and control. Co-founder Kris Pister helped coin the concept of "smart dust" to describe a network of miniture wireless sensors called "motes". The 12 millimeters square devices can run on micro-batteries for up to 10 years. Check out the product lines and technology at http://www.dustnetworks.com/
4. The day before at the Interop 2008 New York conference NeworkWorld reports that today's networks won't cut it as foundations for real-time applications.