One of the presentations I saw at the recent London Cloud Camp (and again at Â Unvirtual) was Justin McCormack (@justinmccormack)’s lightning talk on “re-architecting for the ‘green’ cloud and lower costs” (is re-architecting a word? I don’t think so but re-designing doesn’t mean the same in this context!).
Justin has published his slidesÂ but he’s looking at ways to increase the scalability of our existing cloud applications. One idea is to build out parallel computing systems with many power-efficient CPUs (e.g. ARM chips) but Amdahl’s lawÂ kicks in so there is no real performance boost by building out – in fact, the line is almost linear so there is no compelling argument.
Instead, Justin argues that we currently write cloud applications that use a lot of memory (Facebook is understood to have around 200TB of memory cache). That’s because memory is fast and disk is slow.Â But with the advent of solid state devices we have something in between (that’s also low-power).
Instead of writing apps to live in huge RAM caches, we can use less memory, and more flash drives. The model is Â not going to be suitable for allÂ applicationsÂ but it’s fine for “quite big data” – i.e. normal, medium latency applications. A low-power cloud is potentially a low-cost middle ground with huge cost saving potential, if we can write cloud applications accordingly.
Justin plans to write more on the subject soon – keep up with his thoughts on the Â Technology of Content blog.