In theory, Java is great and Cassandra is great. In practice - Java under a heavy load is a disaster, because it was never designed for it, and Cassandra is a just a hype and propaganda.
Face the reality - it doesn't work in production as it supposed to - as a primary storage engine.
People at the Digg aren't amateur idiots, so I think they do everything as it described in docs, but the damn thing just doesn't work.
Google's heavy use of serverside Java would indicate otherwise.
> because it was never designed for it
Yes it was. Java has a lot of problems, but one thing that isn't a problem is heavy load.
> Cassandra is a just a hype and propaganda
Facebook seems to be using it ok.
System which was designed for being isolated from an OS (leave alone hardware) will have a bottleneck exactly in this level.
Facebook doesn't use it as a primary storage with a high load.
Most languages used in web programming are divorced from the hardware, and either use a virtual machine (e.g. Java, .NET) or an interpreter (PHP, Python, Perl, Ruby).
Virtually no-one uses a low level language for web programming, and the benchmarks say that the virtual machines are faster than interpreters. Java is very fast and efficient once it's running, but initial start-up time is often slower than interpreted languages due to the JIT compilation, but servers rarely "start up". If you're really hitting a barrier with Java's performance, just as with other non-native languages, you can write the performance critical section in C.
http://shootout.alioth.debian.org/u32q/which-programming-lan...