My main problem is - too little memory. My backtest cluster is calculating 1gb / core (we possibly could do with half) - that thing can address 8gb with 60 cores. That makes it very good for a lot of low data stuff (web server) but very bad when you have to keep some data in memory.