6. org.hornetq.core.journal.impl.TimedBuffer is using ~500k of memory (if that can be optimized)
If we optimize it, we will be missing the optimizations we need for performance.
Although we could have the buffer empty (not allocated) until the first usage hits. But that won't prevent the memory usage if persistent messages are being used.
Lazy is better -- there is always a warm-up period for benchmarks anyway. Of course if the entire HQ module in AS 7 could be done lazy it would be best.
But small steps