I believe that these days many big enough latency optimized systems use a hierarchy of storage solutions (magnetic, flash, RAM). Here's a useful rule of thumb:

I believe that these days many big enough latency optimized systems use a hierarchy of storage solutions (magnetic, flash, RAM). Here's a useful rule of thumb: the costs per byte of RAM:flash:magnetic are approximately 100:10:1.