Long-time Slashdot reader 00_NOP is a software engineer (with a PhD in real-time computing) re-visits a historic research paper on the financial trade-offs between disk space (then costing about $20,000 per kilobyte) and (volatile) memory (costing about $5 per kilobyte):
Thirty-five years ago that report for Tandem computers concluded that the cost balance between memory, disk and CPU on big iron favoured holding items in memory if they were needed every five minutes and using five bytes to save one instruction.
Well my estimate is that we should aim to hold items that we have to access 10 times a second.
And needless to say, some techniques for saving data space are more efficient than they were 35 years ago, their article points out.
“The cost of an instruction per second and the cost of a byte of memory are approximately equivalent — that’s tipped the balance somewhat towards data compression (eg., perhaps through using bit flags in a byte instead of a number of booleans for instance), though not by a huge amount.”