Where Next?

what do the trends tell us?

Are the trends sustainable? If so, where do they lead us?

First, let's look at the DRAM vs. disk trend. Unfortunately, I lied when I said the trend was for disk density to improve by a factor of only 2 every 3 years. Since the late 1980s, when mainframe disk technology was displaced as the dominant technology at the high end by mass-market disks, the disk trend has sped up to match the DRAM density trend -- a factor of 4 improvement every 3 years.

What does this tell us? A technology under pressure has to adapt or die. In the high end database market where cost is secondary to speed, there are already DRAM-based disk systems. But the disk manufacturers, at least for now, have conquered the trends for most of their market.

What of the DRAM vs. SRAM trend? That one seems set to continue for at least the next 5 years, while improvements in the speed of transistor-based technologies (such as SRAM) still remain possible. Already, the speed difference between SRAM and DRAM is a factor of more than 10 -- and that gap will widen considerably over the next 5 years.

Perhaps it makes sense to start thinking in terms of a main memory built of SRAM, while using a disk as a fast paging device. After all, if we go back to the origins of virtual memory, when the secondary storage was a fast drum, the speed gap between the paging device and main memory was in about the same ball park as the speed gap of the near future between DRAM and SRAM.

If progress in computer systems means we need to look to the 1940s for inspiration for new memory systems, how long will it before Charles Babbage is back in vogue?

Memories are Made of This 17:30 Thurs 12 September 1996 SHB5, Senate House, University of the Witwatersrand, Johannesburg, South Africa