🚀 go-pugleaf

I'm one who likes to blast people for not seeing past the end of their nose,
but i do think that 64-bit addressing will be stable for a fairly long time.

To re-iterate what John Mashey said, people may think that address sizes must
double every so often, but if you think about it, that doesn't work out right.
That would assume a doubly-exponential growth in memory, which is doubtful.

For physical addresses, i think we can safely assume that they grow linearly,
adding about 1 bit per year.  You can argue about the 18-month doubling time;
maybe it will change some, but i doubt that it will become disastrously wrong.
Old technologies may level off, new ones will appear, and become more mature,
and they will have various tradeoffs of size, speed, and cost.  Nothing new.

For virtual addresses, *if* we assume that they're used the traditional way,
basically as an extension of physical memory, then the same argument applies.
Virtual data sizes should be a small factor more than physical memory sizes,
and they're also limited by the amount of disk space available to the machine.
So you'll run into problems a few years before those for physical addresses.

On the other hand, when you give people more space, they will want to do more
ridiculous things, like map the whole world into their virtual address space.
I've heard many people say, essentially, that 64-bit addresses are limitless,
so we don't need to worry about memory allocation; we can just use up space.
Of course this is wrong, but you can be sure that people will be less careful.

I agree that 64-bit space is, currently, plenty of space *for one machine*,
but once you talk about distributed systems, that becomes not nearly enough.
I like to have persistent identifiers that are unique across space and time,
but i don't think that it should be virtual addresses, and for this purpose,
even 128-bit space is cutting things close, planning not too far in advance.

--
Joe Keane, amateur architect