At 09:23 30.10.98 -0500, Jud McCranie wrote:
>At 09:15 AM 10/30/98 +0100, Harald Tveit Alvestrand wrote:
>
>>With architectures like that, you can burn off an awful number of
>>addressing bits in a hurry.
>
>Not that may, and you still don't need 256 bits for it. Let's say that there
>are 500,000,000 computers, each with 8 GB of disk space. This is somewhat
>more
>than we have now, but it won't be long before we do. 62 bits are
>sufficient to
>address each byte of disk space. 64 bits is enough to address each byte of 1
>billion 16GB disks.
You don't waste bits efficiently by packing them densely :-)
Let's instead imagine a "physical" address where the address is divided
into parts:
- Lower 36 bits address the bytes of a disk drive. This gives a maximum
of 281 terabytes per machine; we can leave the few larger installations
alone for another couple of years.
- Upper 32 bits are the IPv4 address of the machine on the Internet
- Due to the presence of NAT boxes, we need to have 16 bits to disambiguate
between boxes hiding behind the same firewall
- Due to the presence of non-IPv4 networks, we need an address family byte
to distinguish between different families of network identifiers
There! Wasted 92 bits, and didn't even have to think - next version will
introduce the IEEE 64-bit EUID that claims to identify everything, or
the Microsoft/DFS 128-bit UUID that claims the same thing......
it's far easier to waste bits by hierarchy than by adding more items
that are efficiently packed.
But this is growing VERY tangential to Mersenne; I'll shut up now....
Harald A
--
Harald Tveit Alvestrand, Maxware, Norway
[EMAIL PROTECTED]
