daveb@geaclib.UUCP (David Collier-Brown) (04/02/89)
Well, it looks like for instructions and data with the current technologies, alignment of data makes good sense and architectural support for misaligned accesses is not a good use for silicon. But what about in the communications realm? Current packet formats are rather heavily "packed", sufficiently so that one wants to treat them as collections of bits of various lengths (usually even, often powers of two, etc). Two possibilities arise: Communications will be pushing physical transmission-speed limits and central processors will continue to get faster, faster. With processor time available for unpacking, packets will be densely packed, or Processing speeds will peak out sooner than transmission speeds in the near term, and there will be a need to design communications packet formats so they **aren't** an expanding-opcode scheme. In this scenario I'm not suggesting any holes are likely to appear, but I do expect most of the fields at the beginning of the (outer!) packet will be of fixed size and unchanging interpretation. Would anyone with very-high-speed comms experience care to comment on the ratio of development speeds? And does special-purpose hardware make any sense in this context... --dave -- David Collier-Brown. | yunexus!lethe!dave Interleaf Canada Inc. | 1550 Enterprise Rd. | He's so smart he's dumb. Mississauga, Ontario | --Joyce C-B
craig@NNSC.NSF.NET (Craig Partridge) (04/19/89)
David:
This is a belated reply to your note.
I think that for the near future processor speeds and communications
speeds will roughly keep pace. Or better said, they won't be so out of
sync that we need to radically revise our current networking models
(both virtual circuit and datagram) even into the gigabit speed realm.
[Note that some folks are talking about supporting new communications
paradigms on very high speed networks -- those new paradigms may require
us to revise our networking models -- I'm simply saying that speed alone
won't].
On the narrower question of alignment, I think some alignment is
always a win. Note that I don't care if each field is aligned nicely
on a word boundary -- bit masks can grab out stuff from the middle.
What I do care about is not having fields straddle boundaries. Having
the high-order byte of a two-byte field in one word and the low-order byte
in another word is a nuisance...
Craig