From mboxrd@z Thu Jan 1 00:00:00 1970 To: 9fans@cse.psu.edu From: "Douglas A. Gwyn" Message-ID: <3F097FFF.67E4E455@null.net> Content-Type: text/plain; charset=iso-8859-1 References: <3F05D809.7030707@null.net>, <7150078ec6d4a983e1bc125e877704f7@collyer.net> Subject: Re: [9fans] Rstat needs three size fields? Date: Mon, 7 Jul 2003 15:10:48 +0000 Content-Transfer-Encoding: quoted-printable Topicbox-Message-UUID: eb94a286-eacb-11e9-9e20-41e7f4b1d025 Geoff Collyer wrote: > Actually 2=E2?=B3=E2?=B2 is a fairly big number in the context of 9P re= ad & > write byte counts (which is what Rob was talking about). I don't > think I've ever read or written 4,294,967,296 bytes (or even close to > that many) in a single operation. It's not what you have done in the past, but what somebody might reasonably want to do in the foreseeable future. 4GB is no longer unimaginably large, even for main RAM size, nor does it take a ridiculous amount of time to transfer it among hosts in modern networks. In my opinion we ought to be able to access the entirety of any given object without having to artificially segment it. > As for efficient integer encodings, at least for this purpose, if we > restricted I/O transfer lengths to powers of 2, we could just supply > the exponent in 9P ... The trouble is that even if the user accepted the power-of-two restriction, this would waste about 50% of the channel bandwidth. What I had in mind was something like Ken Thompson once devised, a variable-length integer format. Without built-in language support, though, this tends to bog down computation. It seems there should be some nice balance between computability and flexibility..