- From: Daniel W. Connolly <connolly@hal.com>
- Date: Thu, 15 Dec 1994 18:18:02 -0600
- To: Marc Salomon <marc@library.ucsf.edu>, http-wg%cuckoo.hpl.hp.com@hplb.hpl.hp.com
In message <ab1685b2030210040a79@[192.190.111.98]>, Mitra writes: >One thing to consider (not that i think this is a bad idea) is that often >the objects being sent along are images leading to two non-optimisations. > >1) Mime encoding is going to roughly double the number of bytes sent Hello? Mime encoding adds a few bytes between objects for the boundary. HTTP is 8-bit clean, after all. No base64 needed. >2) By explicitly sending them every time, the server wont allow the client >to cache the images, so for example the stupid blue ball for a "bullet" >will get send a zillion times. This is a very good point. The other point is that it is sufficiently expensive to deploy MGET or multipart/* over HTTP at this point that we might as well bite the bullet and do something long-term like Session Control Protocol (which meshes nicely with some security protocols like Secure Sockets Layer). Deploying MGET/multipart looks to me like: * Somebody hacks support into NCSA HTTPD (last I heard, that source code is not to enhancement-friendly. No offense intended.) * Somebody figures out how to manage these multipart files: do you cache them? create them with a cron job? * Bill Perry shows us how it's done with the emacs w3 browser. * Henric adds support in libwww 3.x (which is perhaps sufficiently different from 2.x that the lynx/chimera guys can't upgrade without a lot of pain and hassle?) * Somebody hacks support into NCSA Mosaic. (libwww in NCSA Mosaic bears little if any resemblance to CERN's libwww by now.) * A few information providers maybe start using it (It's 3 months into the future by now) * Other browser/server impelementors get pressured into adding support. Meanwhile, commercial folks are implementing HTTP-NG at lightning speed. Six months from now, all the major vendors are doing interoperable compression and encryption over something like SCP or SSL (not to mention strong authentication). In short: 'taint worthit. I'd rather see more effort spent on (1) Specifying existing practice on HTTP/1.0, keeping an eye on opportunities to gateway/cache/proxy out of HTTP and into next-generation protocols. (2) Carefully crafting HTTP-NG using an interface definition language like ASN/1 or OMG's IDL, so that it can be used with various transports and protocols. (ILU! ILU! ILU! ahem...) (3) Investigating some name service and replication strategies so that links that we write today will continue to work tomorrow. (4) Coordinating implementation efforts, and working on a common base of reusable code. Dan
Received on Thursday, 15 December 1994 16:25:17 UTC