Re: HTTP-NG Server & Client Mechanisms

On Tue, 20 Feb 1996, Ashish Kolli wrote:
> 	As part of a course on World Wide Web Technologies, I am gathering
> a group of students in CMU to implement the HTTP-NG protocol and come up
> with a prototype client and server. The goal of this work is to measure
> the performance benefits of the HTTP-NG protocol and also to make prototypes
> available to the Web community so that people can use them to give further
> shape to the HTTP-NG protocol. The project is going to be executed in three
> phases:
> (1) Implementing communications based on the Session Control Protocol
> (2) Incorporating (1) into existing server/client code. (NSCA or WWWlib)
> (3) Adding some extensions like "load sharing" among servers.
> 	
> 	Before we start off the project I would like to know if you folks
> have any comments or issues that you think we should consider for this work.

Consider altering the SCP protocol just a touch.  As SCP stands right now it 
has a major flaw in that a single packet can be around four gigabites in 
length (the session length field is 32 bits).  Since the whole idea of 
SCP is to use small packets (breaking up large data across several 
packets) to give the capability of multiple 'streams' over a single 
connection, AND because I don't know of a single normal computer that could 
gracefully handle a single four gigabyte packet (considering that the 
computer could do nothing with the packet until it was fully received) I 
would strongly suggest reducing the data length to perhaps 16 bits.  When 
computers can handle 4gb packets I'm sure they will be using a protocol 
other than SCP.

-Brandon Gillespie-

Received on Wednesday, 21 February 1996 11:19:23 UTC