Van Jacobson pbuf's

[revanthn]

Hey guys

Recently I was going through a paper by Van Jacobson "Changes to 4.4 BSD Kernel Architecutre" and some more searching left me with the following results

Van Jacboson prototyped most of the code in 4.4BSD-Lite but most of the code including pbufs implementation are left behind without actually integrating them in the netowrk code and they continued to use the old mbufs style memory managemnt.So did Van released any of his prototype code under the public domain.What changes are integrated
into the 4.4BSD Lite release, that are devised by Van Jacobson and his team.Although his paper is very good I didn't get clear with some of the concepts like are pbufs fixed size where the size is decided by the interface that the packet spents it's life and how can the upper layers call the interface driver for a pbuf to stuff data into that
must be transmitted and such.I can take any info, URL,s personal emails whatever regarding what actually happened to the prototype code and also about the various Implementation details.


Thanx
prizark

 

[bcastner]

I have never read the paper, but I honestly think the answer to your question lies in the protocol used. If you are sending IP packets you cannot stuff or play with them if you expect other devices to understand them. You can Google a lot of information about the various differences in packets, but fundamentally no matter what OS you use or how you do socket services the package has to be framed appropriately, and there are a lot of options.

And there are a lot of options for you as a programmer in how you create and resolve packets as a service.

Again, I never read the paper, but I suspect Van Jacobsen left it to the reader to know that fundamentally no matter how you implement TCP/IP services, you are bound to the protocol for the end result.

 

[revanthn]

That true and nice but what you are talking about is the "Implementation of a TCP/IP stack" and what I am talking about is an "Optimised version of the same implementation" and that's where Van Jacobson's research comes in. He says Memory-to-Memory Copies and Interruptes are too expensive than the actual protocol code.

Google for "Design Changes to 4.4 BSD Kernel Network Architecutre" for his paper.It's a good one and you must take a look. And by the way do you have any experience in implementing protocol stacks or networking programming at the low level.Any help would be greatly appreciated.

Thanx
revanthn