This is apropos a recent discussion on splice() and similar kernel
data-path tuning....

------- Start of forwarded message -------
From: fwmiller@cs.umd.edu (Frank W. Miller)
Newsgroups: comp.os.research
Subject: Re: Zero copy (network) IO question ...
Date: 28 May 1999 19:42:49 GMT
Organization: U of Maryland, College Park, MD 20742
Message-ID: <7imrjp$ru4@darkstar.ucsc.edu>
References: <7gkijb$gdn@darkstar.ucsc.edu> <7h7i9u$806@darkstar.ucsc.edu> <7icgie$mrc@darkstar.ucsc.edu> <7iejr1$5f6@darkstar.ucsc.edu>


In article <7iejr1$5f6@darkstar.ucsc.edu>,
Chris Mikkelson <cmikk@uswest.net> wrote:
>
>
>In the first article of this thread, it looks like the poster was
>also asking about zero-copy network *input*, wheras sendfile is
>network output-only.
>

I recently completed a Ph.D. dissertation on generalizing the concept
used in the sendfile command present in the aforementioned OSs. My work
was based on generalizing the work of Fall at UCSD to support zerocopy
transfers between any pair of I/O devices in the system.  The disseration
is available at:

http://www.cs.umd.edu/~fwmiller/doc/dissertation.ps
http://www.cs.umd.edu/~fwmiller/doc/dissertation/dissertation.html

The work was also presented at the 19th IEEE Real-Time Systems Symposium

http://www.cs.umd.edu/~fwmiller/doc/rtss.ps

In addition, the prototype OS, called Roadrunner, in which the I/O system 
that implements these ideas is currently undergoing a transition from
research system to commercial product.  More information on it is available
at www.cornfed.com/roadrunner.

Later,
FM

-- 
Frank W. Miller
Cornfed Systems Inc
www.cornfed.com


------- End of forwarded message -------