Subject: RE: zbufs for NetBSD
To: None <firstname.lastname@example.org>
From: None <kyle.unice@L-3com.com>
Date: 09/06/2002 10:53:10
Just wanted to follow up and see what decision (if any) has been made about
zbufs and NetBSD.
From: Jason R Thorpe [mailto:email@example.com]
Sent: Friday, August 23, 2002 10:54 AM
Cc: firstname.lastname@example.org; email@example.com
Subject: Re: zbufs for NetBSD
On Fri, Aug 23, 2002 at 10:10:00AM -0600, kyle.unice@L-3com.com wrote:
> would be an inconsequential performance improvment, but for streaming
> on a large packet data link, zbufs could be quite an improvment. I know
> there are ALOT of studies and quantifications that could be done on which
> apps would benefit most from zbufs and does NetBSD run those apps, but
> zbufs would be a plus for NetBSD.
I think you're missing my point, though.
In NetBSD, right now, normal write(2) calls to sockets are *already*
zero-copy, all the way from user application to the wire. It is the
default in NetBSD-current, and an option in the upcoming 1.6 release.
The only thing that is missing is a callback to userspace when the NIC
is finished with the buffer. This could be achieved with a normal POSIX
Async I/O implementation.
As I said, this does not address read(2) on a socket. But again, page
flipping techniques could be used here.
What I'm trying to say is that it's not really clear that a different
API like zbufs are really necessary in order to achieve full zero-copy
Now, all that said, can you provide a reference on the zbuf API?
-- Jason R. Thorpe <firstname.lastname@example.org>