> >
> > Also Andrew Gallatin got this speed with Trapeze && GigE cards about a year
> > ago as well.
> >
> > It's not so much the 960Mbit (or better, actually, which you can get depending
> > on the card). It's how much CPU you eat up doing so. You of all people should
> > know that, Larry :-).
>
> Yeah, I do know it. I think FreeBSD reinventing what SGI did is stupid.
> As soon as you get rid of the copy as the issue the VM maps/unmaps become
> the issue. So they are inching closer to the next problem. Take a look at
> Stephen Tweedies kio bufs, that's better. You can read a little about them
> at ftp.bitmover.com://pub/splice.ps, that's where they came from.
Umm. Yes, that argument has merit. But what it does do, which is almost as
important as the performance issue itself, is to ready the networking stack
for these kind of callout types of tricks. Then you could do something like
kiobufs, or the stream head allocator loanout stuff that's been yityatted
about for so long.
Of course, I also would prefer to use splice!
Now *that'd* torpedo this VI foolishness....
-matt
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Sat Sep 23 2000 - 21:00:17 EST