Speeding up networking, worth a read.

Simon 'corecode' Schubert corecode at fs.ei.tum.de
Wed Mar 8 08:46:58 PST 2006

Miguel Filipe wrote:
I think the dragonflyBSD team must read this, and apply the same
principles to make the DFly networking stack "super-hot".
of course we already read this paper.  i quickly discussed this with 
matt, and i think we agree that we don't feel comfortable with a tcp 
stack in userland.  on the other hand, having the stuff van jacobson 
implemented in dragonfly to try out is something else, but i doubt it 
would go into cvs.

we already do batch processing, basically.  i think bsds always did this 
due to the soft interrupt nature (i might be horribly wrong there). 
while VJ shows some impressive numbers, i'm not sure how they would 
apply to dragonfly.  but in case you are eager to implement this, don't 
let me stop you!  it might turn out to be reall cool[tm], but e.g. for 
tcp we already do something simmilar.  we don't use these lock-free 
lists, but we queue messages to ports.

but of course i appreciate your heads-up to this paper!  maybe you want 
to play with the network stack a little bit?

Serve - BSD     +++  RENT this banner advert  +++    ASCII Ribbon   /"\
Work - Mac      +++  space for low €€€ NOW!1  +++      Campaign     \ /
Party Enjoy Relax   |   http://dragonflybsd.org      Against  HTML   \
Dude 2c 2 the max   !   http://golden-apple.biz       Mail + News   / \

More information about the Users mailing list