[tech] hydra (uccrouter 2)

Leighton Haynes dayta at ucc.gu.uwa.edu.au
Thu Sep 21 16:52:42 WST 2000


On Thu, Sep 21, 2000 at 06:04:54PM +0800, Ian McKellar wrote:
> On Thu, Sep 21, 2000 at 06:38:15AM +0800, Leighton Haynes wrote:
> > Wonder if the drivers are
> > bright enough to work out when a packets just being routed, and transfer 
> > directly card->card instead of card->main memory->card ;)
> 
> Yes, but we want to route traffic not just switch/bridge it. I had to turn
> all the blisteringly fast card to card stuff.

This should be achievable. 
1 Message arrives on card, is buffered by card (probably). Generates interrupt.
2 Read header from card. TCP/IP has fixed header size? 
3 Work out destination.
4 Form new header.
5 Send header to outgoing card.
6 Initiate burst transfer from incoming card to outgoing card.

The real question is whether this would give you any speedup, since 2-5 are
probably going to block incoming messages on the card which has buffered, the
message.  I suspect evil fragmentation stuff may make it even worse :)

Leighton...

--

Part-time student. Full-time Programmer. 
Seeking the 36 hour day and the 10 hour working week.
(08) 9272 9058 (Home - like I'm ever there)
0401 335 136 (Mobile - like it's ever on)




More information about the tech mailing list