[LARTC] (E)SFQ suggestion

[LARTC] (E)SFQ suggestion

[Martin Devera]

Hi,
just simple note. Maybe it is already in progress :)

There are attempts to replace hashing routine in SFQ to
consider IPs or ports.
What about to use HRR - roundrobin around bunch of IP
adresses and then smaller WRR for ports per IP ?
It would solve both problems - fairnes between computers
(IP) and between flows on than single computer ...

regards, devik

_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

Re: [LARTC] (E)SFQ suggestion

[Michael T. Babcock]

> There are attempts to replace hashing routine in SFQ to
> consider IPs or ports.
> What about to use HRR - roundrobin around bunch of IP
> adresses and then smaller WRR for ports per IP ?
> It would solve both problems - fairnes between computers
> (IP) and between flows on than single computer ...

Basically you're suggesting a multi-staged hash bucket ... hash each packet
twice; once against the source IPs, then against the same criteria as now.
Do a loop through the first hash list, pick a bucket, then use the second
hash to pick packets from that bucket.  Any thoughts?
--
Michael T. Babcock
CTO, FibreSpeed Ltd.

_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

Re: [LARTC] (E)SFQ suggestion

[Alexander Atanasov]

On Thu, 6 Jun 2002, Martin Devera wrote:

> Hi,
> just simple note. Maybe it is already in progress :)
> 
> There are attempts to replace hashing routine in SFQ to
> consider IPs or ports.
> What about to use HRR - roundrobin around bunch of IP
> adresses and then smaller WRR for ports per IP ?
> It would solve both problems - fairnes between computers
> (IP) and between flows on than single computer ...

	HRR wouldn't that make things complicated?
I think it's better to have two simple separate solutions and a way to 
combine them.
What you suggest means to put SFQ into SFQ - just different
classification. Some may wish even to have more levels -
SFQ into SFQ recursive. 
Isn't it possible to arrange this with cbq and sfq leafs?
You probably now the IPs but do not know their connections.
There is a suggestion how to combine SFQ and RED with IMQ,
what about doing same but with SFQ and SFQ -
attach ESFQ with dst hash on IMQ and sfq on the network device?
Adding source/destionation port hash in esfq is very easy task.
A better solution would be to make classful SFQ. 
Then you put sfq with one classification and attach to it
sfq but with different classification. Or you can put (G)RED on it.
Would get same as with IMQ but you'll get the N levels.
HRR with WRR would be great when you have to deal with a lot of
IPs, but what is the right way to do it?

-- 
have fun,
alex


_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

RE: [LARTC] (E)SFQ suggestion

[Michael T. Babcock]

> Isn't it possible to arrange this with cbq and sfq leafs?

Yes ... By building a big CBQ/HTB tree with one leaf for each IP ... :)
I'm not sure how efficient / inefficient that is.
-- 
Michael T. Babcock
CTO, FibreSpeed Ltd.

_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/