Hi!

Living on a DSL link is hard when it comes to ALTQ configuration on the
upstream side.
If you are able to configure both sides of a link policy based routing
is no problem (i.e. prioritizing ACKs or icmp/udp/Diablo 2 :)).
In my (and prob. this is the most common setup) I cannot (legally) take
control of the upstream router and its queueing policies.
As a result when my downstream is busy ACKs, icmp errors, nameserver
queries et al. are treated like normal bulk data packets and not
prioritized.

When setting TOS fields on outgoing packets the same bits are set in
the answer thus allowing basic priorization for incoming packets
on the upstream router:

$ nemesis icmp -qE -S 10.10.1.2 -D 195.244.235.251
15:48:25.783347 10.10.1.2 > 195.244.235.251: icmp: echo request (ttl
255, id 42131, len 28)
15:48:25.852952 195.244.235.251 > 10.10.1.2: icmp: echo reply (ttl 245,
id 2534, len 28)

$ nemesis icmp -qE -S 10.10.1.2 -D 195.244.235.251 -t 0x10
15:48:27.033848 10.10.1.2 > 195.244.235.251: icmp: echo request [tos
0x10]  (ttl 255, id 16345, len 28)
15:48:27.168675 195.244.235.251 > 10.10.1.2: icmp: echo reply [tos 0x10]
 (ttl 245, id 33430, len 28)

Since the line is not completly busy as of now there are no noticible
round trip time differences in this example beside the set tos fields.

I'd like to implement a 'scrub' option which basicly works like this:

scrub out on $ext_if proto udp from $internal to $nameservers port 53
keep state tos 0x10

There are many other ways of using this option but the example should
give you a brief overview of my intention.

I talked to Henning since I didn't know if there were any issues since
this has not been implemented as of now but could not remember any
reasons.
Even now when ECN et al. are coming up this feature looks interesting to
me.
Looking at the source (I'm somewhat familiar with the pf FreeBSD port)
inserting the appropriate option right after the random IP ID setting
in pf_norm.c seems to be easy.

Before starting setting up an OpenBSD box I'd like to know if there
are any caveats/reasons since this has not been done already.

Thanks, Hendrik

-- 
Hendrik Scholz - <[EMAIL PROTECTED]> - http://raisdorf.net/

drag me, drop me - treat me like an object 

Reply via email to