bwm-tools-tech — General & Technical Discussion

[bwm-tools-tech] Re: about bwm

[Nigel K. <nk...@lb...>]

On Wed, Jun 18, 2003 at 09:49:31AM +0300, Andriy T. Yanko wrote:
> I am working on specific program that control access traffic flow.
> I am also using QUEUE kernel module.
> Everything was fine  & working but now I got order to add shaping methods=
 and  ...

yea... that gets a little hard ... heh

> I have now problem beause I don't  know how to implement one of famous Qo=
S ( TBF,  HTB ... ) or other.

we basically use first-in-first-out (fifo), anything can basically be
added.

> I also wrote wrapper for  tc  (utilite from iproute2)  .

I also started out on TC a very long while ago, had more drawbacks for
our use than usefullness, so we scrapped it... tried out some of our own=20
in-kernel code, scrapped that aswell... now it looks like we going to
stay with bwm-tools. As you can see from the source we started work on
it close to a year ago. ;)

> But kernel QoS  using  on ethernet  devices and most as egress mode.

yea

> So in this case I must build classes with queuening disciplines and use t=
c filter to match packets that I want shape,
> This  approach is not very comfortable: seting some rate or some rate tra=
ffic flow is not very hard but getting
> cuurant rate or some othe statistics this tc is a problem ( I also tried =
to  use libnetlink library but its's very hard ).
>=20

I agree, this is the drawback I mention above, bwm-tools also supports
"classes" in the sense that it is used in TC as a parent->child->child
scenario we support the exact same thing, with unlimited children
allowed all with their own configurable traffic rates and queues.

> Method of control traffic flow in bwmd looks more better (I don't  know h=
ow good it is for system  performance).
> In this case program may easy get current flow speed and  etc...

Well, a 433MMX Celeron can maintain an average of 5,500 packets per
second with a 84byte total packet size and can maintain about 2,800
packets per second with a 1500 packet size.

I"m not sure exactly how good these results are, but I guess its not bad
at all.

>=20
> Can you explain me  how bwmd working? I saw the code & basic principle i =
understand.

There is a thread which accepts traffic from ip_queue, deciding if it
needs to be dropped (if the queues are all full).  (NF_DROP)

There is then another thread which processes the flows, sending the
packet from a child, to its parent, to its parent and so on, this is
responsible for accepting the traffic to flow through bwmd. (NF_ACCEPT)

The other threads are STAT & REPORT, stat keeps up the averages used for
bursting and rates used in bwm_monitor (which really is cool, I decided
to write bwm_monitor because I would like to see EXACTLY what speed my
traffic flows are using). Report is used to write counters to the log
files.

I would like to use bwm_monitor in future to actually live configure
bwmd, and also create graphs with a nice cursor interface... but that is
for much later when I have volunteers to help me.

> I am VERY  interesting to part code of enqueuenig/dequeuning packets with=
 ip_queue kernel module.

Cool, maybe you can help us out... we just accept from ip_queue... so
what you could do is actually hadd different queueing diciplines to bwmd
(ie. teach it how to queue properly)

> What QoS  You are using as shaping or it is Your own approach?

FIFO, I know its stupid... but was all I could put together in a short
period. What I would like to do is add priorities to bwmd, so there is a
standard order in which packets are accepted, then depending on the
different queueing diciplines you can then determine the priority of the
packet. This is the proper approach because if you have a parent flow
and a child flow, you will need a "standard" way in which packets are
classified so as to beable to have different diciplines used for both
the child and the parent whilst at the same time accepting packets to
pass through in the order which they supposed to.

> maybe you khow some urls about such probles.

I think QOS diciplines are defined in RFC's, I've read a few documents
on the net which I got from a google search. Most of these explain how
the diciplines work... it would be a large job to implement them but
would be worth your while!=20

> Any suggestion is welcomed.

Join our team of developers? hehehe, I would be prepared in any way I
can to help with any queries you have. I must admit that the current
flow processing functions are a little messy and needs cleaning up.
Basically priority of packets must be added. Then when a packet is
received with the netlink (ip_queue) read, they can be classified based
on the queuing dicipline used by the particular flow. The packet can
then either be re-classified by its' parent, or the priority it has can
be used.

>=20
>=20
> THANKS IN ADVANCE.

No problem.

-Nigel

[bwm-tools-tech] principle

[Andriy T. Y. <wir...@wi...>]

Hi people

Can anybody give short  tech overview of work bwmd daemon?

I am interesting in principle how to enqueue/dequeue packet with lipipq in special queue
& how to shape traffic.



-- 
Andriy T. Yanko
wir...@wi...

*   Avoid The Gates of Hell use Linux
**  Wenn ich sterbe, legt mir eine Birne in mein Grab ...
*** Я конєчно діко ізвіняюсь который час? Вже шєсть. ((c) Йшов трамвай N9)

[Bwm-tools-tech] (no subject)

[Jonathan S H. <jon...@wa...>]