LARTC - Feb 2004 - Dynamically changing ceil (and removing filters)

I''m investigating other approaches to managing the very little,
extremely over-used pipe we have feeding the residences of our small
campus.

Previously, I simply allotted them 2Mbit (there are about 100 users,
some full-time users, most not), and did a number of prioritizations to
improve responsiveness.  I used HTB and attached ESFQ qdiscs to evenly
distribute bandwidth based on IP, rather than per-session.  This worked
quite well until recently, where the number of users has increased and
demand has gone up.

One of the goals is that heavy users (basically p2p users) shouldn''t be
able to ruin it for the others.  The average user should never have to
sit and wait for 5 seconds while google loads, or get 4000ms pings to an
upstream router.

One of the approaches I''m fiddling with now is creating a separate
class
for each IP, guaranteeing them a small portion of bandwidth, to at least
keep their latency down.  I''ve also introduced weekly quotas, and am
significantly penalizing users who go over their quota (such as
rate-limiting them to 5KB/s).  It''s awfully harsh, but the proper
solution (buying more bandwidth) is not my call, and so I have to make
this 2Mbit work for now.

Rather than rate-limiting each over-quota user to 5KB/s, I''ve been
considering creating a separate, larger class, and just tossing each of
the over-quota users in it to fight for the bandwidth in that class. 
However, the ceil for this class should depend on the number of users in
it, because I must assume that each user in the class will try to use
all of the allowed bandwidth in that class at all times.

So what I''d like to do is provide a ceil in the HTB class that is
computed based on the number of users in the class.  I''ve rolled my own
scripts that monitors usage of each user via iptables, so I could update
those scripts to change the ceil each time a new user is added (tc class
change dev ...), but I was wondering if there was a more elegant way of
doing this -- that is, does HTB somehow support dynamic ceilings based
on the number of users (IPs) it sees in that class?

Another problem I''m having is that I can''t seem to find the
right syntax
for removing filters.  Can anyone help with that?  The most obvious
doesn''t seem the work:

   [root@hawk ipacct]# tc filter add dev $ETH_RES protocol ip parent 2: u32
match ip dst 10.0.0.100/32 flowid 2:10
   [root@hawk ipacct]# tc filter del dev $ETH_RES protocol ip parent 2: u32
match ip dst 10.0.0.100/32 flowid 2:10
   RTNETLINK answers: No such file or directory


Thanks much,
Jason.

-- 
Jason Tackaberry  ::  tack@auc.ca  :: 705-949-2301 x330 
Academic Computing Support Specialist
Information Technology Services
Algoma University College  ::  www.auc.ca


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

> One of the goals is that heavy users (basically p2p users)
shouldn''t be
> able to ruin it for the others.  The average user should never have to
> sit and wait for 5 seconds while google loads, or get 4000ms pings to an
> upstream router.
Well, again someone struggeling p2p and it''s side effects. What about
not
letting P2P users download anymore? There is a netfilter extension called IPP2P
( http://rnvs.informatik.uni-leipzig.de/ipp2p/ ) that can be used to drop
certain P2P packets thus making down- and uploads of the supported p2p-networks
almost impossible:

iptables -t mangle -A PREROUTING -p tcp -m ipp2p --ipp2p --bit --apple -j
DROP

If this would be an option to you give it a try and come back to me with the
results.  We ran such a system for about 6 weeks at a university link and
found it beeing pretty stable. If you don''t want to recompile the
kernel and
iptables grab the source code do a make and copy the modules to the appropriate
place. Look at the documentation section at the webpage and the README
included in source tarball. Hope that helps.

Regards,
Mike

-- 
GMX ProMail (250 MB Mailbox, 50 FreeSMS, Virenschutz, 2,99 EUR/Monat...)
jetzt 3 Monate GRATIS + 3x DER SPIEGEL +++ http://www.gmx.net/derspiegel +++

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

> Previously, I simply allotted them 2Mbit (there are about 100 users,
> some full-time users, most not), and did a number of prioritizations to
> improve responsiveness.  I used HTB and attached ESFQ qdiscs to evenly
> distribute bandwidth based on IP, rather than per-session.  This worked
> quite well until recently, where the number of users has increased and
> demand has gone up.
I solved that problem by measuring all clients, and give them optional
average individual bandwidth.

The bandwidth is calculated between different part of day /eg. 8-16, 16-24,
24-8 / and is increased or decreased by individual algorithm for each user
too.

All the work with network-coordination, user-problem-reporting,
managing-and-controll, devices-location-access-controll, dhcp-dns-etc,
/somehow CRM for ISP / does my application called SuperEdit. It works in
company with 3000 PC''s over LAN''s .
you can read about it /sorry for large .jpgs/ at
http://nsm.pl/~arek/superedit . I have a version of that in english.

Access is via www and/or with telnet/ssh - together/the_same_options

http://nsm.pl/~arek/superedit/superedit-software-10.jpg - you see brief
list of user with problems

http://nsm.pl/~arek/superedit/superedit-users2-widok-abonenta-i-szafki-2003
-12.jpg - you see some of user fields, which we can edit, you see traffic
report collected for that user from network. You see that we set there
external ports, external IP addresses. Linux router can be anywhere, he is
learning of that DB via special written by me bash Agent.

Browse this site /download images/ then you can learn a lot.

I can share my project in special cases.

It is tested, and working / i organized a medium copany on it/ They have
only this programm and billing program /which i intend to swap him soon
too/


Arkadiusz Binder

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

On Friday 06 February 2004 16:36, Jason Tackaberry
wrote:
> I''m investigating other approaches to managing the very little,
> extremely over-used pipe we have feeding the residences of our small
> campus.
What wbout wrr qdisc:
http://wipl-wrr.sourceforge.net/
The more you download, the slower it goes.

Stef

-- 
stef.coene@docum.org
 "Using Linux as bandwidth manager"
     http://www.docum.org/
     #lartc @ irc.openprojects.net
_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/