All of lore.kernel.org
 help / color / mirror / Atom feed
* [LARTC] Small ISP problems (CBQ)
@ 2001-12-07 23:06 Paul Wisén
  0 siblings, 0 replies; only message in thread
From: Paul Wisén @ 2001-12-07 23:06 UTC (permalink / raw)
  To: lartc


Hi all!

We are a small ISP in Sweden. We have built a radiobased network and shares
2Mbit full duplex among roughly 50 users. (so far..)
To ensure a good quality of service we are today running CBQ on a central
machine in our network. I'm not sure it works the way we want though and
have a couple of questions that maybe some of you alread know and can help
me/us with.

First of all, this is what we want (in network priority order):
1: SSH - to be realtime always.
2: HTTP to be fast, always.
3-> ftp, direct-connect, kazaa and others to be throttled to X bandwidh per
IP.. (or not disturb http and ssh and use real fair quing.. )

So, what we did was to set up a bunch of CBQ scripts.
Defined a SSH class and gave it prio 1.
Defined a http class and gave it prio 2
all other prio 7.

This looked like a fine idea but we experienced that ssh and http suffers
when the "other" class gets chunked.
So, the Fair Quing algorithm does not seem to work. (SFQ). The other class
borrows bandwith from http and ssh classes and those gets chunked to..
So, our solution was to set up a incoming and outgoing class for each IP in
our network. (kind of a mess..)

A first question then: Is there a way of defining a class that basicly says:
" each and every ip in this class shall have X bandwidth and nothing more" ?
(or do we need to define bounded classes for each and every customer ? )

Next: Is there a way of having multiple classes that could trigger on the
same packet ? (and control witch class will get it) ?
Ex: We have a customer that has bought X bandwidh. So, we define a class on
that IP thats bounded to X bandwidh. This, however shall not be the case if
the packet is a ssh packet. Then it shall never end up in the customer class
at all and instead end up in the high prio ssh class ??
I guess the question ends up with, Is there a way of having multiple
classes/rules and the first class/rule that triggers on the packet will get
it ? Somewhat like in ipchains ..  ?


And since I'm up to speed... If everything gets chunked, we start experience
packet loss.. I guess that's quite normal. We have however a 2Mbit
connection to the Internet.
What happens if we do, as we have done now, and defines the interface to
2Mbit and then defines a lot of classes of 256Kbit (more then 2Mbit total)
..  Will CBQ send more load then 2Mbit to the internet connection (and
thereby make other classes like http to go slow..) ??

Anyone out there that has some input to us ?


Best Regards

/ Paul


_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://ds9a.nl/2.4Routing/

^ permalink raw reply	[flat|nested] only message in thread

only message in thread, other threads:[~2001-12-07 23:06 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2001-12-07 23:06 [LARTC] Small ISP problems (CBQ) Paul Wisén

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.