[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [PATCH] sch_tbf.c sporadically returning EINVAL

> It is not required. mtu can have any value, provided filter
> does not classify larger packets to this qdisc.

IMO "mtu" should always be at least the size of the interface MTU (if nothing 
else, to avoid confusion), but you should be able to set "burst" to a smaller 
value if you know the filter won't classify larger packets to this qdisc. 
This is the case of an ATM link between two routers where all other 
interfaces are ethernet, for example. Of course, the packet size would have 
to be checked...

One other feature I think would be very useful is "classful TBF", i.e., 
allowing some qdisc to be installed as a child of TBF. This would not only be 
useful for emulation of physical links, but would also be an advantageous 
substitute for the "shaper" driver. Someone (devik, I think) has already 
altered TBF to do just this. What are the chances for such a thing to be 
incorporated in the standard Linux kernel?

> So, if it is wrong, nothing happens but deadlocking tbf.
> Well, it is not much worse than printing an error message. 8)

It depends. If you're an ISP you may get quite a lot of not-so-friendly 
"error messages" from the users...   :-)

> It also shows that something is configured wrongly. 8)

To most people, it shows Linux is starting to look like windows... Yuc!

Rui Prior