[Click] Traffic Shaping in Click

[Latency Buster]

> ToDevice drops occur as ToDevice can sometimes insert packets faster than
> the device can process them.  If you want, I can email you a patch > that
> prevents that from happening.  (all the drops would occur in the queue then)

Please do so...Would you able to comment on the stability of the
patch? I'm pushing my univ. to use click as a NAT box ..


> Is that using one thread or two threads?  Is the traffic bidirectional > or
> not?  If you have a single threaded config with two PollDevices, > then you
> will have lower throughput as both PollDevice elements will > compete for CPU
> time.
>
Traffic is bidirectional. I am using two threads scheduled on two
different cores. But the surprising thing is that MT click gives me
lower throughput.. I have not looked into too much details as of now

> You can try to increase the BURST parameter of PollDevice to see if you can
> improve the packet rate.  In addition, you can also vary the TX/RX sizes on
> the device via 'ethtool'.
>

I have not tried this option (burst size) and also with the rx/tx
queues with ethtool. But I shall see if this affects the performance
or not.