Hi,
Given the available bandwidth on my network interface and other tcp parameters, how can I estimate queue length sufficient to avoid overflow?
Thanks,
-V

OK, since no one has replied to this, I'll start the ball rolling, even though this may just display my ignorance. I sort of understand what you are asking but could you clarify a bit what is your concern? It's my understanding that overflow of queues only happens under certain circumstances, for example receiving from an uncontrolled source such as the Internet. For transmission, if the local machine processes dumping data to the queue cause the queue to fill, they will simply block until space is available. If a receive queue overflows, then the incoming packets are discarded but the protocol will act to cause them to be retransmitted. Local networks may be able to respond quickly enough at the lower datalink or hardware level to avoid activating the higher network level or end-to-end protocols. Do you have a general concern to know this, or is there a more specific project being engaged?

Ken

Come on. How can bandwidth control (or ant QoS) could cause queues?
And what are those "other" TCP parametes?

I should have never refer to protocol with congestion control mechanism...
Let's assume "udp" instead.
My questions really are:
- are there multiple queues in the kernel, or the single one?
- how the capacity for input kernel queue is selected?
- Is the capacity set in some file?
With the second clarification: i am working with the linux kernel-based router project.

Unfortunately the details of kernel operation is not in my range of expertise. You would probably be best either to pose the question in terms of "what does the kernel do with TCP/UDP queueing" to attract the attention of those in the know, or ask on a more specialized forum for people working on the kernel.

cheers, Ken