MBUF Cluster Network freeze problem

[Antti P Miettinen]

> What good would time-slicing do? If a task holds the network
> semaphore no other task can obtain the semaphore, so there would be
> no point in trying to run other network-related tasks by enabling
> time slice scheduling for the network tasks.

You are correct. If the network tasks hold the semaphore always when
busy (which more or less seems to be the case) the time-slicing would
not have any effect. Time-slicing might be relevant if the network
tasks would obtain the semaphore only when accessing shared resources.
The current system probably makes writing device driver tasks safer.

I'm not trying to suggest that time-slicing would solve the problem. I
was just curious about why time-slicing is disabled. I was bitten by a
scheduling issue once when I was running ttcp :)

Scheduling is probably the issue for system "freezing" during flood
ping but there's something fishy somewhere if the system does not
recover. My 302 board handles flood ping just fine :)