RE: auth handshake and rendevouz objects

[Stefan Gotz]

> Marcus Brinkmann wrote:
> > You don't want the server do the work for A.  All the burden of
> sending a
> > blocking message to B should be on A, not on S.
>
> Something my prof and I were discussing: why should A block on sending a
> message to B? Doesn't it make more sense for A to create a "courier"
> thread that does the blocking IPC on A's behalf?

Yep - that's how you can emulate asynchronous IPC on L4. But the important
part is that the client A has to deal with blocking on B instead of the
server S.

In case you actually meant S instead of A who should use a courier thread
the problem is stated in Marcus' first sentence: creating new threads makes
a handle transfer a rather expensive operation for the server. Thus it
becomes more susceptible to DOS attacks.

This is actually an accounting problem - as stated, clients should pay for
services they request, not servers for services they provide. I don't think
current L4 versions allow to implement an appropriate accounting scheme to
achieve that. So we have to try to keep resource consumption in servers
minimal (or at least low, to be realistic).

Stefan