large messages

[Bryan Ford]

>  >How is larger-than-4K message transfer generally done in KeyKOS
>  >between a client and a server that doesn't want to trust the client's
>  >promptness?
>
>A shared server domain wouldn't do a ping-pong transfer directly. It
>would create a domain to do this, using the factory mechanism or
>equivalent. It creates the domain using client-supplied resources. That
>is why it is necessary to pass a meter and one or two space banks on the
>very first call.

Does the meter have to be one that is guaranteed not to run out
at an arbitrary point and stay dry forever, or can it be just any meter?

How commonly are two space banks needed?  I heard mentioned previously
that one must be trusted, but the other does not have to be.  What exctly
does "trusted" mean in this context?  That resources allocated from
that space bank can't be arbitrarily revoked by the client or someone else?
And/or that the pages and nodes allocated will not be accessible by anyone
except the server?  (Or is the latter a general property of space banks?)

>By the way, the return code returned by a data key (which is KT+1, not
>KT+2, Bill) is not reserved; a real key can also return that value.

BTW, what exactly is 'KT'?  :-)

Also, it would be very helpful to me if someone could post a short
description of the actual register conventions used for invoking
the primitive kernel messaging operations, on both the i386 and 88000
if possible.  I'm still trying to get a clear picture of the exact
semantics of KeyKOS message passing...

Thanks!
				Bryan