Sebastien Roy wrote:
Much of the information in IRE cache entries can be kept on a per
connection basis (in conn_t), or is already stored elsewhere and can be
referenced from conn_t. This reduces the overhead of maintaining the
IRE cache, whose cost outweighs the potential benefits of a having a
cache in the first place (in these kinds of workloads).
This can be true. But note that conn_t is not shared. This means
that the common info right now in an IRE cache is needed to be
replicated in each conn_t of all connections to the same remote peer
using the above scheme. Also note that an IRE cache stays after a
connection is gone. So the next connection to the same remote peer
can make use of what is learnt from the previous connection stored
in the IRE cache. I think that the current scheme is not good, but
I guess we need more thinking about it.
In scenarios where the system is initiating connections, this obviously
only helps us if the connections are short-lived and many of them are to
the same destination.
An example of this may be an accelerating webcache connecting
to a backend webserver. Thus a lot of connections are made to
the same server.
--
K. Poon.
[EMAIL PROTECTED]
_______________________________________________
networking-discuss mailing list
[email protected]
