On Wed, 11 Apr 2007, Tony Finch wrote:
On Tue, 10 Apr 2007, william(at)elan.net wrote:
How would you see it working with extension ID?
Client sends KEEPALIVE <interval> at the start of the session (or again
later to change the interval). Server sends 1yz reply after interval if
normal reply is taking longer. Client responds to 1yz with CONTINUE or
RSET. This screws up pipelining so the server probably should not offer
pipelining and keepalive, or perhaps the keepalive command turns off
pipelining.
Personally, this seems more like a mental exercise in how 1yz reply codes
could be utilized than a particularly good starting point for dealing with
the timeout/keepalive problem. Why settle for just keepalives, which
don't help in the presence of real connection failure, when something more
along the lines of the CHECKPOINT extension (RFC 1845) would let the
problem be *solved* instead of just made less likely?
Philip Guenther