Hi, experimenting with the MTU patch, I discovered that we basically sit idle for two seconds on the client between "TLS is up!" and "PUSH_REQUEST".
This is part due to the coarse granularity of, well, our coarse timers,
but in part due to *two* timers having to fire...
- TLS is done, we schedule an "wait_for_connect" timer to fire in +1 second
(init.c, do_init_timers())
- when that timer fires, 1 second after TLS completes, we set up *another*
timer inside check_connection_established_dowork(), that is intended to
fire in +1 second
- +1 second later, we send PUSH_REQUEST
due to the calling sequence and check_push_request() being called right
after check_connection_established(), it seems to be fairly safe to just
set up the event to "fire right away" (*having* the event is needed to
be able to reset it to "fire again in +5 seconds").
So, with the patch below on top of the MTU patch, connection times to a
server which is 150ms away from me went down from 5 seconds to 2 seconds
- which I consider quite significant :-)
Anyone better versed in the history of this code who can tell me whether
this is safe, or why the extra delay is there?
gert
diff --git a/src/openvpn/forward.c b/src/openvpn/forward.c
index 6d459d2..e1571ec 100644
--- a/src/openvpn/forward.c
+++ b/src/openvpn/forward.c
@@ -212,8 +212,8 @@ check_connection_established_dowork (struct context *c)
0);
}
#endif
- /* send push request in 1 sec */
- event_timeout_init (&c->c2.push_request_interval, 1, now);
+ /* fire up push request right away (already 1s delayed) */
+ event_timeout_init (&c->c2.push_request_interval, 0, now);
reset_coarse_timers (c);
}
else
--
USENET is *not* the non-clickable part of WWW!
//www.muc.de/~gert/
Gert Doering - Munich, Germany g...@greenie.muc.de
fax: +49-89-35655025 g...@net.informatik.tu-muenchen.de
pgpEGXp7vmauJ.pgp
Description: PGP signature
