summaryrefslogtreecommitdiff
path: root/TAO/ChangeLog-99c
diff options
context:
space:
mode:
Diffstat (limited to 'TAO/ChangeLog-99c')
-rw-r--r--TAO/ChangeLog-99c30
1 files changed, 30 insertions, 0 deletions
diff --git a/TAO/ChangeLog-99c b/TAO/ChangeLog-99c
index 714445a6c04..7cfd3c2542b 100644
--- a/TAO/ChangeLog-99c
+++ b/TAO/ChangeLog-99c
@@ -1,3 +1,33 @@
+Fri Mar 12 00:35:47 1999 Marina Spivak <marina@cs.wustl.edu>
+
+ * orbsvcs/orbsvcs/IOR_Multicast.cpp
+ Fixed the problem of client not being able to resolve Naming
+ Service ior through multicast when client and server are run on
+ different hosts. Thanks to Steve Totten for tracking down since
+ when this was broken, and to Carlos O'Ryan for providing
+ hypotheses as to why.
+
+ Made the server wait a little bit before sending a reply to the
+ multicast as well as send a reply several times. This 1) gives client
+ a chance to start listening for a reply, 2) provides a better
+ protection against packet loss. (The problem was occuring because
+ the server sent a reply before the client had a chance to start
+ receiving. This was happening due to the added dynamic memory
+ allocation in the client code in the interval between
+ multicast_send and reply_receive).
+
+ A more flexible policy-based solution should be added (as opposed
+ to hardcoding intervals and retries). This will
+ probably be a part of Vishal's upcoming Interoperable Naming Service changes.
+
+ Values encoded in the code now should work for most
+ platforms/networks. However, they are not guaranteed to work for
+ all! Part of the problem is that UDP is unreliable, so in theory
+ can be retransmitted 100 times and still not work. Also, specific
+ network configurations (busy network, busy hosts, old routers,
+ etc.) may have special needs. Again, this should be better
+ addressed once hardcoded values are replaced by flexible policies.
+
Thu Mar 11 22:49:35 1999 Marina Spivak <marina@cs.wustl.edu>
* tao/ORB.cpp (multicast_query):