Release Notes JGroups 2.6.2 =========================== Version: $Id: ReleaseNotes-2.6.2.txt,v 1.4.2.4 2008/02/25 16:26:33 belaban Exp $ Author: Bela Ban JGroups 2.6.2 is still API-backwards compatible with previous versions (down to 2.2.5). JGroups 2.6.2 is also binary backwards compatible with 2.6 and 2.6.1. Below is a summary (with links to the detailed description) of the major new features between 2.6.1 and 2.6.2. Shared transport ---------------- [http://jira.jboss.com/jira/browse/JGRP-631] When running multiple channels within the same JVM, we can make all channels refer to the same transport (e.g. UDP, or TCP). The transport then effectively becomes a JVM singleton. Note that the protocols of the individual stacks can be different, but the shared transport is the same. The advantage of this is that costly resources like thread pools are shared between channels, reducing overall resources (e.g. fewer threads). Compared to MuxChannels, shared transports do *not* change semantics, so there are no service views etc. Additionally, there is no separate thread pool needed, like in the Multiplexer. So shared transports can be used as a replacement for MuxChannels. Documentation is at http://www.jgroups.org/javagroupsnew/docs/manual/html/user-advanced.html#d0e2325. FLUSH: flush a subset of the cluster ------------------------------------ [http://jira.jboss.com/jira/browse/JGRP-661] This is needed for buddy replication in JBossCache for example, where we don't want to flush the entire cluster, but just a few members. Advantage: not everyone needs to block during the flush phase; other members can continue sending messages. Eager lock release in NAKACK and UNICAST ---------------------------------------- [http://jira.jboss.com/jira/browse/JGRP-656] Determines when a message can be considered 'delivered'. When set to true, and the message receiver has received it, but calls down() on the same thread, the lock on the sender is released when eager_lock_release is set (default=true). This allows another thread to acquire the lock and process the next message *from the same sender*. This results in performance gains, and does *not* change the ordering of messages from a given sender. Note that messages from different senders are not affected, as they're delivered concurrently anyway. Hooks to provide ones own thread factory and/or thread pools (transport) ------------------------------------------------------------------------ [http://jira.jboss.com/jira/browse/JGRP-683] Users can now replace the thread factories which create all threads. Additionally, the default and OOB thread pools in the transport can be replaced / modified as well. PING can now use multiple GossipRouters --------------------------------------- [http://jira.jboss.com/jira/browse/JGRP-684] This is for redundancy; a member can register itself with multiple GossipRouters, and all will get pinged for initial membership. This eliminates a single point of failure. TCPPING: discovery done in parallel ----------------------------------- [http://jira.jboss.com/jira/browse/JGRP-375] Instead of sequentially pinging all listed members (some of which might block), we do this in parallel. This reduces chances of not getting a response from a member. Manual ------ The manual is online at http://www.jgroups.org/javagroupsnew/docs/manual/html/index.html Bug fixes --------- Multiplexer and merging: incorrect merging in some scenarios [http://jira.jboss.com/jira/browse/JGRP-665] Multiplexer ignores OOB flag [http://jira.jboss.com/jira/browse/JGRP-663] Messages get too big for UDP transport [http://jira.jboss.com/jira/browse/JGRP-667] The complete list of features and bug fixes can be found at http://jira.jboss.com/jira/browse/JGRP. Bela Ban, Kreuzlingen, Switzerland Vladimir Blagojevic, Toronto, Canada Richard Achmatowicz, Montreal, Canada Feb 2008