Re: [jgroups-users] total ordering protocols (sequencer)

[apalade <and...@go...>]

> Note that there is an experimental version of SEQUENCER2 [2], but it needs
some more work 

yes, this (i.e. SEQUENCER2) is based on a variant of the SEQUENCER protocol
which is already implemented in JGroups. The purpose of SEQUENCER2 is to
offload the coordinator.  Both protocols are based on Fixed Sequencer class
(please see [1]). The difference between the two is who handles the
broadcast of the messages after allocation of the sequence numbers. In
SEQUENCER, the leader (i.e. coordinator) broadcasts the messages, wheres as
in SEQUENCER2 the node that initiates the message is in charge of
broadcasting the message to the rest of the nodes in the group. 

SEQUENCER implements the variant unicast-broadcast (see variant UB in [1],
section 4.1). SEQUENCER2 is based on the variant unicast-unicast-multicast 
(see variant UUB in [1], section 4.1).  In variant UUB, a request is made to
the coordinator for each message that the sending node wants to broadcast.
SEQUENCER2 improves the variant UUB by requesting from the coordinator a
range of sequence numbers for the messages accumulated in the sending node.
The coordinator replies with a range and the sending node is in charge of
allocating the numbers to each message in the queue and broadcast the
messages. Hence, only one request is made to the coordinator.

I worked on a paper regarding this issue, but is not published yet. However,
initial tests show that SEQUENCER2 performs a bit better than SEQUENCER for
the proposed use case that the paper is based on.


[1]. Xavier Défago, André Schiper, and Péter Urbán. 2004. Total order
broadcast and multicast algorithms: Taxonomy and survey. ACM Comput. Surv.
36, 4 (December 2004), 372-421. DOI=10.1145/1041680.1041682
http://doi.acm.org/10.1145/1041680.1041682 



--
View this message in context: http://jgroups.1086181.n5.nabble.com/jgroups-users-total-ordering-protocols-sequencer-tp10345p10347.html
Sent from the JGroups - General mailing list archive at Nabble.com.