Connection Factories in a Cluster

Question

Hi Folks,
          I'm trying to get to grips with the precise behavior of connection factories
          (CF) in a cluster. Any corrections/additions appreciated...(I've read the
          perf. guide).
          According to the docs, you should look at two scenarios
          a) JMS app located on same server as cluster.
          Imagine I have a JMS app deployed to the cluster, and docs say I should
          deploy CF to cluster as well.
          I have JMS server only on one server in the cluster.
          Here is what I think happens:
          i. As far as I know, connection factories will only load balance
          connections across servers hosting connection factories.
          ii. When the app looks up a CF using the JNDI of the cluster, will it
          always get the a CF collocated
          with the app (ie. on same server), because of collocation optimization?
          iii. When the app uses this CF to create a Connection, will the CF load
          balance the connection?
          ie. Will I get a different connection (round robin) every time I
          create a connection?
          If this does happen, then presumably my JMS application will route
          all JMS traffic through this connection to this other
          server, which may not even host a JMS server, which in turn has to
          route to the JMS server.
          iv. If the CF was looked up using the local JNDI context, then the
          local CF will be returned and presumably if
          I create a connection then a local connection will be returned. ie.
          no load balancing. The JMS application will
          use the "local" connection which will in turn route to the
          appropriate server hosting the JMS server..
          b) Where the JMS client is external
          When I have an external JMS client, the docs say you should deploy the CF
          only to those servers hosting the JMS servers.
          i) The docs say that the CF (new InitialContext("t3://clusteraddres"))
          will load balance across the servers in the cluster.
          ii) The docs also say that createConnection (on the CF) will load balance
          too.
          iii) This means that there are two servers involved right? The CF host
          (the server hosting the CF), and the C host (the
          server hosting the C). Does this mean I have two connections from
          teh client (one to each server), or does traffic
          flow from client to CF host to C host?
          The docs certainly imply that you get these two load balancing
          operations, which seems to lead to one of these
          scenarios.
          iii) Presumably, if my client creates multiple connections (one after the
          other) using the same connection factory,
          then it may at one time have a connection going to "ServerA", and at
          another time one going to "ServerB" (unless
          of course I have affinity enabled).
          Any guidance appreciated,
          Regards,
          Jon
          

Answer 1

          Jon Mountjoy wrote:
          > Hi Tom,
          >
          > I have inlined some comments on your comments!
          >
          >
          >>>a) JMS app located on same server as cluster.
          >>> Imagine I have a JMS app deployed to the cluster, and docs say I
          >
          > should
          >
          >>>deploy CF to cluster as well.
          >>> I have JMS server only on one server in the cluster.
          >>>
          >>> Here is what I think happens:
          >>> i. As far as I know, connection factories will only load balance
          >>>connections across servers hosting connection factories.
          >>
          >>Yes.
          >>
          >>> ii. When the app looks up a CF using the JNDI of the cluster, will
          >
          > it
          >
          >>>always get the a CF collocated
          >>> with the app (ie. on same server), because of collocation
          >
          > optimization?
          >
          >>Yes.
          >>
          >>> iii. When the app uses this CF to create a Connection, will the CF
          >
          > load
          >
          >>>balance the connection?
          >>
          >>Not if it is collocated.
          >
          >
          > I guess this confuses me a little. My scenario above speaks about the
          > CF+App deployed to a cluster.
          >
          > Are you talking about if the CF+App is collocated with the JMS server whose
          > destination the App is communicating with?
          No.
          > (I'd guess not, cos I wouldn't
          > know the destination until after I create the connection?).
          If the app is co-located with the the CF, the connection
          is always local.
          >
          >
          >>(Note that a producer to a distributed destination,
          >>by default produces to the local physical instance,
          >>and will not round-robin each message to other
          >>physical destination instances, unless the
          >>connection factory is so configured.)
          >>
          >>
          >>> ie. Will I get a different connection (round robin) every time I
          >>>create a connection?
          >>> If this does happen, then presumably my JMS application will
          >
          > route
          >
          >>>all JMS traffic through this connection to this other
          >>> server, which may not even host a JMS server, which in turn has
          >
          > to
          >
          >>>route to the JMS server.
          >>> iv. If the CF was looked up using the local JNDI context, then the
          >>>local CF will be returned and presumably if
          >>> I create a connection then a local connection will be returned.
          >
          > ie.
          >
          >>>no load balancing. The JMS application will
          >>> use the "local" connection which will in turn route to the
          >>>appropriate server hosting the JMS server..
          >>
          >>Yes.
          >>
          >>
          >>>b) Where the JMS client is external
          >
          >
          >
          >>> iii) This means that there are two servers involved right? The CF host
          >>>(the server hosting the CF), and the C host (the
          >>> server hosting the C). Does this mean I have two connections
          >
          > from
          >
          >>>teh client (one to each server),
          Yes. But not for the reason you think.
          The JNDI context that was used to look up
          the CF keeps its connection, and the JMS
          connection generated from the CF.
          >>> or does traffic
          >>> flow from client to CF host to C host?
          Yes.
          >>> The docs certainly imply that you get these two load balancing
          >>>operations, which seems to lead to one of these
          >>> scenarios.
          >
          >
          > Can you perhaps shed some light on the above question?
          >
          >
          >>>Any guidance appreciated,
          >
          >
          >>Just send a portion of your book royalties. ;-)
          >
          >
          > Hehe. Do you think I'd actually make a profit!? What you will earn, is my
          > eternal gratitude of course :)
          >
          > Regards,
          > Jon
          >
          >