This discussion is archived
2 Replies Latest reply: Nov 5, 2010 4:07 AM by 632230 RSS

WAN replication

632230 Newbie
Currently Being Moderated
Hi,

I'm considering various solutions to a latency problem of an app with trading desktops in China where the primary data is held in London, so I was wondering if the community has any advice. I've been looking into a number of solutions, with and without Tangosol.

I've looked into the Push Replication Pattern and it seems like a sensible way forward - I imagine that this means that in my scenario I would publish from the London cluster to the China cluster and get the desktops in China to read from their local cluster.

Also, what transport is used for this replication? Is this just TCMP? I gather that JMS support is being added in a future version - what advantage would an external messaging protocol give?

In this sense I'm also wondering if there is any functional cross over with Extend. Having read through the blurb on Extend it seems as if it does pretty much the same thing - please correct me if I'm wrong. Unless of course the replication events to the remote Extend endpoints are performed synchronously - i.e. the local cache put will not return to the caller until the Extend clients have ack'ed the put. Or is it a case that the Extend proxy ack's synchronously and then brokers this out asynchronously?

Any pointers appreciated,

Ben
  • 1. Re: WAN replication
    Bob Hanckel Explorer
    Currently Being Moderated
    Hi,

    Good questions. The transport for Push Replication is vanilla TCP/IP. I

    The way it works is that the receiving cluster needs to declare a TCP Proxy Service
    with a local address and port.

    Push Replication services which are publishing to that cluster then have to
    declare a remote invocation service and a publishing service (one for every
    cache and destination cluster).

    Typically Push Replication batches up updates and then pushes it out the door
    to the destination by remotely invoking a "local" publisher. This means something
    that is local to the target and knows how to iterate through a batch of updates and
    put them in the correct target cache.

    Batch publishing has two settings which you will want to adjust depending on your load.
    One is batch size (i.e. the number of entries that are batched and published in one visit
    to the target cluster.) The other is the delay. How long does the Publishing Service sleep
    before it wakes up to see if more stuff is around to be published.

    Regards,

    Bob
  • 2. Re: WAN replication
    632230 Newbie
    Currently Being Moderated
    Hi Bob,

    Thanks for your response.

    >
    > Batch publishing has two settings which you will want to adjust depending on your load.
    One is batch size (i.e. the number of entries that are batched and published in one visit
    to the target cluster.) The other is the delay. How long does the Publishing Service sleep
    before it wakes up to see if more stuff is around to be published.
    Is the delay you are talking about on the sending side? If this is the case, doesn't this imply that the publisher is a polling thread - why would the publisher not be event driven off the backing cache? Or is it the case that you want to batch items to be friendlier to the network and were you to do this in an event driven style, you'd need some kind of timer thread to flush the buffer to the wire in case it took a long time to reach the high-water mark?

    Cheers,

    Ben

Legend

  • Correct Answers - 10 points
  • Helpful Answers - 5 points