High latency when cluster with hazelcast plugin

Hi there,

We have successfully setup openfire 3.8.2 cluster with hazelcast plugin on two VMs, with the same database server, and did a load testing like below:

  1. user 1 (on node 1) send messages to 50 users (on node2) in a loop. at the same time, user 2 (on node 2) send messages to 50 users (on node1) in a loop,

  2. wait for 1s, repeat step 1 and step 2.

  3. send message about 2m, then wait for about 3m for receiving all message.

it works great with one single openfire server, all message could be received just after sent, but wiht hazelcase plugin, we only could receive little message afther the 3m, the rest are saved as offline messages.

it takes us lots of time but couldn’t find a clue, tried to increase the hazelcase query thread and event thread but didn’t getter better, only one thing might help is that with the monitoring plugin, when all connected to single server, seems the active conversation is created just after the user connected, but with cluster, the conversations are created after a while, any thoughts about this?

Thanks

Seems it’s caused by the VMs network settings, when we used real PC everything works fine. thanks guys, great work!

Hi,

Im new to clustering so will you help me on how to do it?

I have two openfire server and one mysql server for database

I enabled cluster plugin and its shows as below

Cluster Overview

Below is an overview of your cluster. You have 2 node(s) running and you are licensed to 10,000 node(s) in this cluster. To see more information, click each node. The row in yellow indicates the local node.

sv2.openfiresv.com
e3205cea-eeed-41ff-ab82-97f287b77391
Aug 27, 2013 2:30:25 PM
N/A
N/A
N/A

N/A

sv1.openfiresv.com
a50bf05e-a21f-4638-86d9-95b24e412878
Aug 27, 2013 12:10:48 PM
0 (0%)
0 (0%)
0 (0%)


25.08 MB of 454.38 MB used

can you tell me how it works when a node goes down?

client side which ip address we need to give?

Hi Shamjith,

You need a TCP loadbalancer in front of these two openfire node, like LVS, haproxy, or F5, and you client only need the IP of this loadbalancer, by the correct configuration, the loadbalancer will route request to a live server when another is down, please note under this situation, the user who is on the dead server before should relogin on the live server.

Hope this will help.

1 Like