plugin-cfg.xml LoadBalanceWeight setting

[r***@barclays.com]

Hi,

I want to be able to HEAVILY weight in favour of one group of servers in a cluster vs the others in that cluster.

Does anyone know what the upper limit is (or if there is one) for setting the LoadBalanceWeight value to?

Thanks

Richard

[j***@efirstbank.com]

Have you considered using the Primary and Backup server designations?

PrimaryServers (zero or one element for each server cluster)
Specifies a list of servers to which the plug-in routes requests for this cluster. If a list of primary servers is not specified, the plug-in routes requests to servers defined for the server cluster.

BackupServers (zero or one element for each server cluster)
Specifies a list of servers to which requests should be sent to if all servers specified in the primary servers list are unavailable. The plug-in does not load balance across the backup servers, but traverses the list in order until no servers are left in the list or until a request is successfully sent and a response received from an application server.

http://publib.boulder.ibm.com/infocenter/wasinfo/v5r1//index.jsp?topic=/com.ibm.websphere.base.doc/info/aes/ae/rwsv_plugin.html

[r***@barclays.com]

Hi,
Yes, that was the original configuration we wanted to use but we had 5 primary JVMs and 5 backups, the primaries are load balanced but as soon as they are all unavailable only the first backup JVM received all traffic until it dies. I can't understand the logic of this (perhaps you know) it would seem logical that the backups should be able to round-robin as well.

As such we have decided to just heavily weight to what were the primaries and keep traffic to an absolute minimum on what were the backups until a failure of the primaries occurs.

R

[j***@efirstbank.com]

Here is the explanation from the site that I linked above:

+LoadBalanceWeight (zero or one attribute for each Server)+
+The weight associated with this server when the plug-in does weighted Round Robin load balancing. The algorithm for this attribute decrements all weights within the server cluster until all weights reach zero. Once a particular server's weight reaches zero, no more requests are routed to that server until all servers in the cluster have a weight of zero. After all servers reach zero, the weights for all servers in the cluster are reset and the algorithm starts over.+

+[Version 5.1 and later]When a server is shut down, it is recommended that you set the weight for that server to zero. The plug-in can then reset the weights of the servers that are still running, and maintain proper load balancing.+

So, for example, assume a cluster consisting of a, b, c, and d. If you only wanted traffic sent to a and b primarily you can do it in a n:1 ratio. E.G. for every 1 hit on c and d, a and b receive 100.

[r***@barclays.com]

Thanks for your reply.

I understand the principle here but let me try and explain our situation better.

We have NodeA with servers 01-05, we have NodeB with servers 06-10. We have decided we need 5 JVMs to support our user population so NodeA's JVMs are defined as Primaries, NodeB's JVMs as Backups. Running against the primary Node everything is fine as the traffic is balanced across all five JVMs, however if we lose NodeA the plugin fails over and uses just one of the backup JVMs until that individual JVM fails (which it will quickly because it will be taking the traffic of 5 JVMs).

As such we have thought about just heavily weighting to NodeA so the vast majority of traffic hits NodeA and a small proportion hits NodeB.

What I don't know if there is an upper limit to the LoadBalanceWeight (could we set it to 1000000??).

Also I don't really understand why it wouldn't be coded to round robin balance across the backup servers in a Primary/Backup configuration but that is an incidental question

[j***@efirstbank.com]

Given your explanation, I'm assuming that load balancing and high availability can not be abstracted to another device besides the plugin (Edge, BigIP, etc...). This would allow you to configure your "Backup Servers" as a second cluster and run high availability between the 2 clusters.

I'm not aware of any upper limit to the weight. IMHO, something like 1,000,000 might be a bit extreme. with that setting and given your number of servers 5 out of every 5,000,005 requests would be routed to your backups if all servers were up.