List:Cluster« Previous MessageNext Message »
From:Michael Dykman Date:July 14 2004 4:28pm
Subject:Re: node groups
View as plain text  
I am trying to relocate chapter and verse (the documentation is pretty 
scattered, I'm sure I don't need to tell you) but I just read within the 
last 2 days that:

	NoOfReplicas=<size of node group> and the groups will organize 
themselves according to the cardinality of the node id.

ie. where:
NoOfReplicas=4
DB nodeids= 4-11 (8 nodes)
group 1 will be 4-7
group 2 will be 8-11

  - michael dykman
  - michael@stripped

Joseph E. Sacco, Ph.D. wrote:
> Questions:
> 
> (0) What are the relationships between the number of replicas,the number
> of node groups, and the number of nodes per node group?
> 
> (1) Should there be a configuration tag that assigns a db node to a
> specific node group?
> 
> (2) Are node groups immutable?
> 
> =======================================================================
> It would appear that:
> 
> * (number of db nodes) % (number of replicas) = 0
> * (number of db nodes) % (number of node groups) = 0
> * in a given configuration all node groups must have the same number of
> db nodes:
> (number of db nodes per node group) = 
> 			(number of db nodes)/(number of replicas)
> 
> Are these relation sips correct? Are there any other constraints?
> 
> I have noticed that db nodes are assigned sequentially to node groups.
> For example, consider a cluster with two replicas and four db nodes with
> node Id's 2->5. The system will create two node groups and assign in
> sequence two nodes to each node group:
> 
> 	node group 1: DB2 + DB3
> 	node group 2: DB4 + DB5
> 
> It is important for an administrator to know that the assignment of db
> nodes to node groups is sequential, particularly if the administrator
> wishes to distribute each node group across multiple computers. In the
> current example, DB2 and DB4 should be configured to execute on one
> computer while DB3 and DB5 should be configured to execute on the other.
> 
> I guess the answer to my second question depends upon how many DB nodes
> and how many replicas are in the cluster. If the number of nodes is not
> huge, the cluster configuration task is manageable. Adding appropriate
> comments to the config.ini file should suffice.
> 
> My third question comes from thinking about cluster maintenance and
> cluster extensibility. I have conducted a few [albeit incomplete]
> experiments. It would appear at present that changing the cluster
> configuration requires rebuilding the cluster from scratch. 
> 
> I can easily envision situations where it becomes necessary to add more
> db nodes to an existing cluster. How will this be done?
> 
> 	dump -> reconfigure -> restore  ???
> 
> Whether or not this is a big issue, a small issue, or a non-issue comes
> down to the amount of time required [cost] to perform the task. The old
> mainframe data center operational model where there was a scheduled
> eight hour window of downtime each week for maintenance is rapidly
> becoming a thing of the past in the 7x24 e-world.
> 
> 
> Thoughts???
> 
> 
> -Joseph
>  

Thread
node groupsPh.D. Joseph E. Sacco9 Jul
  • Re: node groupsMichael Dykman14 Jul