Re: Setting up partitioning (5.0)

[Joseph Kouyoumjian]

Jonathan:

I also have been following this thread and I am also struggling a little. Your answers got me almost all the way there. Based on your example, would the following be correct?

6 computers
3 replicas
Each computer running 1 data node

Node configuration: 2 node groups

Node group 1:
db1
part 1
part 3
part 5

db2
part 3
part 5
part 1

db3
part 5
part 1
part 3

Node group 2:
db4
part 2
part 4
part 6

db5
part 4
part 6
part 2

db6
part 6
part 2
part 4

1) You could lose up to two machines from each node group and still run.
2) Each machine services the first partition listed, maintaining a copy of the other two in case it has to take over them as well.

Is this correct?

What is a little unclear is the relationship betwen the number of replicas, the number of nodes and the number of node groups.

The example from the manual uses two replicas with two nodes (on two computers). This results in a single node group with two nodes. Each node must have a complete copy of the database, correct?

If I had a single replica with two nodes, I would have two node groups with each node having 1/2 of the database, but no redundancy, correct?

What would happen if I had 6 nodes and two replicas? Would I end up with 3 node groups of two nodes each? How would the data be partitioned in this case?

Would it be somehing like the following?

Node group 1:
db1
part 1
part 3

db2
part 3
part 1

Node group 2:
db3
part 5
part 2

db4
part 2
part 5

Node group 3:
db5
part 4
part 6

db6
part 6
part 4

Is this correct?

Doing something like this would reduce the amount of memory needed per node to store a database of a given size, correct (memory per node = size of db * # of reps * 1.1 / # of nodes)?


Thanks,

Joe



Edited 2 time(s). Last edit at 02/18/2006 04:52PM by Joseph Kouyoumjian.