I am looking at Teradata Manager after having added two new nodes and I am seeing a lot of virtual utilization and physical utilization skew. I suspect it is because of the additional resources in the new nodes, but it does raise some questions: 1. Are the new nodes working at the speed of the old ones due to the fact that they need to wait? 2. Is there anything that can or should be done to balance the system better?
New nodes mean, more vprocs. If you have some tables which have a very low degree of cardinality on a primary index value, this condition could arise.
Think about it like this: if you have 8 amps per node, and then double your system size, you will go from 16 to 32 amps.
Now let's imagine that you have a very large table on your system with only 20 unique values in the primary index. Once the data gets redistributed into your new configuration, it could very well increase the degree of physical skew (and therefore query skew) for the objects themselves, or queries referencing those objects.
What I am noticing in Teradata manager is a change in the relationship between the max, avg and min resouce utilization on the node and amps. The three lines would graph very closely before the merge. Now, the high is much higher than the minimum. I believe I know why this is; evidently, the processing power of the new nodes is not spread out acrosss enough amps to bring the per amp processing capability down to an equal level across nodes.
so, is there anything that can be done to more evenly distribute the increased resources on the new nodes to more amps so that each one is more compatible with the amps on the old system