Thank you for the post Dieter.
1. Each AMP will hold the secondary index values for their rows in the base table only for NUSI. How does the hash value point to the same AMP for the rows that are in same AMP on which NUSI is defined!! (Hash alg in NUSI is independent of NUSI data where as it is dependent on data on USI as it may be mapped to another AMP)
2. "As a first step you would need to do a broadcast to all AMPs in the system because the exact AMP is unknown and this is much more overhead"
What does broadcast to all AMP's in system mean? I hope that, it has already been done via PI (it has nothing to do with either USI or NUSI) as which rows are to be distributed to which AMP is decided by PI.
As the rows are already distributed based on PI, now what all it requires is just to hash the column on which USI is defined and store the row-id and other info. in the sub-table of the same AMP (same as NUSI) which is faster than hashing the USI column and map to another AMP and store it that sub-table.
↧