Scale-Out HANA: Where to put tables?

[i033659]

Hi,

If I have a BW system on HANA with say three nodes, then there is a master node with the rowstore tables and the master column store. On the slave nodes we have the business data, i.e. the column store with infocubes, DSO, ...

Question: What if I have in my BW HANA DB a further schema with e.g. replicated tables from ERP used by HANA live?

a) On which node do those tables live?

b) Do I have to take care where to put those tables or does the system take the work?

c) Explain why ...

Thanks,

Ingo

[Bojan-lv-85 · Accepted solution]

Hello Ingo,

a) if no table partitioning is specified the tables remain on the master node.

b) actually, in first instance the application which is "bringing" the data to HANA usually should take care of table partitioning. I assume that e.g. during the BW migration to HANA the DDL statements are already containing the partitioning specifications for "large" tables.

In an SLT scenario this is not handled automatically and one has to take care of table redistribution manually e.g. in defining the partition specification in transaction IUUC_REPL_CONTENT.

c) on HANA, whether it is beneficial to have the data distributed evenly throughout all available nodes, depends highly on the application which will be accessing the data, e.g. in an BW envorinment if you load from one infoprovider to another keeping both on one node might increase performance.

On the other hand, if you are harvesting from gigantic datasets increasing parallelity by distributing table partitions to different nodes might assist. Such decisions mostly still need to be evaluated manually.

BR,

Bojan