BW tables distribution in HANA SP6 and later

[Former Member]

There are so few information about very various theme as HANA table placement and table distribution.

In SPS 06 and later

http://scn.sap.com/community/hana-in-memory/blog/2013/09/03/sap-hana-landscape-redistribution-with-s...

answer from SAP Support:

By default when BW creates a DSO table it sends an estimated row count  of 1 Mio. records.

As according to the table placement rules a DSO  tableis created with 3 parts if it exceeds 40 Mio. records (which is not

fullfilled) the table is created with one part.

Sap notes: 1908073  1908075   1908082   1908133

My question to SAP - does it' mean that we have always all tables on one node and other will be idle.

Answer from SAP  -  Different DSO tables will usually be placed  on different slave nodes (depending on the hash value of the group

name).

---==---

So Be aware  that for default in your Scale-out BW on HANA systems DSOs tables will be created on one worker node and without any partitions.

[Former Member]

The reason for this is that there aren't so many HANA Scale-Out installations out there (in the hundreds), and most of those fall into one of two categories:

- Small enough that table placement issues weren't a problem

- Big enough that an expert advised on table placement

I'm not sure specifically what you want to know though - it's not clear from your question. But...

- Small tables will always sit on one node (on a round robin basis, so all nodes will be used)

- Bigger tables will be partitioned automatically and will distribute across a number of nodes

- The row store objects e.g. Basis tables always sit on the master node (row objects on slave nodes perform badly)

- The column store objects (DSO, Cube) always sit on slave nodes which is why you need at least 3 active nodes for BW

As you can imagine, the above strategy can cause problems with very large tables, in which case you may need expert advice.

John

[Former Member]

Hello John,

Currently, I am researching the case BWonHANA extension from One Node to Scale-out.

As you said, basis tables sit on the master node and the DSO and Cube sit on slave nodes. I would like to know how I could move DSO, Cube data from Master to Slave in my case.

The Table Redistribution will do it or shall I move them manually?

Regards, John Zhang

[Former Member]

You should do a table redistribution.

[Former Member]

+

"However, as a general principle, the system never creates more

partitions than the number of available hosts. For

example, if a distributed SAP HANA database only has five hosts, the

repartitioning described above, from three

partitions to six partitions, would not take place. The partitioning

rules also apply only to tables for which Hash or

RoundRobin partitioning is specified at level 1. Tables without a

partitioning specification are also not

automatically partitioned when the threshold values are exceeded."

If you want to have more partitions than number of hosts than you shouldsemantic partitioning. There is a good blog regarding this topic: http://www.saphana.com/community/blogs/blog/2013/04/15/bw-on-hana-and-very-large-tables