SAP HANA Infocube and DSO Memory Management

[carlosbasto]

Hii Experts,

I have been reading many articles and papers of SAP BW on HANA Memory Management, but I did not find any solid or clear content about a doubt I have. Therefore, I hope you all can help me to understand it:

We have a non-active data concept, after SP6, for management of SAP BW objects staging in SAP HANA Main Memory; however I didn't see any possibility of choosing which SAP BW objects I can put in memory or not.

I do know that SAP HANA, by default, understands Hot and Warm data and set them accordingly (after SP6). However, I think it would be nice if I had the possibility of setting an infocube, for exemple, to be out of memory (only in disk).

In RSHDBMON, I can set "unload from the memory", but I do not think it's a permanently setting.

Let’s go to the questions:

1) In RSHDBMON, can I set an infocube to be out of memory (in disk only) permanently?

2) Infocubes in HANA have 4 partitions. Is any of them non-active data? In addition, if it is, can I set it manually?

3) After I unload data from PSA, only new partitions can be created. Does it mean that I would not access these data anymore?

I hope I had been clear.

Thank you all in advance.

Regards.

Carlos Basto

[vivekbhoj · Accepted solution]

Hi Carlos,

Have you read all documents given in below link:

http://www.saphana.com/docs/DOC-3075

1.) If you permanently want to keep an Infocube out of memory, then you can unload its fact table from memory but if that Infocube is accessed again, say for reporting then the columns of the fact table that have been queried will be loaded into memory.

2.) Partitions don't have anything to do with non active data. Four partitions are for compressed requests, non-compressesd requests, reference points of inventory data and historic movement of inventory data

Regards,

Vivek

[carlosbasto · Accepted solution]

Hi, Vivek.

Thank you for you fast reply to my question.

Yes, I've read all of these documents.

1) You mean that, in RSHDBMON I can set an Infocube to be out of memory and this setting will work permanently while data from this Infocube is not accessed, right? Ok, I got it.

2) About partitioning, I understand that partitions is used to stage compressed data for performance staging reasons, but let's suppose I have an Infocube with data from the last 3 years and all my queries access only the last 2 years. Data that is not being accessed will not be considered as non-active data?

with best regards,

Carlos Basto

[vivekbhoj · Accepted solution]

Hi Carlos,

1.) Yeah using RSHDBMON, if you unload Infocube from memory, it will be permanently out of memory unless you start doing some operations with it - like loading data to Infocube or using it for query etc.

Check below link for more details on this transaction:

http://help.sap.com/saphelp_nw73/helpdata/en/bb/fd2d8a0bbe4714aae378c4ae561652/frameset.htm

Alternatively, you can manually unload Infocube fact table from HANA Studio also like

UNLOAD /BIC/F/AA; -

http://help.sap.com/hana/html/sql_unload.html

2.) Data that is not being accessed will not be considered as non-active data? No

To know more Check SAP Note 1767880 - Non-active data concept for BW on SAP HANA DB

Also read:

https://help.sap.com/saphelp_nw73ehp1/helpdata/en/1a/c43532bb124b2fb62c3e5ad8b1b442/frameset.htm

Regards,

Vivek

[carlosbasto · Accepted solution]

Hi, vivek.

Thank you for your helpful answer.

About my question 3, don't you have any knowledge about it?

Regards.

Carlos Basto

[vivekbhoj · Accepted solution]

Hi Carlos,

After I unload data from PSA, only new partitions can be created. Does it mean that I would not access these data anymore? - I didn't get it clearly.

PSA is partitioned on basis of request. Older request and partitions are unloaded from memory but if you want to access older data, you can access it and can also load PSA table into memory.

Regards,

Vivek

[carlosbasto · Accepted solution]

Hi Vivek,

Probably I didn't understand a part of the sap note you've sent:

"Persistent Staging Areas and write-optimized DataStore objects are also

partitioned by request:

Partitions that have once been displaced are no longer loaded because new data is loaded only to the newest partition, and older data is normally no longer accessed.

However, if old data should be accessed, this data is loaded to the main memory. Typically, it is load processes that are used for setting up new target objects, or data has to be reloaded. For such processes, it is acceptable that the data must first be loaded to the main memory."

Could you please explain the sentence above in more details?

Regards,

Carlos Basto

[vivekbhoj · Accepted solution]

Hi Carlos,

As per my understanding,

It says that PSA and WODSO tables will be partitioned based on request. This will help in minimizing Delta merge time. 

For every new request that is being loaded to PSA or WODSO, a new partition will be created and older partitions will be unloaded from memory.

Thus saving memory as generally we don't need older data from PSA.

But if older data from older requests is needed to be accessed again then the partitions will be again loaded into the memory

Regards,

Vivek

[carlosbasto · Accepted solution]

I got it now, Vivik.

Thank you a lot for sharing your knowledge.

Regards.

Carlos Basto

[vivekbhoj · Accepted solution]

You are welcome Carlos

Regards,

Vivek