are there two levels, L1 and L2 of delta stores

[Former Member]

dear all

i have been reading up on the delta merge process, and come across a document that may explain the reason why i though delta storage was row based, in the face of many recent documents that say the delta store is a column based.

it states that there are actually 2 levels of delta store.

L1 - row based, uncompressed

L2 - column based, dictionary compressed

can anyone categorically confirm this, and maybe point to further information.

thanks

[lbreddemann · Accepted solution]

This notion of what L1 and L2 delta are is correct on a conceptual level.

The L1 delta concept however has not been implemented in SAP HANA so far.

[Former Member · Accepted solution]

hi lars

thanks for that

is the L2 store, dictionary compressed, with unsorted indexes.

or am i talking gibberish

and what about chandan's response below ?

ta

[lbreddemann · Accepted solution]

Sorry, but the details on the delta store implementation are not externally documented, so I cannot go into more detail here.

As mentioned before, the understanding of the L1 delta concept is correct - but it's not active in any productive SAP HANA code.

[Former Member · Accepted solution]

hi lars, thanks for that

we restarted a failed instance a short time ago, and noticed, in a log, that a "fusion delta merge" was occurring upon restart.

can you shed any light on that ?

[lbreddemann · Accepted solution]

never saw this message - no idea.

[lbreddemann · Accepted solution]

Ah well... looked it up and this seems rather new (SPS11).

WIth the continuous log replay feature there is also the ability to perform delta merges during log recovery (makes a lot of sens in replication scenarios). Upon crash recovery this is also done and works primarily on the delta store as far as I understand. The old delta gets cleaned out of outdated row versions and "fused' into a new delta.

[chandan_praharaj]

Yes you are correct, there are 2 level of delta, in which HANA internal architecture designed, and this is one of the best design for handling Write Optimised and Read Optimised scenario in a Single Data base. This is the speciality of HANA.

When ever Data update/Insert happens, it happens as below step.

1. Data written to L1 Delta(Row Based), so writing is faster, This data is copied to REDO Log(for any issue)

2. System Writes(Incremental Merge) L1 Delta data to L2 Delta(Column Based).

3. L2 Delta data is Persisted in HDD and In Parallel Merge with the main store(Column Based) (RAM/Primary Mem).

4. Main Memory Data is also periodically comitted to HDD.

This Process is called Delta merge technique.

You can find multiple document for this in Google.

This one is the best described with case.