1. Why Table Buffering Matters Even on SAP HANA

A common misconception in SAP S/4HANA projects is that because SAP HANA is an in-memory database, application server table buffering is obsolete. This assumption is dangerously false:

• Network Roundtrip Latency: Even with high-speed 10Gbps interconnects, an SQL call from the ABAP application server to HANA takes 1 to 5 milliseconds of network and protocol serialization overhead. A local table buffer hit takes ~0.02 milliseconds.
• HANA CPU & Thread Starvation: Offloading high-frequency configuration reads (e.g. currency conversion rates, tax codes, posting periods) preserves HANA database threads for mission-critical analytical and transactional writes.

2. Buffering Types & Architectural Decision Matrix

Choose the buffering strategy based on data volume and access pattern:

1. Single-Record Buffering (#SINGLE):
   • Best for large master data tables (e.g. > 10,000 rows) where programs access isolated records using the complete primary key.
   • Each record is loaded individually upon request.
2. Generic Buffering (#GENERIC):
   • Best for tables organized by organizational units (e.g. Client + Company Code, or Client + Country).
   • A query matching the leading generic keys loads the entire sub-segment into memory.
3. Full Buffering (#FULL):
   • Best for small lookup tables (< 1,000 records) accessed by multiple non-key combinations.
   • The entire table is loaded on the first SELECT.

3. Cross-Instance Invalidation via DDLOG

In a scaled-out SAP landscape with multiple application server instances (PAS and AAS), modifying a buffered table requires cache synchronization:

• The modifying instance invalidates its local buffer immediately and writes a timestamped record to database table DDLOG.
• Other application server instances periodically poll DDLOG (governed by profile parameter rdisp/bufreftime, typically every 1 to 2 minutes) and purge invalid entries.

4. SQL Constructs That Bypass the Buffer

Developers must structure ABAP SQL carefully to avoid unintentional buffer bypasses:

" ❌ Bypasses buffer: Aggregate function
SELECT COUNT(*) FROM zclean_config INTO @DATA(lv_cnt).

" ❌ Bypasses buffer: Incomplete primary key on single-record table
SELECT SINGLE config_value FROM zclean_config WHERE company_code = '1010' INTO @DATA(val).

" ✅ Hits buffer: Full primary key provided
SELECT SINGLE config_value FROM zclean_config 
  WHERE config_id = 'TAX_RATE' AND company_code = '1010' 
  INTO @DATA(val).