Concept card

Replication Flows

As of 2026-07-24T14:00:00Z

Edited by Cédric-Olivier Mâry, sole editor and publication director — 27 years in the SAP ecosystem. How these pages are produced, and where automation is used.

What is Replication Flows?

Without native CDC at the source, Replication Flow falls back to full-load and the batch-window math turns ugly fast — a 500GB fact table eats 1.7 hours of source time every single day.

Replication Flow is SAP Datasphere's native, managed pipeline for moving data from a source system into the platform's storage layer. Think of it as the tenant-internal alternative to the two heavyweight SAP integration tools consultants have used for two decades — System Landscape Transformation (SLT) and Smart Data Integration (SDI) — purpose-built for the analytics use case rather than full-blown enterprise integration. You configure a Replication Flow once, per source connection and per set of source objects, and Datasphere handles scheduling, monitoring, restart-on-failure, and, where the source supports it, incremental change capture, without you writing or maintaining custom extraction code.

What it does and why it matters

A Replication Flow moves rows from a source object — a table, a CDS view, a database — into a target table inside Datasphere, following one of three modes. Initial load is a one-time bulk copy, used to seed a target before incremental replication starts. Full-load periodic re-reads the entire source object on a schedule — simple, but expensive at volume. Delta, the production-grade mode, uses change-data-capture to move only what changed since the last run. The mode you choose is not a stylistic preference; it is the single decision that determines whether your pipeline scales to enterprise data volumes or grinds the source system to a halt on day one.

Why it matters

Choosing full-load over CDC-driven delta on a large fact table can stall a pipeline at the first 50GB table instead of scaling to enterprise volume.
Sustained delta throughput (50-200k rows/sec) versus full-load (1-5GB/min) is the gap between a design that scales and one that doesn't.
Tables over 100GB need partition-by-partition initial load — single-bulk loads bottleneck on source CPU.

Key points

Three modes: initial-load (one-shot), full-load periodic, delta (CDC). CDC is the production-grade default.
Throughput: 50-200k rows/sec delta, 1-5 GB/min full. Source CPU is the bottleneck, not Datasphere.
S/4 has native CDC; ECC needs SLT or ODP-CDS; cloud DBs use their native streams.
Initial-load > 100 GB: partition by fiscal year + LoB, load 4-16 partitions in parallel.
Target storage: HANA local for hot (SAC > 10×/day); managed Delta (BDC) for cold/historical at 10× compression.
Schema drift = silent risk. Wire source release pipeline to Datasphere REST API introspection refresh.
CU cost: 0.5-2 CU per active pipe baseline; consolidate small dim pipes to control floor cost.
Replaces SLT for most analytics use-cases; SLT remains for legacy + complex transformations + > 200k rows/sec.
Replication Flows is mastered only when it changes a named buyer decision.
Start with the semantic contract and control model before demonstrating the tool.

Terms used on this page

Replication Flow: Datasphere's native managed pipeline for moving data from source to target. CDC-aware, throughput 50-200k rows/sec.
Delta load: Incremental load mode — only changed rows since last run. Requires CDC at source.
Initial load: One-time bulk copy of source state into Datasphere. Foundation for delta to take over.
Full-load periodic: Re-reads entire source on schedule. Fallback when CDC isn't available; doesn't scale beyond 50 GB tables.
CDC (Change Data Capture): Source-side mechanism that emits change events (insert/update/delete). Required for delta mode.
ODP-CDS: ECC extraction protocol — Operational Data Provisioning over CDS views. The canonical CDC path from ECC.
Schema drift: Source-side schema change (column added, type changed, field renamed) that the replication metadata cache hasn't yet absorbed.
Pipe parallelism: Number of concurrent worker threads per Replication Flow. Default 4, max OOTB 8, raisable via support.

Sources

Full card available to members. What the full card adds: the full decision framework · the SAP vs Snowflake / Databricks / Fabric comparison · the common pitfalls and their fix · the cheat sheet · the architecture schemas · the code blocks · the facts worth quoting.

Open in the app →