Single service comparison

Service boundary differs sharply. ADF is an orchestration platform: pipelines compose activities, triggers, control flow, and Azure/self-hosted/SSIS Integration Runtimes that can invoke Databricks, Stored Procedures, HDInsight, or custom code. Huawei Cloud Data Migration (CDM) is a narrower, connector-driven batch data movement service. CDM exposes a wizard-defined job over roughly twenty source types (relational, file, NoSQL, OBS, MRS, DWS) and a distributed runtime that migrates tables, files, and directories; it has no equivalent to ADF's DAG, branching, or SSIS package runtime, so end-to-end pipeline logic must be re-homed elsewhere.

API and data model diverge. ADF exposes Azure Resource Manager resources (factories, pipelines, linked services, datasets, triggers) and a typed pipeline JSON with parameterization. CDM exposes RESTful APIs over jobs and clusters: you provision a CDM cluster, create a migration job bound to source and destination links, and run it; increment is expressed via WHERE clauses or time-variable functions rather than change-data-capture or trigger metadata. Expect to refactor pipeline definitions into CDM jobs plus an external scheduler (e.g., Data Ingestion Service, MRS workflow, or FunctionGraph) for dependency chains.

Scaling, HA, and billing differ. ADF scales via Integration Runtime node counts and bills by activity executions, DIU-hours, and self-hosted IR compute with managed VNet options. CDM scales by cluster specification and bills primarily per instance specification and runtime period, with a pay-for-what-you-use model and no per-activity unit. CDM is a managed Elastic cluster with built-in fault handling for dirty data archived to OBS, but cross-region HA, source-side CDC, and lineage/governance are not native CDM features and require adjacent services such as DRS for CDC, MGC for governance, and DLI/DWS for analytics integration.

Run assessment before selecting a target. Inventory ADF pipelines, linked services, Integration Runtime types, trigger cadence, and source/sink matrix. CDM is the closest equivalent for the data-movement subset (batch and WHERE-based incremental moves across ~20 source types), but it is not feature-parity for orchestration, SSIS, or managed CDC. For pipeline orchestration, plan a composed pattern: CDM for data movement plus an external scheduler (Data Ingestion Service, FunctionGraph, or MRS workflow) for dependency control, and DRS where real-time or CDC-style ingestion is required.

Re-platform, do not lift-and-shift. Convert each ADF pipeline into discrete CDM jobs: recreate linked services as CDM source/destination links, replace COPY activities with CDM table/file/directory migration jobs, and translate incremental WHERE clauses using CDM's WHERE and time-variable functions rather than ADF tuming-window triggers. Self-hosted IR responsibilities move to a CDM cluster plus VPC/VPN/Direct Connect to the on-premises source; plan the network path, security group, and key (KMS) equivalents explicitly since CDM does not host a self-hosted IR analog.

Validate and execute a controlled cutover. Build staged migrations: full load, then increment via WHERE, then a final delta and cutover with rollback criteria. Validate row counts, checksums, dirty-data archives in OBS, and end-to-end orchestration timing against acceptance criteria. CDM does not provide ADF's pipeline monitoring lineage natively, so attach Cloud Eye, LTS, and the scheduler's audit log to reconstruct pipeline run history and alerting before production cutover.

Recompute TCO and mind the gaps. ADF bills activity executions, DIU-hours, and IR compute plus source/target infrastructure; CDM bills per cluster specification and runtime, with additional cost for DRS, DLI, or DWS where the composed pattern needs them. Recalculate for peak load, volume, retention, and cross-region/interconnect traffic, and budget for orchestration, CDC, and governance services separately since they are not included in CDM metering. Confirm regional availability of CDM and dependent services in the target region before commitment.