Ecoservity · Independent Benchmark · January 2024
ETL extracts 212 raw database tables and calls it a business object. STREAM reads the business object directly. That architectural difference (not cloud choice) explains the entire cost gap. Verified on AWS S3 and Azure ADF with identical results.
AWS Glue, AWS DMS, FiveTran, and any other ETL pipeline share the same architectural constraint: they extract database tables. SAP does not organize business data in tables: it organizes it in business objects (TCODEs, reports, transactions) that each touch dozens or hundreds of underlying tables. The STREAM Engine reads at the object level, not the table level. This single difference eliminates the Build phase entirely.
ETL tools access SAP at the database layer. MB51 (Material Document List) is one SAP transaction, but it reads 212 underlying database tables. Every ETL tool must extract all of them, then engineers reconstruct the business logic in AWS Glue, Databricks, or Redshift. That reconstruction is Phase 3 (Build), costing $974,368 for 5 TCODEs.
USB4SAP STREAM Engine connects at the SAP application layer, not the database layer. It reads the same output that SAP already produces for MB51, complete with business column names, filter logic, and computed fields, and streams it directly to AWS S3 via a standard REST call. No tables. No reconstruction. No Build phase.
Same customer, same project scope as the Azure ADF benchmark. The AWS comparison was run independently to verify the result is not cloud-specific. It is not.
The benchmark was run independently on both Microsoft Azure Data Factory and AWS S3. The cost result is $1,149,630 versus $112,093 (90% reduction) on both clouds. The time result is 21 weeks versus 10 weeks (59% reduction) on both clouds. The gap is architectural, not cloud-specific. See the ADF + USB4SAP benchmark for the Azure comparison.
Every project phase on AWS alone versus AWS with the STREAM Engine plug-in. The Build phase tells the entire story.
| Phase | AWS ETL alone | AWS + STREAM Engine | $ saved | % saved |
|---|---|---|---|---|
| 1 · Plan | $1,920 | $640 | $1,280 | 67% |
| 2 · Setup | $1,664 | $576 | $1,088 | 65% |
| 3 · Build | $974,368 | $704 | $973,664 | 100% |
| 4 · Go-live | $6,566 | $2,560 | $4,006 | 61% |
| 5 · Operations | $165,111 | $107,613 | $57,498 | 35% |
| Total | $1,149,630 | $112,093 | $1,037,537 | 90% |
| Phase | AWS ETL alone | AWS + STREAM Engine | Wks saved | % |
|---|---|---|---|---|
| 1 · Plan | 4 wks | 2 wks | 2 wks | 50% |
| 2 · Setup | 6 wks | 2 wks | 4 wks | 80% |
| 3 · Build | 7 wks | 2 wks | 5 wks | 73% |
| 4 · Go-live | 4 wks | 4 wks | 0 | 0% |
| Total | 21 wks | 10 wks | 11 wks | 59% |
USB4SAP STREAM Engine plugs into your existing AWS infrastructure. No replacement of AWS Glue or your current pipeline. No on-premise agent to manage. SAP connectivity goes through STREAM Engine; everything downstream (API Gateway, S3, Lambda, Redshift, Bedrock) stays exactly as it is.
Scheduled or on-demand extraction of full tables and field-level CDC to S3 for data lake, analytics, and AI workloads.
On-demand REST calls for live SAP data, routed through Amazon API Gateway into downstream AI and automation services.
MCP Catalog layer enables AI agents to discover and call SAP data on-demand without pre-staging. Supports RAG architectures, agentic workflows, and real-time operational AI on live SAP data.
The STREAM Engine's AWS security model combines SAP's own authorization layer with AWS pre-signed URL. No credentials stored outside either system, no custom security layer to build or maintain.
Every STREAM Engine request executes under the requesting user's SAP authorization profile. If a user cannot see a plant's material movements in SAP, they cannot see them through STREAM Engine either. No custom row-level security to build in AWS. No shadow copy of your SAP data without the security model attached.
STREAM Engine uses AWS pre-signed URLs to write to S3. A pre-signed URL grants time-limited, operation-scoped access to a specific S3 location. No AWS access key or secret stored in the STREAM Engine configuration. The URL expires after use. No persistent credentials. No blast radius if a URL is intercepted.
Traditional ETL tools require AWS IAM credentials with broad S3 write permissions, stored persistently in the ETL platform configuration. STREAM Engine replaces this with pre-signed URLs scoped to the exact operation. Audit trail stays intact in both SAP and AWS CloudTrail simultaneously.
No custom VPC endpoint rules for SAP connectivity. No IAM role for the ETL platform with broad data access. No data masking layer in Glue because sensitive SAP fields were dumped into S3 without column-level filtering. STREAM Engine applies SAP's own column-level visibility before data leaves SAP.
The cost gap is a capability gap. These are the six things STREAM Engine delivers on AWS that any ETL pipeline architecturally cannot.
TCODEs, ABAP reports, ALV output, BW extractors, CDS views all stream to S3 as complete business objects. Not 212 raw tables that must be joined: the final output, ready for Athena, Redshift, or Bedrock to query directly.
Change Data Capture at field level, not table level. Only changed fields stream to S3, not a full table reload on every run. No change pointers activation in SAP required. Works on SAP ECC, S/4HANA, BW, and CRM on HANA or non-HANA databases.
Standard REST API with no SAP SDK. AWS Lambda functions, Step Functions workflows, and Amazon Bedrock agents can query live SAP data on-demand, not from a staging copy in S3 that may be hours old.
STREAM Engine exposes SAP objects through MCP (Model Context Protocol), the standard that AI agents use to discover and call tools. Amazon Bedrock agents can query SAP inventory, orders, and financials without any custom connector development.
The same STREAM Engine REST API handles non-SAP sources alongside SAP: Oracle Hyperion, IoT data streams, unstructured data. One API for mixed enterprise data landscapes, not separate ETL pipelines per source.
At 5 SAP objects, STREAM Engine costs $112K versus $1.15M for ETL. At 400 objects, ETL exceeds $20M. STREAM Engine scales near-flat because the REST call is the same for any object. No per-object custom code that compounds the Build phase.
The $1.1M saving above is for 5 TCODEs and 100 tables. Every additional TCODE or BW report multiplies the ETL Build cost while STREAM Engine stays near-flat. Talk to us about your specific SAP object footprint and AWS target architecture.