Four GoldenGate product lines, one per situation. Which line covers a non Oracle target, a Kafka stream or a mainframe source, and where the processor count actually lands.
Oracle GoldenGate licenses per processor on every source and target it touches. Replicate into a non Oracle database or a streaming platform and the GoldenGate line follows the data, whatever the target costs.
Which line it follows is the question this page answers. There are four separate GoldenGate products on Oracle's price list, they cover different ends of a flow, and buying the wrong one is the most common finding we write up.
The canonical metric, the hub arithmetic and the Foundation Suite math live in our Oracle GoldenGate licensing guide. This page is the heterogeneous case: non Oracle targets, streaming and lake targets, and the tool swap that does not remove the license.
Oracle GoldenGate is licensed per processor on every system where it captures or delivers data. Both the source database server and the target server need licenses, even when the target is not an Oracle product.
That symmetry is the whole commercial story. A replication estate is priced by the number of ends, and nothing in the metric responds to how much data actually moves.
GoldenGate counts processors using the Oracle Processor Core Factor Table. Multiply physical cores by the core factor for the chip to get licensable processors, and apply the same rule to non Oracle target hardware.
Read the factor off the current table for the exact chip family. In our reviews the wrong row was used often enough to move the count by 15 to 35 percent, in both directions.
On a managed cloud database you cannot install anything on the server, so the delivery process has to run somewhere else. That somewhere else is the licensed footprint, and you choose its size.
This is the one place in GoldenGate licensing where the buyer has direct control over the count. A delivery host with 8 vCPUs is a smaller license line than one with 32, and the difference is a launch time decision nobody revisits.
Size the delivery host to the throughput the flow actually needs, then pin it. If the platform team later resizes it for headroom, the license line moves with it and no invoice tells you.
Four line items cover four different situations, and the choice is made by what sits at each end of the flow rather than by what your team calls the project.
GoldenGate product lines by path, list per processor
| Path | Product line | List per processor | What it does not cover |
|---|---|---|---|
| Oracle to Oracle | Oracle GoldenGate | $17,500 | The database license at either end |
| Oracle at one end, non Oracle at the other | Both lines, one per end | $17,500 each | Nothing is bundled. Two products, two counts |
| Non Oracle to non Oracle | GoldenGate for Non Oracle Database | $17,500 | Any Oracle database end, plus Active Data Guard and XStream rights |
| Kafka, object store, lake or warehouse target | GoldenGate for Distributed Applications and Analytics | $20,000 | The relational ends of the same flow |
| Mainframe source, DB2 for z/OS and similar | GoldenGate for Mainframe | $100,000 | Open systems ends, which stay on their own line |
List prices from the Oracle Technology Global Price List, before discount. Support runs at 22 percent of net.
If an Oracle database sits at either end of the flow, that end is licensed on the Oracle GoldenGate line, and every non Oracle end is licensed on its own line. There is no combined product and no second leg discount.
The GoldenGate licensing information manual is where the entitlement text lives. Read the section for the exact line you are buying, because the prerequisites differ between them.
Both base lines include a restricted use license to GoldenGate for Distributed Applications and Analytics, and the restriction is narrow enough that most teams breach it without noticing.
Read those two conditions carefully. The restricted rights are only useful if you already own the Oracle product at the far end, which most estates streaming into Kafka do not.
GoldenGate for Mainframe lists at $100,000 per processor, which is 5.7 times the open systems line. It covers DB2 for z/OS and the other mainframe sources Oracle supports.
Check the core factor row for the processor family before you model it. The 0.5 multiplier that halves x86 counts does not apply across the board, and on a mainframe family the count can land far closer to raw cores than a team used to Xeon pricing expects.
Yes. Replicating into PostgreSQL, SQL Server, MySQL or any supported non Oracle database requires GoldenGate licensing on that target server. The target being free software does not make the GoldenGate path free.
This is where the audit finding usually lands. The Oracle source was licensed properly at project time, and the targets multiplied afterwards as teams asked for their own copy of the data.
GoldenGate supports a long list of non Oracle databases, documented on the Oracle GoldenGate product page. Each supported target type still draws a license, and the supported version matrix moves between releases.
Confirm the version matrix before you design the path. A target on an unsupported version is both a technical risk and an argument you will lose in a support case.
Most heterogeneous GoldenGate purchases start as a migration. Move Oracle to PostgreSQL, run both in parallel for a cutover window, decommission the source.
Then the flow stays, because someone wants fallback, and a temporary tool becomes a permanent line with 22 percent support attached. In our review file that pattern was still running two years after the cutover in several estates.
The buyer move is to price the exit at the start. Ask for term licensing scoped to the migration window, agree in writing what happens at the end of it, and put a date in the project plan for switching the flow off.
Streaming and analytic targets sit on GoldenGate for Distributed Applications and Analytics, the line Oracle previously sold as GoldenGate for Big Data. It lists at $20,000 per processor and it is a different product from the relational lines, not an add on to them.
Licensing a stream target under the classic product is the second most common error we correct. It looks like the same software, the handlers ship in the same distribution, and nothing at deployment time asks which line you bought.
The relational ends of the same flow stay on their own lines. An Oracle source feeding a Kafka topic is two products: the Oracle GoldenGate line on the source, and the Distributed Applications and Analytics line on the delivery side.
Streaming targets are almost never servers you can install on. The delivery process runs on a host you provision, and that host is what carries the processor count.
Two flows delivering into the same Kafka cluster from one delivery host is one licensed footprint. The same two flows split across two hosts for operational tidiness is two. Consolidate the delivery tier before you buy.
Cloud warehouse and object store targets follow the same logic as Kafka. There is no server of theirs to license, so the licensed footprint is your delivery tier, and its size is your decision.
That makes the analytics case unusually controllable. A tightly sized delivery tier feeding a warehouse can be a modest GoldenGate line, while the same flows spread across a fleet of oversized workers can cost more than the warehouse.
Not on the Oracle side of the flow. If the source is an Oracle database, the alternative tool still has to read Oracle change data, and the mechanism it uses is the one Oracle licenses on.
ENABLE_GOLDENGATE_REPLICATION is the switch. Oracle's database reference states that when it is true you need a valid GoldenGate license, and that the requirement extends to third party and end user tools using the feature.
So a team that replaces GoldenGate with a competing product against an Oracle source, and sets that parameter to make it work, has swapped the vendor without swapping the license claim. Our GoldenGate licensing guide works through how Oracle finds it.
A genuinely clean swap avoids the Oracle end of the flow entirely, or reads the source through a mechanism the parameter does not govern. That is a technical determination, and it belongs in writing before the migration business case is signed.
Where the source is not Oracle, none of this applies and the alternative market is wide open. The trap is specific to Oracle sources, and it is the reason so many heterogeneous cost cases fail to deliver.
For many non Oracle and streaming paths, native replication, open source change data capture and cloud provider services replace GoldenGate at a fraction of the cost. Fit depends on latency, throughput, transformation needs and who carries support.
Log based open source tooling handles a large share of relational to relational and relational to streaming paths. The license cost approaches zero and the operational cost does not, so budget the engineering ownership honestly.
Where it fits, the saving is not marginal. A single 16 processor GoldenGate target line at list is $280,000, which buys a lot of engineering time.
Make the trade explicit rather than arguing the tool. "This path is 16 processors, which is $280,000 at list and $61,600 a year in support after a mid band discount. What does the alternative cost us in engineering time over three years?"
Sometimes GoldenGate still wins, and that is a fine answer once it is a priced decision. The failure mode is choosing it by default and discovering the line at renewal.
The common assumption is that GoldenGate licensing follows the Oracle database, so a team that owns Oracle Database licenses believes its replication is covered. We disagree. In the estates Fredrik Filipsson reviewed, the unlicensed exposure sat on the non Oracle targets and the streaming delivery side, not the Oracle source. The buyer side move is to map every source and target pair, license GoldenGate where it actually runs, and test whether a native or open source tool can carry the heterogeneous paths. Treating GoldenGate as an Oracle only line item is how the audit finding gets written.
GoldenGate is licensed where it runs, not where you bought it. The non Oracle target is still an Oracle license.
Source: Redress Compliance advisory engagement file, 2024 to 2025.
Yes. GoldenGate is licensed per processor on every server where it captures or delivers data. Both ends need licenses, including non Oracle targets, and the count is taken on each end's own cores.
Yes. Replicating into PostgreSQL or any supported non Oracle database requires GoldenGate licensing on that target. The target being open source does not remove the requirement, and the non Oracle line lists at the same $17,500 per processor as the Oracle line.
Oracle GoldenGate for Non Oracle Database. It covers supported non Oracle open systems databases at both ends and does not cover an Oracle database at either end, so a mixed flow needs both product lines.
GoldenGate for Distributed Applications and Analytics, which lists at $20,000 per processor. It delivers change data into streaming and analytic targets such as Kafka, object stores and cloud warehouses, and it is a separate line from the relational products.
Only a narrow restricted use. The base lines allow JMS traffic solely into separately licensed Oracle Coherence Grid Edition and Kafka traffic solely into separately licensed Oracle Stream Analytics. Any other stream or lake target needs the full Distributed Applications and Analytics line.
On the host where the delivery process runs, because you cannot install on a managed service. Size that host to the throughput the flow needs and pin it, since a later resize moves the license line without any warning.
By multiplying physical cores by the Oracle core factor for the chip family. The same calculation applies to non Oracle target hardware, and on an authorized cloud the vCPU rule replaces the core factor.
Unlicensed targets. Estates license the Oracle source properly and then run live delivery into non Oracle or streaming targets without matching licenses, which surfaces as a true up at renewal or audit.
For many heterogeneous and streaming paths, yes. Open source change data capture, native replication and cloud services cover common cases. Against an Oracle source, check whether the tool needs the replication parameter, because that keeps the Oracle claim alive.
It lists at $100,000 per processor, which is 5.7 times the open systems line, and covers DB2 for z/OS and other supported mainframe sources. Check the core factor row for the processor family, because mainframe families do not carry the x86 multiplier.
No. Redress Compliance is 100 percent buyer side. We do not resell or implement Oracle software. We review the GoldenGate footprint and licensing position for the customer.
Per processor GoldenGate math, the options it drags in, and where replication licensing quietly doubles.
Used across more than five hundred enterprise clients. Independent. Buyer side. Built for procurement leaders running the next renewal cycle.
