Data replication deployments sizing and capacity planning guide
This article provides baseline resource and process sizing guidelines for OCI GoldenGate Data Replication deployments. You and your implementation teams can use this as a structured starting point for selecting deployment capacity and beginning configuration and integration testing.
This article covers:
- OCI GoldenGate managed Data Replication deployments on OCI, Azure, and Google Cloud.
- Supported Oracle Database and non-Oracle relational sources and targets, using the deployment technologies and builds available in the selected cloud and region.
- Extract, Distribution Path, Receiver Path, and Replicat capacity within each managed deployment.
- Trail retention, recovery backlog, network demand, and production-representative sizing validation.
This article doesn’t cover:
- Self-managed GoldenGate installations on physical servers, virtual machines, or Kubernetes.
- Data Transforms, Data Verification, and Stream Analytics deployments, which require separate capacity models.
- Database hardware sizing and detailed design of complex federated or active-active architectures.
For hybrid topologies, size each component according to its operating model. A customer-managed source-side Extract still requires host and database capacity even when it sends changes to OCI GoldenGate.
Note: This document provides baseline sizing guidance but does not guarantee performance, scalability, or capacity. Actual resource consumption varies with transaction volume, transaction size, transformations, topology, latency requirements, and target performance. Complete a proof of concept using production-representative workloads before deploying to production.
Provide reasonable headroom during initial testing. Undersized deployments can experience resource pressure, trail growth, replication lag, or process instability. Monitor the limiting stage before increasing resources or parallelism.
Basis for these recommendations
The sizing recommendations in this document are not based on a single benchmark or synthetic workload. GoldenGate processes a wide range of transaction profiles, data models, and replication topologies, so no single throughput figure can represent all customer environments.
Instead, the guidance in this document reflects a combination of:
- Oracle Product Development performance testing and scalability validation activities
- Oracle Support and performance engineering investigations
- Observed deployment patterns from customer implementations
- Oracle consulting and field engineering experience
- Operational best practices from production GoldenGate environments
These recommendations provide a practical starting point for infrastructure provisioning and deployment planning. Actual resource requirements depend on workload characteristics including transaction volume, transaction size, redo generation rates, latency requirements, transformation complexity, network conditions, and target-system performance.
Customers should treat these recommendations as baseline guidance and validate them through proof-of-concept testing and operational monitoring before production deployment.
Workload classification tiers
Before selecting a sizing profile, classify the workload using the table below. These labels describe planning complexity; they do not map redo rates or table counts to a guaranteed OCPU requirement. When demand varies, test the busiest sustained period and the recovery workload.
| Attribute | Small | Medium | Large | X-Large |
|---|---|---|---|---|
| Target Workload Profile | Dev / QA or low volume | Department OLTP | Enterprise OLTP / DW | High volume or complex topology |
| Avg Redo per Hour (GB) | < 1 GB | 1-100 GB | 100-300 GB | > 300 GB |
| Replication Scope | Limited source and target scope | Several workload partitions | Multiple sources or targets | High concurrency or extensive fan-out |
| Sizing Emphasis | Functional validation | Sustained demand | Peak and recovery capacity | Partitioning and failure boundaries |
Note: The size of the database, TPS (Transactions Per Second), and table count alone are not reliable predictors of resource consumption. Collect source redo/log bytes, transaction characteristics, captured trail bytes, and wire traffic separately. For Oracle sources, validate redo measurements with the DBA and avoid counting duplicate archived-log copies or RAC threads more than once.
Initial sizing workflow
-
Measure the source workload.
- Collect the redo generation rate.
- Understand transaction characteristics.
- Identify the source and target count.
-
Select a workload profile.
- Use the Workload Classification Tiers table to identify the required test cases.
-
Create data replication resources.
- Select base OCPUs and autoscaling. Confirm the cloud, region, build, and storage limit.
-
Configure GoldenGate.
- Configure supported Extract, Distribution Path, and Replicat processes.
-
Establish a monitoring baseline.
- Capture consumed OCPUs, memory, process lag, and trail growth from day one.
-
Validate sizing.
- Run normal, sustained peak, initial-load, and outage-recovery workloads.
-
Adjust resources as needed.
- Change the limiting resource or supported process setting, and repeat the test.
Compute and storage sizing
OCI GoldenGate allocates resources to managed deployments. The following table summarizes example base configurations and the documented allocation model. Development and production values are initial planning guidance. The larger values are configuration examples rather than workload tier recommendations.
Sizing philosophy:
- Start with the service baseline and concurrent process requirements, then validate actual demand.
- Provide headroom for peaks, large transactions, maintenance, and catch-up while source changes continue.
- Size each deployment independently and retain separate source database and target database capacity plans.
| Resource | Development / Test | Production Start | Larger Example | Maximum Base |
|---|---|---|---|---|
| Base OCPUs | 1 | 4 | 8 | 24 |
| Base Memory Allocation | 16 GB | 64 GB | 128 GB | 384 GB |
| Base Network Allocation | 1 Gbps | 4 Gbps | 8 Gbps | 24 Gbps |
| Storage (Trails, other files) | 500 GB | 2 TB | 4 TB | 12 TB |
OCPU guidance
- Oracle recommends starting with 1 OCPU for development/test and 4 OCPUs for production. Enable autoscaling if needed to handle peak workloads. You can also scale deployments up or down manually.
- Autoscaling can use up to three times base OCPUs, capped at 24 OCPUs. Test sustained demand and recovery at the selected configuration.
- Start with 1 OCPU for each concurrent Extract, Replicat, or Distribution Path as process planning guidance. Adjust the OCPU count during testing.
- Extract CPU demand varies with captured changes, transaction characteristics, filtering, transformations, and source mining.
- Oracle also recommends allocating 1 OCPU for the user interface.
Memory guidance
- The service allocation is 16 GB of memory per OCPU. Validate memory pressure together with consumed compute and temporary-space demand.
- Memory requirements depend on uncommitted transactional data, long-running transactions, LOBs, concurrent processes, and spill activity. Provide headroom until representative testing establishes demand.
- By default, each Big Data handler typically uses 8 GB of memory. Ensure
jvm.bootoptionsheap settings (-Xmxand-Xms) are sized appropriately to preventOutOfMemoryerrors under peak load. CACHEMGRis internally self-configuring and self-adjusting. Retain the defaults and consult Oracle Support before changing cache-size controls.- Integrated Extract uses the source database logmining server and streams pool. Increasing service OCPUs does not increase source SGA capacity.
- Database Parallel Replicat uses Mappers, a Master, and Appliers. Assess process and target database memory rather than applying a generic JVM-heap-per-thread estimate.
Storage and I/O guidance
- Storage and filesystem implementation are service managed. Customers plan retention and recovery capacity rather than selecting disk types, mount points, or IOPS.
- Count every physical trail copy within each deployment and reserve space for reports, discard files, temporary spill, operational files, and backup-related requirements where applicable.
- Note that initial load trails tend to consume more storage space than regular trails.
Trail retention policy
Base trail retention on business recovery objectives, operational procedures, and the maximum expected outage duration of downstream systems. The required window must also cover restart, catch-up, operator response, and any additional replay requirement.
Example starting retention targets:
| Deployment Type | Recommended Trail Retention |
|---|---|
| Development / QA | 4-8 hours |
| Production (single target) | 24 hours |
| Production (critical workloads) | 48-72 hours |
| Cross-region / WAN deployments | 72+ hours |
- Keep checkpoint protection enabled for live trails and validate all Extract, Distribution, Receiver, and Replicat consumers.
- A purge task deletes eligible files.
- Confirm source log retention as well as trail retention so capture can restart at the required position.
These are example starting targets. The selected window must cover the actual recovery procedure and fit within the confirmed deployment storage limit.
Trail storage calculation
When sizing trail storage, account for:
- Peak Trail Generation Rate x Retention Period x Safety Factor
- Oracle recommends applying a minimum safety factor of at least 1.5x to accommodate workload spikes, maintenance windows, and unexpected target outages.
- Do not purge trail files based solely on age. Purge files only after all downstream Replicat processes have successfully processed the required trail data and their checkpoint positions have advanced beyond the files you plan to remove.
- For business-critical environments, retention should be sufficient to allow recovery from planned maintenance activities, target database outages, network disruptions, and troubleshooting investigations without requiring Extract repositioning or re-instantiation of target systems.
Use this formula:
Required Trail Storage = Peak Trail GB/hour x Retention Hours x Safety Factor
Example: 100 GB/hour x 24 hours x 1.5 = 3,600 GB (3.6 TB)
The safety factor is a planning baseline, not a managed-service allocation rule. Use measured trail-generation bytes for each local trail copy, add the non-trail reserve, and include the complete outage, restart, catch-up, and response interval without counting the same data twice.
Note: The example illustrates the calculation, and it is not a single-deployment capacity recommendation. Validate retention and workload partitioning against the confirmed capacity of each deployment.
GoldenGate process count recommendations
The the following table provides recommended process counts per OCI GoldenGate Data Replication deployment for each workload tier. These counts assume a single-source, multi-target deployment pattern. For multi-source configurations, multiply the Extract count by the number of source databases and adjust the OCPU count accordingly.
| GG Process | Small | Medium | Large | X-Large |
|---|---|---|---|---|
| Extract (Integrated) | 1 | 1-2 | 1-2 | 2-4 |
| Distribution Paths | 1-2 | 2-4 | 4-8 | 8-16 |
| Replicat (Classic) | 1-2 | 2-5 | 5-15 | 15-30 |
| Replicat (Parallel) | 1 | 1-2 | 2-4 | 4-8 |
| Replicat (Coordinated) | 1-2 | 2-5 | 5-10 | 10–0 |
| Max Concurrent Processes | 2-4 | 8-14 | 16-34 | 36-70 |
For multiple sources, account for each supported capture scope and adjust compute accordingly. Apply the console and concurrent-process allowance in OCPU Guidance, then validate memory, target capacity, and peak demand. Partition the workload across deployments when one deployment cannot meet the tested requirement.
Extract process guidelines
- Start with one Integrated Extract per supported source capture scope. For CDB/PDB sources, use the capture configuration supported by the selected deployment build and database.
- Introduce additional Extracts only for a valid workload partition or a measured capture bottleneck. Test mining overhead, source database resources, transaction consistency, and recovery.
- A single Extract can write to multiple trail files. This allows a single Extract to feed multiple targets or split a large workload into multiple trails that multiple Replicats can process.
- Confirm platform restrictions before using multiple trails; for example, MySQL Extract with GTID set positioning supports only one trail.
Distribution Service guidelines
- A single Distribution Service can sustain multiple concurrent paths. Configure a path for each required trail-transfer route and account for fan-out traffic and each downstream consumer.
- For WAN distribution, review the supported path buffer and TCP flush settings to compensate for network latency. Change settings only after route testing demonstrates the constraint.
Replicat process guidelines
- Select the right Replicat for your workload. For example, see Types of Replicat for Oracle databases.
- Oracle recommends Parallel Non-Integrated Replicat for Oracle databases and Parallel Replicat for all other GoldenGate platforms that support it.
- Classic Replicat applies serially and suits development, light-workload DR scenarios, and non-Oracle targets that lack parallel apply support.
- A coordinator manages a Coordinated Replicat’s multiple apply threads, making this Replicat suitable for structured OLTP workloads that require dependency tracking.
- Assign one Replicat group per target schema or logical workload partition to maximize parallelism and isolate failure domains.
OCPU consideration
- Parallel Replicat generally requires additional OCPU resources as parallelism increases.
- Integrated Replicat performs parallel apply inside the target Oracle Database, so include target database resources in capacity planning.
Disaster recovery topology guidance
OCI GoldenGate operates the service infrastructure. Select a managed recovery topology according to business recovery objectives and feature availability in the chosen cloud and region. Database, network, replication, and application recovery remain part of the complete design.
Single deployment
- Suitable as a starting deployment for development, QA, pilots, and workloads with recovery procedures that meet the agreed objectives.
- Configure supported process auto-start and restart controls and test their behavior.
- Use service deployment backups and checkpoint-protected trail retention. Include source and target database recovery in the runbook.
Primary and peer deployments
- You can create local or cross-region peers to increase availability and minimize downtime during planned maintenance and unplanned outages.
- A primary deployment can have one peer deployment. The standby always remains the same size as the primary.
- Run peer prechecks and recovery exercises. RPO (Recovery Point Objective) is zero. RTO (Recovery Time Objective) depends on the amount of data stored on the deployment. The higher the storage utilization, the longer the RTO.
- For cross-region recovery, validate Distribution Path endpoints, truststore certificates, secrets replication, connection routing, and process startup. See Manage peer deployments for more information.
Deployment backups and maintenance
- The service retains daily automatic backups for 12 days and manual backups until you delete them. The backup content can include metadata only or metadata and trails. See Manage deployment backups for more information.
- A deployment backup does not replace source and target database backups or required live trail retention. Test the restore sequence at a consistent recovery point.
- Plan the stop/restart and backlog associated with deployment upgrades. Review known issues and the deployment-specific maintenance information.
Monitoring and metrics baseline
Establish a monitoring baseline from day one. Track service consumption, process performance, database resources, and the complete source-to-target route continuously for production deployments.
OCI Monitoring metrics
- Extracts: capture lag and process state.
- Distribution and Receiver Paths: lag and status.
- Replicats: apply lag and process state.
- Deployments: OCPU utilization, OCPU consumption, memory and storage utilization.
- Application: heartbeat lag.
You can review the full list of available metrics in Data Replication Metrics.
Configure alarms in OCI Monitoring for process failure, freshness breaches, stalled progress, and forecast storage exhaustion. Thresholds should allow time for operator response. The OCI Notifications service delivers these alarms.
Initial sizing validation criteria
After deployment and representative workload testing, validate sizing against the application objectives. Define numerical lag, resource, and catch-up thresholds for each workload instead of treating them as universal service limits.
A properly sized deployment should exhibit:
- End-to-end freshness and per-process lag within the agreed objective during normal and sustained peak demand.
- No persistent trail backlog growth and sufficient retained space throughout the planned outage and recovery.
- Memory and temporary-space behavior that remains stable through large transactions and concurrent activity.
- No unexplained data errors, recurring process restarts, or resource-related abends.
- Backlog recovery within the agreed deadline while new source changes continue.
- Successful restart, backup/restore, and peer recovery tests where applicable.
Include initial-load overlap, backups, maintenance, and forecast growth in the test. Record achieved rates, worst-case lag, retained storage, recovery time, consistency results, and the conditions requiring a new sizing exercise.
Pre-provisioning checklist
Complete the following checklist before provisioning the initial OCI GoldenGate deployment. Gathering this information in advance reduces deployment delays and post-deployment resizing.
- Collect peak and average redo log generation rate from the source DB (per hour, per day).
- Identify replicated tables, row sizes, LOBs, largest transactions, filtering, and transformations.
- Determine source and target counts, deployment placement, and LAN / WAN / cross-cloud routes.
- Confirm supported endpoint combinations and deployment technology.
- Confirm the latency SLA (data freshness requirement) with the application owner.
- Select base OCPUs and autoscaling; record console and concurrent process allowances.
- Size each retained trail copy, recovery backlog, non-trail reserve, and checkpoint-protected purge policy.
- Validate measured wire throughput and route latency during normal, peak, and recovery demand.
- Measure supplemental logging impact and validate source mining and target apply capacity.
- Define source/target privileges, credentials, access policies, DNS, routing, and required listener ports.
- Choose a recovery topology and confirm peer, backup, and build availability in the selected cloud.
- Plan the initial load and a consistent synchronization boundary. Retain logs/trails through catch-up.
- Establish monitoring, alarms, operator ownership, and representative test acceptance criteria.
Disclaimer
Oracle based these recommendations on common deployment patterns and internal testing. Actual resource requirements may vary depending on workload characteristics, including transaction complexity, DML/DDL mix, network latency, target database performance, and supplemental logging configuration.
These recommendations reflect common deployment characteristics across Oracle GoldenGate customer environments and Oracle internal engineering experience. They intend to provide a practical starting point for capacity planning rather than a substitute for workload-specific performance validation.
Oracle Corporation and its affiliates make no warranties, express or implied, regarding the accuracy, completeness, or suitability of these recommendations for a specific environment. Customers are responsible for validating all recommendations and sizing guidance against their own production workloads prior to go-live.