Scaling Palm Vein Recognition from 100,000 to One Million Users: What Changes?
Your palm vein project is preparing for growth. More customers will register, more terminals will be deployed and more locations may use the service.
Does reaching one million registered users mean rebuilding the terminal application?
Within BioWavePass’s Large Model, an increase in user count alone does not require changing the default device SDK. The main changes concern server resources, licensed user capacity and QPS configuration.
The first step is to separate two planning questions: how many users the system must hold and how many requests it must process at peak times.
1. Registered Users and Request Rate Are Different Requirements
A database containing one million users does not necessarily receive more simultaneous requests than a database containing 100,000 users.
A smaller deployment across busy retail locations may generate concentrated demand. A larger membership platform may have users who authenticate only occasionally.
Both user population and request rate matter, but they describe different requirements.
| Planning factor | What it measures | What your team should establish |
|---|---|---|
| Registered users | The population the system must support | Expected launch population and future growth |
| QPS | Queries processed per second | Required peak identification request rate |
| Concurrent activity | Operations occurring at the same time | How requests overlap across terminals and locations |
| Response time | How long an operation takes | Acceptable waiting time for the application |
| Storage | Data retained by the deployment | Capacity needed for registration data and operational records |
For palm vein payment, estimate demand around busy checkout periods. Daily transaction totals alone do not describe the peak workload.
2. What Changes on the Terminal?
The default Large Scale device SDK does not need to change solely because the registered population grows from 100,000 to one million users.
The terminal application continues to participate in the supported registration and identification workflow.
However, this does not mean that every part of the application can be ignored during expansion.
Your team should validate:
- Connectivity from the intended locations.
- Response handling through your backend, if used.
- Application timeouts and error handling.
- Behaviour during service interruptions.
- The complete user journey under representative load.
A stable SDK interface reduces redevelopment work. Production testing still needs to reflect the expanded operating environment.
3. Review the Server Hardware
As the population and workload increase, review the server configuration against the requirements of the intended deployment.
The relevant resources include:
- CPU.
- GPU.
- Memory.
- Storage capacity and performance.
Use BioWavePass’s configuration recommendations for the target user population and QPS requirement. Do not assume that a server suitable for an initial pilot will support every future workload.
Size for the workload you expect
A useful infrastructure brief should include:
- Current registered users.
- Expected user population over the planning period.
- Number of active locations and terminals.
- Estimated peak request rate.
- Registration activity during normal operations.
- Response-time and availability requirements.
These details make server planning more useful than selecting hardware from user count alone.
4. Update Licence Capacity and QPS Configuration
Growing within the Large Model requires the appropriate authorisation for the intended user capacity and request-rate configuration.
These should be reviewed separately.
| Requirement | What to confirm |
|---|---|
| User capacity | The licence supports the intended registered population |
| QPS | The authorised configuration supports the required request rate |
| Infrastructure | The deployed hardware can support the planned workload |
| Licence validity | Purchased algorithm licences are lifetime-valid within their authorised scope |
A higher user-capacity licence should not be interpreted as unlimited throughput.
Similarly, adding server resources does not automatically change the authorised user capacity or QPS configuration.
Coordinate licensing and infrastructure planning before expanding the rollout.
5. Single Server or Kubernetes Cluster?
A cluster is not automatically necessary for every palm vein project.
For populations around 50,000 to 100,000 users, a single-server deployment is generally the recommended starting point.
For populations above one million users, or where scalability is a priority, a Kubernetes cluster should be evaluated.
The final choice should also consider availability requirements and the team’s ability to operate the environment.
A single-server deployment
A single-server approach can suit a project whose planned workload fits the recommended configuration and whose operational requirements do not call for a cluster.
It provides a more contained deployment for the customer’s infrastructure team to manage.
A cluster deployment
A cluster can provide a foundation for distributing services and expanding resources.
However, the actual benefits depend on the deployment design. A cluster label alone does not establish application availability, database resilience or recovery capability.
Confirm how the relevant services, storage and operational procedures will be configured.
6. Test the Complete Request Path
For a palm vein payment project, algorithm processing is only part of the customer’s waiting time.
The complete experience may also involve:
- Terminal capture.
- Network communication.
- Your backend.
- Algorithm processing.
- Payment authorisation.
- The application’s final confirmation.
If the terminal communicates through your backend, that backend must forward the required operations and return responses in the format expected by the SDK.
A successful algorithm query does not by itself demonstrate that the complete checkout meets your requirements.
Measure the full application journey alongside the algorithm-service workload.
7. Plan Registration Alongside Identification
Expansion often brings two kinds of activity at once: existing users identifying themselves and new users registering.
Include both in the workload discussion.
For example, launching additional locations may create a temporary increase in enrolment activity while established locations continue normal operations.
Your test plan should reflect the expected mix of activity rather than testing identification in isolation and assuming registration has no operational effect.
In the Large Model, your platform continues to supply the stable, unique user_id used during registration. Growth does not transfer ownership of that identifier to the SDK or Algorithm Server.
8. Define Operational Responsibilities Before Expansion
A larger deployment needs clear ownership of the production environment.
Agree who is responsible for:
- Monitoring service health and resource usage.
- Investigating failed or delayed requests.
- Managing user registration and deletion.
- Maintaining backups.
- Testing recovery procedures.
- Coordinating configuration and licence changes.
The customer owns and controls the deployed system and data. Access, retention and operational procedures should support the project’s compliance requirements and GDPR alignment.
BioWavePass does not require ongoing access to the server environment after deployment. Your team should therefore be prepared to manage routine operations and provide relevant information when requesting technical support.
9. Separate Capacity Expansion from Changing Algorithm Systems
Moving from 100,000 to one million users within the Large Model is capacity expansion inside that system.
It is different from beginning to use the Large Model after a Small Model project.
BioWavePass’s Small Model and Large Model are entirely separate, decoupled systems:
- The Small Model uses featureId to identify a palm.
- The Large Model uses a platform-supplied user_id to identify a user.
- Their SDKs should not be included together in the same application build.
Existing Small Model records do not automatically become Large Model registrations.
Where suitable images and corresponding user IDs have been retained, they can be submitted through individual Large Scale registration API calls. That separate registration process should not be confused with expanding an existing Large Model deployment.
A Practical Expansion Checklist
Before increasing the deployment’s capacity, confirm the following:
| Area | Readiness question |
|---|---|
| Population | Have we defined the required registered-user capacity? |
| Demand | Have we estimated peak requests separately from user count? |
| Hardware | Has the server configuration been reviewed for that workload? |
| Authorisation | Do licence capacity and QPS configuration match the plan? |
| Architecture | Does the deployment topology meet our operational requirements? |
| Application | Have we tested the complete request-and-response path? |
| Operations | Are monitoring, backup, recovery and support responsibilities assigned? |
Use these questions to prepare a concrete expansion plan before adding more locations.
Frequently Asked Questions
Do we need a new device SDK when growing from 100,000 to one million users?
An increase in user count alone does not require changing the default Large Scale device SDK.
Does a one-million-user licence provide unlimited QPS?
Do not assume that user capacity and request rate are the same entitlement. Confirm the authorised capacity and QPS configuration for your project.
Can we expand by upgrading hardware alone?
Hardware, licence capacity and QPS configuration must be considered together. Increasing one does not automatically update the others.
Is Kubernetes required for every deployment?
No. A single-server deployment is generally recommended around 50,000 to 100,000 users. Evaluate a cluster for populations above one million or where scalability requirements justify it.
Can we test the Large Model before purchasing its algorithm licence?
Yes. BioWavePass provides a Large Scale SDK Demo and public-cloud testing environment before algorithm purchase. Use that evaluation to understand the workflow, then validate production capacity against the intended infrastructure and workload.
Plan Your Next Stage with BioWavePass
Start with a realistic user forecast and a separate estimate of peak demand.
From there, review hardware, authorisation and deployment architecture together. This helps your team expand its palm vein recognition service without assuming that population growth requires unnecessary terminal redevelopment.
Partilhar este artigo
Sobre o autor
Também pode gostar
Small Model and Large Model Palm Vein Systems: What Developers Must Keep Separate
Can a Palm Vein SDK Capture Data Without an Algorithm Server?
Palm Vein Image Encryption: Why AES-256-GCM Alone Does Not Guarantee Compatibility
Large-Scale Palm Vein Recognition: How to Test the SDK and Plan for Growth
Palm Vein Payment SDK: Direct Algorithm Server Integration or Your Own Backend?
Palm Vein Deployment Support: Who Can Access Your Servers?
Scaling Palm Vein Payment: Single Server or Kubernetes Cluster?
Before You Scale Palm Vein POS: A Terminal Management Checklist
Why Saving Original Palm Images Matters for Biometric Recovery and Future Upgrades By BioWavePass