Post-Quantum Cryptography in ADSS Server ‑ Performance Benchmarks

What's the performance impact of adopting Post-Quantum Cryptography inside ADSS Server?

This blog presents controlled software-based benchmarking results, comparing classic algorithms (RSA and ECDSA) with ML-DSA within ADSS Server 8.4.

The objective is not marketing optimisation — it is architectural transparency.

In the previous articles in this series, I examined:

• Structural integration of ML-DSA within X.509 and CMS
• Interoperability validation using OpenSSL and Bouncy Castle

Now, let's look at performance benchmarks.

1. Benchmarking Scope

The following operations were measured:

All above tests were conducted using ADSS Server 8.4.0 in software‑only mode (no HSM acceleration).

2. Test Environment

• ADSS Server 8.4
• Microsoft Windows Server 2022 Standard (10.0.20348 N/A Build 20348) – RAM = 16 GB
• JVM default settings
• 1 KB payload for signing tests
• Detached CMS signing
• OCSP Response signing
• Timestamp Response signing
• Warm JVM (steady‑state performance measured)

Each test was executed in batches of 10,000 operations, and average performance was calculated.

3. Digital Signature – Signing Performance

3.1 Detached CMS Signature Performance Overview

3.2 OCSP Response Performance Overview

3.3 Timestamp Response Performance Overview

Observations

• ML‑DSA signing is slower than ECDSA‑P256 but comparable to, or moderately slower than, RSA‑2048 in many cases.
• The performance increase scales with security level (2 → 3 → 5).
• Signature size increases significantly compared to classical algorithms.

From an architectural standpoint: signing performance remains practical for server‑side use, but throughput planning must account for increased computational cost.

4. Signature Verification Performance

Verification performance seems more practical as we done some raw signature testing in POC. Verification cost is critical for:

• OCSP responders
• High‑volume validation services
• Document validation platforms

4.1 Signature Verification Performance (POC)

Observations

• ML‑DSA verification is generally faster relative to its signing cost.
• In many cases, ML‑DSA verification is closer to RSA‑2048 than to ECDSA‑P256.
• For high‑volume validation workloads, ML‑DSA remains operationally viable.

Importantly:

Verification scaling remains linear and predictable under load.

5. Artefact Size Impact

Beyond CPU cost, PQC affects:

• Certificate size
• CRL size
• CMS signature size
• OCSP response size

5.1 Certificate Size Comparison

Observations

• Certificate size increases significantly.
• CRLs grow proportionally due to larger signature fields.
• CMS signatures may increase by multiple kilobytes.

Architectural implications:

• Increased storage requirements
• Increased bandwidth usage
• Potential impact on high‑frequency OCSP environments

This is often a larger operational concern than CPU cost.

6. Throughput Considerations

Under sustained load:

• ML‑DSA signing throughput is lower than ECDSA.
• CPU utilisation increases proportionally.
• Horizontal scaling becomes more relevant for high‑volume signing services.

However:

ADSS Server's stateless signing architecture allows:

• Load‑balanced scaling
• Policy‑based algorithm selection
• Hybrid classical + PQC deployment strategies

This mitigates performance concerns at scale.

7. Practical Deployment Guidance

Based on testing:

For High‑Volume Signing:

• Consider ML‑DSA‑2 where regulatory constraints allow.
• Scale horizontally rather than vertically.

For Root / CA Certificates:

• ML‑DSA‑5 provides maximum security margin.
• Performance impact is negligible due to low issuance frequency.

For Validation Services:

• Provision additional CPU capacity compared to ECDSA‑only environments.
• Monitor CRL and OCSP response sizes in bandwidth‑sensitive deployments.

8. Architectural Conclusion

The performance impact of PQC adoption in ADSS Server can be summarised as:

• CPU cost increases moderately
• Artefact size increases significantly
• Structural semantics remain unchanged
• Operational viability remains intact