HSM / PKCS#11 / HSM Capacity Calculator
What you will do: Explore fictional per-site workloads, assumed operation rates, utilisation headroom and local redundancy. Inspect actual site allocations and separate payment-fleet results.
Worked example: Enter ten general-purpose devices and three sites: the allocations are 4/3/3, not twelve devices. Compare each site against the model requirement at 70% target utilisation, then apply N+1.
Runtime and privacy: This educational calculator performs arithmetic in your browser; it does not run cryptographic benchmarks or measure hardware. Use synthetic planning inputs.
For your role
- Executive / Business Leader
- Explore which workload and resilience assumptions drive an illustrative device count. Fictional profiles A and B are sensitivity examples, not a procurement forecast, vendor comparison or hardware-replacement ratio.
- GRC / Risk & Compliance
- Export the model version, operation recipes, rate assumptions and site allocations as an assumption record. Separate payment scope and unsupported capabilities remain visible; the result is not measured evidence, certification or production readiness.
- Security Architect
- Explore per-site workload recipes, TLS signing and hybrid key-exchange offload, and independent payment capacity. Compare assumed shared device time at 70% utilisation before N+1 or 2N; these fictional profiles do not establish vendor capability or whole-site failover.
- IT Ops / DevOps
- Enter per-site transaction rates and total inventory. Check each site allocation, including uneven splits, and keep utilisation headroom separate from local spares. Export the assumptions, then validate real workloads and failure scenarios on the target system before making an operational sizing decision.
HSM workload and capacity explorer
Educational model with fictional performance profiles. Demand is per site; all sites receive equal traffic. Sum assumed device-time shares, divide by the utilization target, round up, then apply local N+1 or 2N. Independent operation rates do not validate concurrent hardware performance. Local redundancy does not model whole-site failure or traffic redistribution. Displayed utilisation averages demand across all allocated devices, not only active devices in a cold-standby design. TLS signing and key exchange are HSM-offloaded; one server signature and one classical exchange including ephemeral key generation, plus one server ML-KEM encapsulation for hybrid exchange. Certificate issuance and peer verification are separate workloads. No queueing, latency, session, partition or product-limit prediction. Payment and general-purpose fleets are independent. User-entered rates remain assumptions, not verified benchmarks.
PQC TLS example: entirely PQC certificate chain, hybrid classical + ML-KEM exchange. This is a hypothetical scenario, not a product deployment recommendation.
Utilisation target leaves 30% nominal headroom before local spares. The 70% default is a teaching choice, not a universal threshold. Changing the preset replaces all per-site transaction rates, including payments.
General-purpose workloads
6,645 transactions/s per site; 6,645 transactions/s across 1 equally loaded sites. Transactions can require several cryptographic operations.
Full server handshakes with signing and key exchange inside the HSM. One server signature, one classical exchange, plus one ML-KEM encapsulation in the PQC example. The classical exchange rate includes ephemeral key generation. PQC certificates use an entirely PQC chain; the exchange remains hybrid. CA issuance and peer verification are separate.
Operation recipe assumptions
Classical recipe per transaction: 1 × ECDSA P-256 + 1 × ECDH P-256. Hypothetical PQC recipe: 1 × ML-DSA-65 + 1 × ML-KEM-768 + 1 × ECDH P-256. Multiply each recipe by per-site TPS.
Fictional per-site workload. Organization presets illustrate sensitivity; replace the transaction rate to explore your own assumptions.
One artifact signature per transaction. ML-DSA or SLH-DSA is a hypothetical algorithm choice, not CNSA or product-compliance advice. SP 800-208 specifies LMS/XMSS families, not SLH-DSA.
Operation recipe assumptions
Classical: one RSA-2048 signature per artifact. PQC: one ML-DSA-65 signature per artifact. Multiply by the selected per-site signing rate.
Fictional per-site workload. Organization presets illustrate sensitivity; replace the transaction rate to explore your own assumptions.
One document signature per transaction. This model makes no qualified-signature or eIDAS status claim.
Operation recipe assumptions
Classical recipe per transaction: 1 × RSA-2048. Hypothetical PQC recipe: 1 × ML-DSA-65. Multiply each recipe by per-site TPS.
Fictional per-site workload. Organization presets illustrate sensitivity; replace the transaction rate to explore your own assumptions.
Illustrative external key-service requests: one symmetric operation and 0.1 asymmetric key-delivery operations per request. This is not a claim that database connections directly cause TDE key operations.
Operation recipe assumptions
Classical recipe per transaction: 1 × AES-256 + 0.1 × RSA-2048. Hypothetical PQC recipe: 1 × AES-256 + 0.1 × ML-KEM-768. Multiply each recipe by per-site TPS.
Fictional per-site workload. Organization presets illustrate sensitivity; replace the transaction rate to explore your own assumptions.
One new CA or responder signature per transaction. Use actual signing events, not certificate validations: cached OCSP responses do not require a fresh signature for every request.
Operation recipe assumptions
Classical recipe per transaction: 1 × RSA-2048. Hypothetical PQC recipe: 1 × ML-DSA-65. Multiply each recipe by per-site TPS.
Fictional per-site workload. Organization presets illustrate sensitivity; replace the transaction rate to explore your own assumptions.
Illustrative recipe: one symmetric wrap/unwrap plus 0.2 key-delivery exchanges per request. ML-KEM needs an appropriate derivation and wrapping construction; it is not a direct replacement for an encryption API.
Operation recipe assumptions
Classical recipe per transaction: 1 × AES-256 + 0.2 × ECDH P-256. Hypothetical PQC recipe: 1 × AES-256 + 0.2 × ML-KEM-768. Multiply each recipe by per-site TPS.
Fictional per-site workload. Organization presets illustrate sensitivity; replace the transaction rate to explore your own assumptions.
Fictional shared fleet operating both tunnel endpoints: two AUTH signatures and an assumed exchange service per tunnel. Remote-access clients may use a different fleet. This simplified recipe excludes peer verification and protocol processing costs.
Operation recipe assumptions
Classical recipe per transaction: 1 × ECDH P-256 + 2 × ECDSA P-256. Hypothetical PQC recipe: 1 × ML-KEM-768 + 1 × ECDH P-256 + 2 × ML-DSA-65. Multiply each recipe by per-site TPS.
Fictional per-site workload. Organization presets illustrate sensitivity; replace the transaction rate to explore your own assumptions.
Fictional HSM-resident host signing and hybrid exchange. The classical curve rate is a modelling proxy, not a measured X25519 rate or an SSH configuration recipe.
Operation recipe assumptions
Classical recipe per transaction: 1 × ECDSA P-256 + 1 × ECDH P-256. Hypothetical PQC recipe: 1 × ML-DSA-65 + 1 × ML-KEM-768 + 1 × ECDH P-256. Multiply each recipe by per-site TPS.
Fictional per-site workload. Organization presets illustrate sensitivity; replace the transaction rate to explore your own assumptions.
One RRset signature per transaction. Input is a signing rate, not DNS request traffic. PQC signatures here are a hypothetical end state, not a deployable DNSSEC recommendation.
Operation recipe assumptions
Classical recipe per transaction: 1 × ECDSA P-256. Hypothetical PQC recipe: 1 × ML-DSA-65. Multiply each recipe by per-site TPS.
Fictional per-site workload. Organization presets illustrate sensitivity; replace the transaction rate to explore your own assumptions.
Educational scenario comparison
Results depend on the selected fictional rates and recipes across 1 sites. They do not establish production readiness or a hardware replacement ratio.
Classical · profile A: 3 modelled devices required; 10 allocated. All sites meet the selected model.
PQC · profile A: 71 modelled devices required; 10 allocated. At least one site falls short of the selected model.
PQC · profile B: 3 modelled devices required; 2 allocated. At least one site falls short of the selected model.
| Scenario | Site | Required incl. headroom | Required incl. redundancy | Allocated | Model result | Assumed utilisation |
|---|---|---|---|---|---|---|
| Classical · profile A | 1 | 2 | 3 | 10 | Meets assumptions | 8.92% |
| PQC · profile A | 1 | 70 | 71 | 10 | Shortfall | 485.97% |
| PQC · profile B | 1 | 2 | 3 | 2 | Shortfall | 60.61% |
How TPS becomes HSM count
- Multiply per-site transactions by each recipe's operation counts.
- Sum operation demand ÷ assumed rate across algorithms. This assumes shared device time.
- Divide that sum by the target utilisation, then round up.
- Apply local redundancy: N+1 adds one device to a nonzero requirement; 2N doubles it. The selected setting is N+1.
- Assess each site's actual allocation. Aggregate demand is per-site traffic × sites. Equal traffic and local spares do not establish whole-site failover capacity.
PQC · profile A: assumed device time 48.6 ÷ 0.7, rounded up gives R = 70.
N+1: 71 per site; 2N: 140 per site.
Total fleet = 1 × 71 = 71 HSMs
The total above uses the selected local redundancy. Headroom is separate from spare devices; no spare is required for zero demand.
Fictional general-purpose performance profiles
Fictional profile A
Fictional inputs, not vendor measurements. Signature rates mean signing; ECDH means an exchange including ephemeral key generation; ML-KEM means server encapsulation for TLS and an assumed effective exchange rate for other recipes; AES means the selected recipe operation, not bulk bytes/s.
Zero means unsupported. Edited values are user assumptions.
Fictional profile B
Fictional inputs, not vendor measurements. Signature rates mean signing; ECDH means an exchange including ephemeral key generation; ML-KEM means server encapsulation for TLS and an assumed effective exchange rate for other recipes; AES means the selected recipe operation, not bulk bytes/s.
Zero means unsupported. Edited values are user assumptions.
Separate payment-HSM fleet
Separate fictional payment fleet. Two symmetric operations per transaction and an illustrative 0.05 key-management operations per transaction. The ratio is not a PCI/EMV requirement; RSA-to-KEM mapping is a hypothetical key-delivery scenario. Spare general-purpose devices cannot satisfy payment demand.
500 payment transactions/s per site; 500 across all sites.
Classical recipe per transaction: 2 × AES-128 + 0.05 × RSA-2048. Hypothetical PQC recipe: 2 × AES-256 + 0.05 × ML-KEM-768. Multiply each recipe by per-site TPS.
| Scenario | Site | Required incl. headroom | Required incl. redundancy | Allocated | Model result | Assumed utilisation |
|---|---|---|---|---|---|---|
| Classical · payment profile | 1 | 1 | 2 | 2 | Meets assumptions | 6.25% |
| PQC · payment profile | 1 | 1 | 2 | 2 | Meets assumptions | 18.75% |
Fictional payment performance assumptions
Fictional payment profile
Fictional payment operations and rates. Not a PCI/EMV recommendation or product measurement. Zero denotes an unsupported operation.
Zero means unsupported. Edited values are user assumptions.
HSM object storage assumptions
Separate from throughput and TLS-chain sizing. This estimates bytes for an expanded private key and, optionally, its certificate. Vendor object limits, alignment, wrapping and storage policies can change capacity. Composite/dual material is a byte-sum illustration, not an encoded certificate format or a mixed-chain recommendation.
Modelled object: 10,412.85 B including overhead. Estimated objects fitting in this byte budget: 6,444.
RSA private-key encoding is estimated; PQC uses expanded key material. Seed-only storage is outside this example. A KEM key cannot sign: any included ML-KEM certificate uses a separately assumed ML-DSA issuer. The wrapper and template allowances are editable estimates, not measured encodings. Storage settings are separate from the throughput report export.
Production Tool Capabilities & Verification Requirements
What this sample check establishes: Calculates projected HSM NVRAM storage, partition object counts, and throughput impact when migrating from RSA/ECDSA keys to larger ML-KEM, ML-DSA, or LMS/XMSS keys.
Browser sandbox limitation (non-certification boundary): Performs deterministic arithmetic sizing from parameter-set byte sizes; it does not benchmark real HSM hardware latency, bus contention, or firmware memory fragmentation.
What to require and test in your production tools:
- HSM firmware with vendor-documented NVRAM object overhead for ML-KEM-768/1024, ML-DSA-65/87, and stateful LMS/XMSS trees (PKCS#11 v3.2 defines the mechanisms but not storage overhead).
- Load-tested signing/decapsulation ops/sec under multi-tenant partition concurrency on your actual HSM appliance model.
- High-availability HSM cluster replication and backup capacity sized for substantially larger post-quantum key and signature objects (for example an ML-DSA-65 signature is 3,309 B against 256 B for RSA-2048).
Try it
- 1. Predict before you run it.
- 2. Run or change the tool above.
- 3. Explain what the result shows.
A workload consumes one nominal device under the assumed rates. At 70% target utilisation and N+1, how many devices does the model require per site?
Next step
Turn it into a plan: Infrastructure Modernization PlannerThis tool practises the PKI module, phase 6 (Infrastructure & Performance); Infrastructure Modernization Planner produces a deliverable of that phase.
Related content
Next in HSM / PKCS#11