Migration Evaluator
Prerequisites: temporary-credential IAM roles, VPC routing and the rollback plan.
Migration Evaluator builds a data-informed migration business case. Its agentless assessment covers supported physical/virtual Windows and Linux profiles, including SQL Server discovery; it is not a VMware-only agent installation service.
Scope and distinctions
Do not conflate the Migration Evaluator Collector with Application Discovery’s Agentless Collector, Discovery Agent or historical Connector. Evaluator assessment/sizing outputs support a business case; they do not automatically choose a workload’s migration strategy or replace application-owner dependency analysis.
Inventory licensing, peak utilization, memory pressure, storage throughput/IOPS, support deadlines, network connectivity and migration effort. A low average CPU value alone is not a safe downsizing decision.
Assessment exercise
- Define the business question, source scope, data classification and representative observation period. Identify data owners and obtain collection authorization.
- Follow the owning request/collector workflow; validate hypervisor/physical source support, read-only credentials, connectivity and collection permissions. Alternatively use the supported data-import process for an eligible dataset.
- Compare collected server counts and observation periods with authoritative inventory. Investigate missing hosts and collection failures before calculating savings.
- Review current-state and proposed-state assumptions with finance and application owners: licensing model, Region, target instance family, storage, operational overlap and commitments.
- Create two scenarios (conservative baseline and optimized), record assumptions and uncertainty, and compare peak-capacity requirements. Do not use an invented CLI namespace/assessment command or fictional CSV schema.
- Stop/remove the disposable collector and revoke its collection credentials when finished. Preserve approved business-case artifacts under the organization’s retention rules.
Output quality
A defensible TCO includes infrastructure, licenses, labor/operations, network, disaster recovery and migration overlap. Separate observed usage from projected savings and commitment risk; a modeled discount is not a guaranteed invoice.
Translate findings into seven-R decisions and migration waves. Pair this work with source-specific dependency discovery rather than claiming the evaluator supplies detailed process/network telemetry.
Cost and lifecycle
Review current service eligibility and assessment terms before starting. The business-case tool’s terms do not make collection hosts, migration staging or target resources free. Reference current AWS pricing inputs rather than retaining stale universal instance rates.
Assessment architecture and evidence model
authorized physical/virtual inventory + representative utilization
↓
agentless collector / supported imported assessment data
↓
quality checks → licensing/capacity assumptions → AWS scenarios
↓
business-case report → owner/finance review → migration strategy| Evidence | Question | Incorrect shortcut |
|---|---|---|
| Inventory/support | Which systems/versions/owners are in scope? | Assume VMware-only or every OS is supported |
| CPU/memory profiles | What peaks/headroom must the target absorb? | Size from CPU average alone |
| Storage/network | What capacity/IOPS/throughput/latency are needed? | Compare only RAM and vCPU |
| Licensing | What contracts/mobility/SQL options apply? | Apply one blanket OS discount |
| Dependencies | What must move together or remain reachable? | Infer complete topology from a sizing report |
| Observation quality | Which windows/hosts are missing? | Treat an incomplete collector run as the portfolio |
Server-level recommendation walkthrough
For one synthetic server record, keep observed and projected values separate. Suppose average CPU is 10% but p95 is 70%, memory peak is 14 GB and a batch window is latency-sensitive. A candidate target must satisfy CPU architecture/performance, memory headroom, disk/network and support/license requirements. “10% CPU therefore one tenth the vCPU” is not a defensible sizing rule.
Compare an initial compatible configuration, a right-sized configuration and a managed-platform alternative at equal availability. State confidence and missing measurement. Test the candidate using representative application/batch data before accepting the report’s estimate.
Business-case worksheet and dimensional example
| Cost row | Current state | AWS scenario | Evidence |
|---|---|---|---|
| Compute | Owned/leased capacity and remaining commitments | Running hours/purchase model/Region/OS | Utilization plus current scoped tariff |
| Storage | Capacity, performance, backup and support | Volumes, I/O, snapshots/copies/restore | Retention and measured demand |
| Software | Existing licenses/support | BYOL/managed/licensed options | Contract and support eligibility |
| Network | Circuits/firewalls/egress | Transfer, gateways/endpoints/circuits | Direction, GB and hours |
| Operations | Current people/tooling | Migration/steady-state responsibilities | Explicit ownership |
| Overlap | Existing environment | Staging/test/parallel-running period | Cutover/decommission timeline |
At hypothetical $0.20/hour, twenty instances for 730 hours are $2,920 before dependencies—not twenty × $0.20 as a monthly estimate. A claimed 30% savings applies only to the cost rows actually reduced; it does not discount license/support/network automatically. Calculate baseline/scenario totals and show the assumption change.
Assessment variants and migration handoff
A preliminary business case can use approved imported inventory; a measured assessment improves evidence but still needs quality checks. These are engagement/data-quality stages, not an invented agent-based versus VMware-only product catalog.
Validate licensing, criticality, effort and dependency requirements with owners before seven-R classification. Record assessment revision, data period, assumptions, recommended tests and unresolved systems. Feed the approved evidence into a supported tracking workflow; do not invent export-to-Migration-Hub API calls.
Stop/revoke disposable collection after completion, retain approved reports and remove sensitive raw exports according to policy. Target/staging and any collector hosting resources remain separate cost dimensions.
References: Evaluator capabilities, Evaluator resources, migration business case.