Skip to main content
GIFRÖST / Compare

Choose the right comparison depth for the job.

Each Compare algorithm balances speed, precision, infrastructure cost, and transferred data differently. Start with a light volume check, move to sampled control for large tables, use agent-side chunk hashing when payload must stay low, or run a full row audit when the result needs to stand up to a cutover or investigation.

4 comparison modesFrom a quick sanity check to agent-side chunk diagnostics and a full value-by-value audit.
Different execution costMatch the algorithm to table size, SLA, and the time window you have.
Easy escalation pathBegin light, then increase confidence only where the data or risk justifies it.

Algorithm overview

Compare offers four ways to validate consistency. The right choice depends on whether you are optimizing for speed, confidence, repeatable monitoring at scale, or low data movement in agent-based runs.

Row Count icon

Fast baseline

Row Count

Checks whether the source and target contain the same number of rows. It is the fastest way to confirm that a migration or replication flow did not diverge at the volume level.

Best when

you need a quick signal after an initial load, batch migration, export/import, or nightly processing.

Cost profile
Lowest system impact and the shortest execution time.
Blind spot
It will not detect content mismatches when the row totals still match.
FastestLow overheadSmoke test
Row Level icon

Full audit

Row Level

Performs a full row-by-row, value-level comparison across the table. Use it when certainty is more important than runtime and the output may need to support a cutover, audit, or incident review.

Best when

you need hard validation before go-live, after an incident, or whenever discrepancies must be explained precisely.

Cost profile
Highest confidence, but also the highest infrastructure load and longest runtime.
Tradeoff
Large tables may require deliberate scheduling and a controlled execution window.
Most preciseHigh costAudit-ready
Row Nlevel icon

Sampled control

Row Nlevel

Compares every n-th row using a configurable interval. It offers a practical balance between confidence and execution time for large tables that are checked repeatedly.

Best when

you want regular control over high-volume tables without running a full scan every time.

Cost profile
Moderate overhead with much better scalability on large datasets.
Blind spot
Differences that fall between sampled rows can remain undetected.
BalancedLarge tablesMonitoring
Chunk Hash icon

Agent-first diagnostics

Chunk Hash

Available only in agent mode. Agents send hash signatures for consecutive data chunks instead of shipping full rows. Compare advances while hashes match. At the first mismatch it requests the decrypted chunk from both sides, compares that window in detail, and reports exactly which records differ inside it.

Best when

you want low network transfer on large tables, but still need a precise explanation of the first persistent difference.

Cost profile
Very efficient on in-sync data because only chunk hashes move until a mismatch appears.
Tradeoff
It stops at the first mismatching chunk, so a rerun is needed to continue beyond that point.
Agent-onlyLow transferFirst mismatch

Quick selection guide

01

Need to confirm only the volume?

Choose Row Count when the main question is whether the target has the same record total as the source.

02

Working with large tables and limited runtime?

Choose Row Nlevel for recurring controls where a representative sample is good enough for operational monitoring.

03

Need low transfer and the first exact mismatch?

Choose Chunk Hash in Agent-Based Connection mode when you want hashes to flow chunk-by-chunk and full values only for the first differing chunk.

04

Need evidence-grade validation?

Choose Row Level when the output must support a cutover decision, an audit, or root-cause analysis.

Typical use cases

After initial load

Fast smoke test after a batch migration

Start by checking whether the target carries the same row volume as the source before you spend time on deeper analysis.

Recommended: Row Count
Recurring monitoring

Large tables in active replication

When the table is big and the comparison runs regularly, sampled control keeps the feedback loop fast without scanning everything in full.

Recommended: Row Nlevel
Go-live or incident

Validation before cutover or after a discrepancy

When the result must be defensible and specific, use the most detailed comparison and work from exact row-level differences.

Recommended: Row Level
Agent-based diagnostics

Large or remote tables where transfer must stay low

Keep traffic minimal by exchanging chunk hashes first, then pull the first mismatching chunk in decrypted form only when detailed diagnosis is required.

Recommended: Chunk Hash
Operational pattern

Escalate instead of auditing everything

A practical workflow is to start with Row Count, move recurring controls to Row Nlevel, use Chunk Hash when agent mode should stop at the first real drift, and reserve Row Level for critical tables and escalated analysis.

Recommended flow