Continuous real-time sync
Every change on the old system is applied to the new one as it happens, for the whole parallel run. More on our real-time replication page.
iblync · Parallel run and cutover
Run the old and the new system side by side on live data, compare them, and switch only when your own go or no-go criteria are met.
A parallel run keeps the old system and the new one working on the same live data for an agreed period before cutover. It turns the switch from a leap into a checked decision: you compare results, rehearse the cutover, and move only when the criteria are met. iblync keeps the two sides in sync during that period.
A parallel run is a period, agreed before it starts, in which the old system and the new one both work on the same live data. The old system stays in service and remains the reference. The new one receives every change as it happens and is checked against the old one, again and again, until the business is ready to switch.
The pattern suits a move to a new platform, a migration between different database engines and a change of replication tool. What stays live in each phase, and what the final switch can still ask of you, is set out on our zero downtime migration page.
Relatedmigration between different database engineschange of replication toolzero downtime migration
A parallel run is worth its effort because it answers three questions with evidence before the cutover night, not after it.
If any of the three is not yet true, the parallel run simply continues. Nothing has been switched, and the old system is still serving.
Six stages, in this order. How long each one lasts is set in the assessment, for your own systems.
Stage 01
The existing data is copied once to the new system, with transfer speed limited so that production is not slowed.
Stage 02
From then on, every change on the old system is applied to the new one as it happens. If the new system is unavailable, changes wait in a queue and are applied when it returns.
Stage 03
Record counts and agreed samples are compared on a set schedule, and every difference is explained and closed before the next comparison.
Stage 04
The whole runbook is carried out end to end, on a copy or in a quiet window, including the way back. Whatever failed is fixed and rehearsed again.
Stage 05
The people who decide meet and check the criteria agreed in advance. Any criterion not met means no-go, and the parallel run continues.
Stage 06
The runbook is carried out. The old system stays available through an agreed monitoring period, and is switched off only when the business confirms the new one.
Eight conditions, each one measurable, written down and signed before the night. On the night they are checked one by one, not debated.
| Step | Owner | Evidence it is done | If it fails |
|---|---|---|---|
| Confirm the go decision | Decision owner | The go or no-go criteria checked and the decision recorded | No-go: the old system stays in service and the parallel run continues |
| Announce the start of the window | Project lead | Message sent on the agreed channel to every team and to users | Hold the start and announce the new time |
| Stop writes on the old system | Application owner | No new transactions recorded on the old system | Reopen writes and stop the cutover |
| Wait for the last change to arrive | Replication team | No change left waiting between the two systems | Reopen writes on the old system and return to the parallel run |
| Run the final comparison | Data owner | Record counts and agreed samples match | Follow the fallback plan: the old system resumes service |
| Switch the connections | Infrastructure team | Applications, integrations and reports point to the new system, each one ticked off a list | Point the connections back to the old system |
| Run the smoke test | Business testers | The agreed key transactions pass on the new system | Follow the fallback plan |
| Open writes on the new system | Application owner | The first real transactions are recorded on the new system | Follow the fallback plan |
| Announce the result | Project lead | Message sent on the agreed channel | Announce the return to the old system and the next date |
| Start the monitoring period | Operations team | The agreed checks run on schedule, and the old system is still available | Decide on the fallback plan with the decision owner |
Copy the table into your own plan and replace the roles with names. The order matters more than the wording: writes stop before the last change is awaited, and the final comparison comes before any connection moves. How to rehearse the night and what a way back means are covered in our migration checklist.
iblync, Interkey's platform for migration and continuous sync, keeps the two sides in step. It is delivered with the Interkey team that operates it.
Every change on the old system is applied to the new one as it happens, for the whole parallel run. More on our real-time replication page.
Transfer speed and resource use can be limited, so the old system keeps serving users at its normal pace while the new one is filled.
If the new system is unavailable, for maintenance or a fix, changes wait in a queue and are applied when it returns.
Both databases accept writes, and each side's changes are applied to the other continuously. More on our bidirectional sync page.
A direct move can be enough
A parallel run costs effort: two systems to run, compare and watch. It is not always worth it. If the data is small and the business can accept a downtime window long enough to copy it once and check it, a direct move with a careful check is simpler. The same is true of a read-only system, such as an archive, where nothing changes during the move. Our migration checklist helps you decide.
Relatedmigration checklist
A parallel run is a period in which the old system and the new one both work on the same live data before cutover, while their results are compared. Its length is agreed in advance. The old system stays the reference throughout, and the switch happens only when criteria agreed beforehand are met, so the decision rests on evidence rather than on a date in the plan.
Long enough to cover one full business cycle that includes your most important operations, such as a month-end close. There is no fixed number: the length depends on your own systems and processes, and it is set in the assessment. If the go or no-go criteria are not met at the end, the parallel run simply continues until they are.
A cutover plan is the document that sets out the steps of the cutover night in order, who carries out each step, the evidence that each one is done, and what to do if it fails. It is written before the night and rehearsed at least once. The runbook table on this page is a template you can copy into your own plan.
They are measurable conditions, agreed and written down before the cutover night, that decide whether the switch goes ahead. Typical ones are matching record counts, matching agreed samples, sync lag within the agreed limit and a successful rehearsal. On the night they are checked one by one, and the decision follows the criteria, not the mood in the room.
The old system keeps serving as before. With iblync, changes wait in a queue while the new system is unavailable and are applied when it returns. The comparison then resumes, and every stage affected by the stop is checked again before the go or no-go decision, so the interruption is visible in the evidence rather than hidden by it.
Yes. iblync applies every change from the source to the target as it happens, for the whole parallel run, with transfer speed and resource use limited so production is not slowed. Where both systems must accept writes, it also supports synchronisation in both directions. Our real-time replication page explains how it works.
Relatedreal-time replication
Next
Next step
Tell us the source system and the new one, the rough data volume, the cutover window you have in mind, and what the parallel run must prove before you decide.
What happens next
Built and supported by Interkey
An Interkey product, delivered and supported by Interkey
One platform for migration and continuous sync
The move and the replication that follows it
Written quote after one meeting
The price is discussed in a meeting about your systems
Or directly
+966-11-2180999 [email protected]Tawuniya Towers, North Tower, 7th Floor, King Fahad Highway, Olaya, P.O. Box 56835, Riyadh 11564, Saudi Arabia