Techbanq

Cloud · migration

The overrun is in the systems nobody could describe

Planned workloads move roughly on schedule. What consumes the timeline is the batch job with no owner, the appliance with a licence tied to a hardware address, and the integration that turns out to be a file dropped on a share.

Serving Saudi Arabia from our Riyadh office and our engineering hubs in Lahore and Karachi

Market
Saudi Arabia
Written for
CTO, CIO
Decision
Plan and execute a migration to cloud infrastructure
Languages
EN · AR

Information security for cloud delivery

  • ISO

    ISO/IEC 27001

    ISO/IEC 27001 certified

Certifications & registrations

Discovery

Documentation describes the architecture. Traffic describes the system

We do not have an office in Saudi Arabia

We serve organisations there from our Riyadh office and our engineering hubs in Lahore and Karachi, travelling for the parts of an engagement that need to happen in person.

Every organisation has an architecture diagram and none of them is complete, because the connections that get drawn are the ones somebody designed. The ones that break a migration are the ones that accumulated: a report that queries the production database directly, a scheduled task on a workstation under a desk, an integration implemented as a file written to a share at two in the morning.

So discovery is done by observing traffic for a period rather than by interviewing people. Two to four weeks of flow data shows what actually talks to what, including the connections that only appear at month end — which is why a two-week window that misses a month boundary produces a map that is wrong in an expensive way.

The output is a dependency map and, from it, a disposition per system: retire it, replace it with a service, rehost it as it is, replatform it onto a managed component, or leave it where it is. That last option belongs on the list. A system that will be decommissioned in two years, or that depends on hardware that cannot move, is cheaper left alone, and a migration plan without a "leave in place" column is not a plan that was allowed to reach the right answer.

Execution

Landing zone first, then waves

STAGE 01

1 · Landing zone

Account structure, identity federation, network boundaries, logging and cost allocation, established before the first workload arrives. Retrofitting identity onto a populated estate is the single most common source of rework. A destination that does not need rebuilding at workload thirty.

STAGE 02

2 · A first wave chosen to be boring

Low-dependency, low-consequence systems that exercise the whole path — build, deploy, monitor, back up, restore — without a business impact if a step is wrong. A rehearsed process, and a team that has done it once.

STAGE 03

3 · Waves by dependency cluster

Systems that talk to each other move together. Splitting a cluster across waves means running the connection across the boundary, and that temporary arrangement is where latency and cost surprises live. Fewer boundary crossings, and a shorter dual-running period.

STAGE 04

4 · Decommission deliberately

The saving arrives when the old thing is switched off, and it frequently is not — because nobody is certain nothing still uses it. Set a date at migration time and verify with the same traffic observation that ran at discovery. The saving the business case assumed.

What surprises people

Four costs that are not in the business case

  1. 01

    Dual running

    Both environments run through the migration, and the old one cannot be switched off until the last dependent system moves. Budget for the overlap explicitly; it is a known cost that is routinely left out.

  2. 02

    Egress and inter-region traffic

    Data leaving the platform is charged, and a split dependency cluster generates far more of it than the design anticipated. This is the line item that makes a first cloud bill unrecognisable.

  3. 03

    Licences that do not travel

    Software licensed per physical core, or tied to a hardware identifier, or priced differently on cloud infrastructure. Check the licence terms of every system in the plan during discovery, not during the wave.

  4. 04

    Skills, before the tooling helps

    A team that ran servers now runs a platform, and the first six months are slower while that is true. Training and a period of lower throughput are real costs, and hiding them makes the first quarter look like failure.

Scope boundary

What we will say before you commit

What we do

  • Run discovery from observed traffic across a period long enough to catch month-end behaviour.
  • Give every system a disposition, including "leave in place" where that is the cheaper answer.
  • Model the cost through the migration, including dual running, rather than only the destination cost.

Which region a workload may run in, and where its backups and logs may be replicated, follow from data protection and sector rules rather than from architecture preference. Settle the residency position before the landing zone is designed — it constrains the account structure, not just the placement.

Primary source

Common questions

What buyers ask

Should we lift and shift or modernise as we go?+

Per system rather than as a policy. Lift and shift is right for anything with a short remaining life or a dependency that cannot be untangled quickly; modernising during the move is right where the system is staying for years and the managed alternative removes work you are currently doing. What consistently goes wrong is modernising everything, because it converts a migration with a date into a rewrite without one.

How long does a migration take?+

Discovery is four to six weeks, the landing zone another four to six, and the waves depend almost entirely on the dependency structure rather than the workload count. Two hundred loosely coupled systems move faster than forty tightly coupled ones. The honest answer after discovery is a range with the two or three unknowns named, and anybody giving a date before discovery is giving you a template.

Can we migrate without downtime?+

For most systems, with replication and a cutover window measured in minutes. For a few — usually a database with no replication path, or an appliance with a single instance — a maintenance window is the honest answer and it is cheaper than the engineering to avoid one. Identify those during discovery so the downtime conversation with the business happens months early rather than the week before.

Next step

Start with dependency discovery

A discovery that produces a dependency map and a disposition per system, which is what makes a migration plan a plan rather than a wish.