Skip to content

Hosting architecture · server migration

Find the right hosting architecture for your business.

Dedicated, virtualised or private cloud is an architecture decision, not a price comparison. Getting it the wrong way round is why migrations cost more than they should.

Plan a migration Contact: hosting or migration questions

Independent reference for the people who buy and run infrastructure. No supplier is recommended here, and no prices or uptime figures are quoted.

The same three workloads on three hosting architectures, drawn as engineering elevationsA flat front-elevation drawing that changes in stages. View A: three workloads, each in its own dedicated server with local disks and idle headroom. View B: the same three workloads as virtual machines inside one host, divided by a hypervisor layer, on one shared storage array that forms a shared fault domain. View C: the workloads placed across a pool of four hosts by an orchestration layer, over distributed storage, inside a provisioned pool. Illustrative, not to scale.VIEW C · PRIVATE CLOUDORCHESTRATION LAYERPROVISIONED POOLHOST 1HOST 2HOST 3HOST 4DISTRIBUTED STORAGEVIEW B · VIRTUALISEDONE HOST, SEVERAL VIRTUAL MACHINESHYPERVISOR LAYERSHARED STORAGE ARRAYSHARED FAULT DOMAINVIEW A · DEDICATEDIDLE HEADROOMLOCAL DISKSSERVER 1IDLE HEADROOMLOCAL DISKSSERVER 2IDLE HEADROOMLOCAL DISKSSERVER 3WORKLOADWORKLOADWORKLOADELEVATION · ILLUSTRATIVENOT TO SCALE VIEW A · DEDICATEDSERVER 1WORKLOADIDLEHEADROOMLOCAL DISKSSERVER 2WORKLOADIDLEHEADROOMLOCAL DISKSSERVER 3WORKLOADIDLEHEADROOMLOCAL DISKSEACH SERVER IS ITS OWN FAULT DOMAINVIEW B · VIRTUALISEDONE HOST, SEVERAL VIRTUAL MACHINESHYPERVISOR LAYERWORKLOADWORKLOADWORKLOADSHARED STORAGE ARRAYONE SHARED FAULT DOMAINVIEW C · PRIVATE CLOUDORCHESTRATION LAYERPROVISIONED POOLWORKLOADHOST 1WORKLOADHOST 2WORKLOADHOST 3HOST 4DISTRIBUTED STORAGEFAULT DOMAIN SET BY POOL DESIGNILLUSTRATIVE · NOT TO SCALE
Fig. 1 · the same three workloads, three architectures · illustrative, not to scale
  1. A · DedicatedOne workload per physical server, with local disks.Trade-off accepted: capacity is bought in whole servers, so headroom sits idle and growth means new hardware.
  2. B · VirtualisedVirtual machines share one host, through a hypervisor layer, and one storage array.Trade-off accepted: a host or storage fault now touches several workloads at once unless redundancy is built in.
  3. C · Private cloudA pool of hosts, placed and scaled by an orchestration layer.Trade-off accepted: elasticity is limited to the pool you provisioned, and the platform itself must be operated.

The position

Choose the architecture, then the supplier.

Dedicated, virtualised or private cloud decides who controls what, how capacity grows and what one failure can take down. A supplier can only deliver the architecture you chose, so settle that first.

Control
Highest on dedicated hardware. Under private cloud it moves from machines to policy.
Elasticity
Slow on dedicated servers, quicker on virtualised hosts, highest on a pool, and always bounded by what was provisioned.
Failure domain
One server, a host with its shared storage, or a design choice. Know which before you sign.

Moving

A migration is five stages, and the last one is how you leave.

Most failed moves skip the unglamorous first two stages or never rehearse the fifth. The server migration guide covers each one, and the planner shows the transfer arithmetic.

Migration from an existing environment to a target environment Five stages. Discovery scans the existing environment. Dependency mapping draws the links between its components. Seeding copies data across to matching slots in the target environment while the source keeps serving. Cutover moves the traffic path from the source to the target. Rollback keeps a return path from the target back to the intact source. Illustrative. EXISTING ENVIRONMENT TARGET ENVIRONMENT WEBAPPDBFILESAUTH WEBAPPDBFILESAUTH SEEDING · SOURCE KEEPS SERVING CUTOVER · USERS ROUTED TO TARGET USERS RETURN PATH · SOURCE KEPT INTACT
Fig. 2 · migration sequence · illustrative, not a prescription for any particular environment

Failure domains

One fault. How far does it travel?

The same three workloads, W1 to W3, in each architecture, and the same single fault. The architecture decides the blast radius, which is why the table below has a row for it.

A Dedicated

SERVER 1SERVER 2SERVER 3W1W2W3!ALL WORKLOADS RUNNING1 OF 3 WORKLOADS DOWN
One server fails and takes its own workload with it. The other two are untouched, because nothing is shared.

B Virtualised

ONE HOSTHYPERVISORW1W2W3SHARED STORAGE ARRAY!ALL WORKLOADS RUNNING3 OF 3 WORKLOADS DOWN
The host, or the storage array under it, fails and every machine on it goes down together, unless redundancy is built in.

C Private cloud

ORCHESTRATION LAYERHOST 1HOST 2HOST 3HOST 4W1W2W3W3DISTRIBUTED STORAGE!ALL WORKLOADS RUNNINGHOST 3 FAULT, W3 DOWNW3 RESTARTED ON HOST 4
A host fails and the platform restarts its workload on another host, but only if the pool was provisioned with spare capacity.

Fig. 3 · illustration of a general pattern, not a customer environment. Not to scale.

The decision, compressed

Three architectures, five questions

The same requirements, answered three different ways. These are qualitative statements about the architecture classes, not claims about any product. The full comparison matrix adds colocation and public cloud and shows how to weight each row.

Fig. 4 · architecture class against question
QuestionDedicatedVirtualisedPrivate cloud
What you buyWhole physical servers.Capacity on shared hosts, divided into virtual machines.A pool of resources plus the automation that allocates it.
ControlHighest: hardware, firmware, operating system and storage layout are yours to choose.High at the guest. At the host, only as much as the host’s owner allows.Moves from machines to policy: quotas, placement rules and templates.
ElasticitySlow. More capacity means more hardware.Moderate. New machines are quick while host headroom lasts.Highest of the three, bounded by the pool that was provisioned.
Failure domainOne server, unless you build a cluster.A host or its shared storage, taking several workloads with it.Set by design (pool, rack, site). The platform layer is a new failure domain.
Trade-off acceptedIdle headroom is paid for, and growth is stepwise.Shared dependencies concentrate risk unless redundancy is built in.The platform needs operating, and capacity still has a ceiling.

Where to start

Start from the question you are actually asking

01

Choosing

Which architecture fits?

Settle control, elasticity, failure domains and licensing before comparing suppliers.

02

Moving

How do we migrate safely?

Discovery, dependency mapping, seeding, cutover and rollback, with the transfer arithmetic shown.

03

Keeping it running

What protects us, and who runs it?

High availability, backup and disaster recovery are three different things. So are a supplier and a partner.

Planning a move? Work through the numbers first.

The migration readiness planner is a rule-based questionnaire. It shows the transfer arithmetic and the questions to settle, and it is not an audit of your systems.

Plan a migration Contact: hosting or migration questions