Back

Server Virtualisation (VMware / Hyper-V) Services

Workloads virtualised 38% to 92%, host CPU utilisation 18% to 68%, VM density per host 25% to 86%, provisioning automated 20% to 88%, supported firmware baseline 45% to 97%.
Figure 1. Platform efficiency before and after a server virtualisation programme, each measure expressed as a share of the estate in the target state. Workloads virtualised rise from 38 to 92 per cent, host CPU utilisation from 18 to 68 per cent, virtual machine density per host from 25 to 86 per cent, provisioning automation from 20 to 88 per cent and a supported firmware baseline from 45 to 97 per cent. Utilisation is the number that pays for the programme; density is the number that decides whether the licensing bill goes down as well as the hardware bill.
Chart values are an illustrative planning model based on typical UMC engagements and published industry benchmarks. They are not a guarantee of results on any specific project.

Most enterprises do not have a server problem. They have forty-eight servers that are each eleven per cent busy, four of which matter, and a licensing position that was negotiated against a hardware estate which no longer exists. Virtualisation is not a technology purchase. It is the decision to stop buying capacity by the box and start managing it as a pool — and the platform you choose to hold that pool now costs more than the hardware it replaced.

Server virtualisation at Unified Management Consulting covers the whole programme: the inventory that establishes what you actually run, the platform decision that follows from it rather than from a renewal notice, the cluster design that decides what happens when a host fails, the migration waves that move production without stopping it, and the operating model that keeps the estate honest afterwards. We work across VMware vSphere, Microsoft Hyper-V, Proxmox VE, Nutanix AHV and KVM-based platforms, and we are equally willing to recommend consolidating onto the platform you already own.

The commercial context has changed enough that the platform question can no longer be deferred. Perpetual hypervisor licensing has largely given way to subscription bundles sold against per-core minimums, which means the cost of a virtualisation estate now tracks the number of physical cores you carry rather than the number of virtual machines you run. That inverts a decade of behaviour: the discipline that used to reduce hardware cost while increasing licensing cost is now the discipline that reduces both, because fewer, denser hosts carry fewer licensed cores. An estate that was consolidated to 2019 assumptions can be paying for cores it does not need, and a platform evaluation that ignores this is evaluating the wrong thing.

🖥️ Server Virtualisation (VMware / Hyper-V) — Quick Summary

Component What We Deliver
Estate inventory and assessment Every physical and virtual machine catalogued with measured utilisation, dependencies, operating system support status and criticality — not nameplate ratings, and not whatever the last migration spreadsheet claimed.
Platform recommendation A written comparison of VMware vSphere, Hyper-V, Proxmox VE, Nutanix AHV and KVM against your workload profile, skills, licensing exposure and five-year cost, with a recommendation and the reasoning behind it.
Cluster and host design Cluster sizing, HA admission control, DRS or equivalent scheduling, anti-affinity rules, failure domains, and the blast radius each decision creates.
Host build and firmware baseline Hypervisor installation, management and production network separation, multipathing, a documented firmware and driver baseline, and secure boot where the platform supports it.
Storage attachment Shared storage, vSAN or hyperconverged storage, or local storage with replication — chosen against the workload’s latency and recovery needs rather than habit.
Migration execution P2V and V2V waves sequenced by dependency and risk, with rollback plans, cutover windows and post-migration verification per workload.
Licensing and right-sizing Core counts rationalised, entitlement evidence assembled, and the subscription position modelled against the consolidated estate.
Operations handover Patching and lifecycle process, backup integration, monitoring thresholds, capacity reporting in three units, and the runbooks the operations team will actually use.
Decommissioning Physical hosts retired with data sanitisation certificates, licences surrendered or reassigned, and rack space, power and cooling reclaimed and reported.

What Virtualisation Actually Changes

Virtualisation is the substitution of a scheduling problem for a procurement problem. Instead of buying capacity in fixed units and stranding most of it, capacity becomes a pool that is allocated to workloads on demand and reclaimed when they do not need it. The technology is mature and well understood; what separates a good programme from a bad one is whether the inventory was honest, whether the cluster was designed for failure rather than for capacity, and whether anyone planned what happens after the last physical server is switched off.

What virtualisation changes Before After The catch
Capacity procurement Per server, with each workload owning its peak Per pool, sized to the aggregate peak An undersized pool fails all at once rather than one workload at a time
Hardware utilisation Typically 8-15 per cent of CPU on distributed servers Typically 55-75 per cent on a consolidated cluster Measurement has to be real: a quarterly average hides the month-end peak
Provisioning time Weeks, gated by procurement and racking Minutes, from a template Templates that are never updated become a patching liability
Recovery from hardware failure Manual rebuild, hours to days Automatic restart on a surviving host, minutes Only for workloads that were designed to be restarted, and only within HA admission limits
Patch and lifecycle work Per machine, with per-machine windows Per cluster, with rolling evacuation Requires enough spare capacity to evacuate a host, which costs money
Licensing exposure Per server or per socket in the old models Per core against minimums in current subscription models Consolidation reduces cores only if the hosts are actually retired, not repurposed

The catch column is the one worth reading twice. Every benefit of virtualisation is also a concentration of risk: capacity that fails together, a patching process that depends on spare headroom, and a licensing model that rewards decommissioning rather than migration. A programme that delivers the benefits and ignores the catches produces a smaller, cheaper estate that is more fragile than the one it replaced.

Assessment: Measured Inventory Before Any Platform Decision

We assess before we recommend, and the assessment is measured rather than surveyed. Nameplate specifications and the opinions of application owners are both systematically wrong in the same direction: owners overstate what their workload needs, because understating it once caused an outage they would rather not repeat. Measured utilisation over a period that includes the worst week — month end, quarter end, the seasonal peak — is the only defensible basis for a sizing decision.

Assessment output How we produce it What it decides
Machine inventory with dependency map Agent-based collection where permitted, agentless or hypervisor-level where not, plus network flow analysis for dependencies nobody documented Migration wave order, and which workloads cannot be separated
Measured CPU, memory, disk and network utilisation At least four weeks of collection, including the worst week, at a granularity fine enough to show peaks Host sizing, memory overcommit ratio, and the storage tier each workload needs
Operating system and application support status Version census against vendor support matrices What must be upgraded before it can be migrated, and what cannot be virtualised at all
Physical-to-virtual suitability Driver, licensing, hardware dongle, clustering and latency review per machine The short list of workloads that stay physical, and why
Recovery requirement per workload Agreed RPO and RTO with the business owner, not with IT Cluster design, backup schedule, and whether a second site is needed
Licensing position Entitlement evidence collected against the actual estate Whether consolidation reduces or increases cost, and which platform is affordable
Capacity headroom requirement Modelled from the consolidated peak plus N+1 host failure plus growth How many hosts the cluster needs, which is more than the arithmetic suggests

Two findings appear in almost every assessment. The first is that the estate is larger than anyone believed, because machines accumulate without an owner and without a decommission process. The second is that a small number of workloads — typically fewer than ten per cent — account for most of the resource consumption, which means the migration sequence should start with the easy majority and handle the demanding minority deliberately rather than accidentally.

Platform Choice Under Current Licensing Realities

Platform choice is now a commercial decision with a technical constraint attached, rather than the reverse. All the mainstream hypervisors are technically capable of running an enterprise estate; what differs is the licensing model, the management tooling, the skills your team already has, and the cost of leaving. We model five-year cost including subscription, support, training and the migration labour itself, because a platform that is cheap to license and expensive to leave is not cheap.

Platform Licensing model Management and ecosystem Best suited to What to weigh
VMware vSphere / ESXi Subscription bundles with per-core minimums; perpetual licensing largely withdrawn vCenter with mature DRS, HA, vMotion and the widest third-party ecosystem Large mixed estates that need proven scheduling and broad vendor support Current bundle pricing against your core count, and the cost of the features you actually use
Microsoft Hyper-V Included with Windows Server licensing, plus System Center or Azure-based management Failover Clustering with live migration; strongest where the estate is already Microsoft Windows-dominant estates, organisations with existing Microsoft agreements, and Azure-hybrid designs Whether your team can operate clustering without the VMware tooling it is used to
Proxmox VE Open-source core with optional paid subscription for support and the enterprise repository Web-based management, KVM hosts with LXC containers, built-in backup and clustering Small and mid-size estates, cost-sensitive consolidations, and teams comfortable operating Linux Support arrangements and in-house Linux depth, since the ecosystem is narrower
Nutanix AHV Subscription per node or per core, sold with the hyperconverged stack Prism management across compute and storage in one plane Estates that want compute and storage converged and simplified to operate That the platform and the storage layer are bought together, which changes the exit cost
KVM and OpenStack-based platforms Open source, with support from a distributor or built in-house Varies widely; powerful but requires real engineering ownership Organisations with platform-engineering capability, and cloud-style internal offerings The operating cost of the team, which is where the saving is spent

The question that decides it

Not “which hypervisor is better” — they are all adequate — but “which platform can our team operate at three in the morning, at a cost we can forecast for five years, without a second migration in the middle of that period?” A platform the team cannot operate produces an estate nobody patches. A platform whose cost cannot be forecast produces a budget conversation every renewal. Both failures are more expensive than the licensing difference that prompted the evaluation.

Production A 86 VMs, Production B 74, VDI 120, dev and test 58, management 22; 360 VMs across five clusters.
Figure 2. Virtual machines hosted per cluster after consolidation: Production A 86, Production B 74, VDI 120, development and test 58, and management 22 — 360 virtual machines across five clusters. The distribution matters more than the total: VDI is the largest cluster and the most latency-sensitive, while the management cluster is the smallest and the one that must never be placed on the same failure domain as the workloads it manages.
Chart values are an illustrative planning model based on typical UMC engagements and published industry benchmarks. They are not a guarantee of results on any specific project.

Cluster Design: HA, Scheduling and the Blast Radius

Cluster design is where availability is actually decided, and it is decided by arithmetic rather than by intention. High-availability admission control reserves capacity so that the surviving hosts can absorb the workload of a failed host; if the cluster is sized without that reservation, HA will restart what fits and leave the rest down, which is not the behaviour anyone expected when they bought it.

Design decision What we set Why it matters
Cluster size Enough hosts that losing one leaves the aggregate peak served, with the growth horizon included A cluster sized to today’s load fails over into an overcommit that produces latency rather than an outage — which is worse, because it is hard to diagnose
HA admission control Host failures tolerated set explicitly, with the reservation measured against real memory demand Determines whether failover is complete or partial, and it is the setting most often left at default
Scheduling policy (DRS or equivalent) Automation level chosen per cluster, with affinity rules for clustered workloads Fully automated movement is wrong for workloads with licensing or latency constraints
Anti-affinity Applied to the two halves of any clustered application, and to management components Otherwise a single host failure takes both nodes of a cluster that was bought for resilience
Failure domain separation Management, production and VDI on separate clusters or separate storage where the budget allows A shared failure domain turns a platform incident into a total outage
Memory overcommit Set from measured demand, not from a ratio copied from a reference architecture Overcommitment beyond measured headroom causes swapping, which users experience as an unexplained slowdown
Blast radius Documented per cluster: what stops if this cluster is lost The number the business needs in order to fund the right level of redundancy

We publish the blast radius of every cluster as part of the design, in plain language: if this cluster is lost, these applications stop for this long. It is the single most useful document in a virtualisation design, because it converts an engineering decision into a business decision that can be funded or accepted knowingly.

Storage and Networking Underneath the Cluster

A hypervisor cluster is only as good as the storage and network underneath it, and both are routinely under-specified because they are invisible to the people approving the budget. Storage decides whether a live migration completes in ninety seconds or times out; the network decides whether a host failure produces a graceful restart or a storm of retries.

Layer Design rule we apply Common failure we are designing against
Host networking Management, vMotion or live-migration, storage and production traffic on separate VLANs with dedicated physical uplinks, teamed with an explicit load-balancing policy Migration traffic sharing uplinks with production, so maintenance windows take four times as long
Switch configuration Jumbo frames consistently enabled end to end or consistently disabled, with MTU verified on every hop An MTU mismatch that breaks storage traffic intermittently and only under load
Storage attachment Multipathing configured and tested, with the failover policy matched to the array’s capabilities A single path that nobody noticed because it never failed during working hours
Storage tier Latency-sensitive workloads on flash with measured IOPS headroom; archives on capacity tiers Database workloads placed by convenience rather than by latency requirement
Converged storage (vSAN or equivalent) Only where the failure domain, witness and capacity model are understood, and where the team can operate it A converged cluster whose usable capacity is consumed by its own resilience policy
Backup path Backup traffic isolated from production, with a target that is not on the same storage as the source Backups that read from and write to the same array, and therefore fail together with it
Firmware and driver baseline One documented baseline per cluster, tested before rollout, with a rollback path A driver update applied host by host that changes behaviour halfway through a cluster

Where storage is converged onto the compute cluster, the design has to account for the fact that the storage layer now shares the failure domain of the hosts it serves. That is a legitimate architecture with real operational advantages, and it changes the capacity and resilience arithmetic in ways that must be modelled before purchase rather than discovered at eighty per cent full.

Hardening the Platform: The Management Plane Is Tier Zero

A virtualisation cluster concentrates value, and it therefore concentrates attack. The management plane is the single most valuable credential set in the estate: whoever controls it controls every workload on the platform, can read the memory of any machine, and can make the whole cluster unavailable. Ransomware campaigns have targeted hypervisor management interfaces specifically, because encrypting the hosts is more effective than encrypting the guests and does not require touching any of them.

Control What we implement Why it is not optional
Management plane isolation Management interfaces on a dedicated, non-routable network segment with jump-host access only, never reachable from the production VLAN An exposed management interface is an invitation, and it is how hypervisor-targeting ransomware gets in
Administrative access Named accounts with role-based permissions, no shared root-style credentials, and multifactor authentication on the management console A shared administrative password cannot be rotated after a leaver, and cannot be attributed in an incident
Host hardening baseline A documented baseline per platform: shell access disabled, unnecessary services off, secure boot and TPM attestation where supported, lockout policies set Drift from the baseline is invisible until an audit or an incident makes it visible
Firmware and hypervisor patching Advisories tracked against the cluster baseline, with a defined window for out-of-band security fixes Hypervisor vulnerabilities are patched by attackers faster than by most enterprises
Virtual machine isolation Device access restricted, copy-paste and shared virtual disks disabled between trust zones, and sensitive workloads on separate clusters These are the settings that turn a guest compromise into a platform compromise
Encryption Encryption at rest for datastores where the workload warrants it, and encryption in flight for migration traffic Live migration carries entire machine memory across the network in the clear unless it is configured otherwise
Snapshot and template integrity Templates and snapshots stored on paths with restricted write access, and reviewed for age A compromised template propagates to every machine built from it
Audit logging Management actions logged to a store the platform administrators cannot alter, with retention agreed for audit purposes Without it there is no way to establish what an administrator account did, or who used it

We treat the hypervisor as a tier-zero system: the layer whose compromise defeats every control above it. That classification changes how it is patched, who may administer it, where its interfaces are reachable from, and how its logs are retained — and it is the reason a virtualisation programme that ignores security produces a smaller, cheaper and materially more dangerous estate than the one it replaced.

Migration Waves: The Order That Keeps Production Alive

Migration is sequenced by dependency and risk, not by convenience. The first wave is deliberately unimportant: it exists to prove the process, the tooling, the cutover window and the rollback path on workloads whose failure costs nothing. The last wave is the workload everybody cares about, and by then the process has been executed enough times that the cutover is routine.

Wave Contents Purpose Exit criteria
Wave 0 — pilot Two to four non-critical machines with representative operating systems and storage profiles Prove the tooling, the cutover window and the rollback path Migration, verification and rollback all executed successfully and documented
Wave 1 — development and test The largest group by count and the least risky Build repetition and train the operations team on real work Runbook stable, verification automated, no manual steps left undocumented
Wave 2 — departmental production Workloads with a single owner and a tolerable window Extend the process to production change control Change records clean, business owners signed off per workload
Wave 3 — clustered and stateful workloads Database clusters, application clusters, anything with shared storage or quorum Handle the migrations that cannot be treated as a copy operation Quorum preserved, replication healthy, failover re-tested after migration
Wave 4 — critical and regulated The workloads whose outage is a business event Complete the estate with the most rehearsal behind the least risk Recovery tested on the new platform, evidence retained for audit
Wave 5 — decommission Physical hosts retired once nothing depends on them Realise the licensing, power, cooling and space saving Data sanitisation certificates issued, licences surrendered or reassigned, capacity reclaimed and reported

Physical-to-virtual conversion is the easy part of a wave; the hard part is everything around it. Drivers must be injected or the machine will not boot. Machine identity must survive — hostnames, IP addresses, service principal names, certificate bindings and licensing that is tied to hardware. Clustered applications must keep quorum. And the rollback path must be tested, because an untested rollback is not a plan, it is a hope.

Physical hosts fall 48, 34, 21, 12 across four quarters while virtualisation hosts rise 6, 9, 12, 14.
Figure 3. Physical hosts and virtualisation hosts across a twelve-month consolidation. Physical hosts fall from 48 in quarter one to 34, 21 and 12, while the hosts carrying the consolidated pool rise from 6 to 9, 12 and 14. The estate ends the year with 26 machines instead of 54 — and the saving is only realised in the licensing and power columns when the retired hosts are actually removed from service rather than kept as spares.
Chart values are an illustrative planning model based on typical UMC engagements and published industry benchmarks. They are not a guarantee of results on any specific project.

Licensing and Cost: Where the Programme Pays for Itself

Licensing is where a virtualisation programme either pays for itself or quietly becomes more expensive than the estate it replaced. Three mechanisms drive the outcome, and all three are decided by the design rather than by the procurement team.

Cost lever How the design controls it Typical effect
Licensed core count Fewer, denser hosts carry the same workloads; core minimums per processor and per contract mean host count matters as much as utilisation Consolidation from 48 hosts to 14 can reduce licensed cores substantially — but only if the old hosts are retired
Feature tier Only the features actually operated are purchased; advanced scheduling and automation are worth paying for only where the team uses them Avoids paying for capabilities that sit unused, which is common after a platform change
Operating system and application licensing Virtual machine counts, affinity and host assignment reviewed against the software vendors’ own virtualisation terms Frequently larger than the hypervisor cost, and frequently overlooked
Support and subscription term Term length aligned to the refresh cycle rather than to the fiscal year Avoids paying for support on hosts that are being decommissioned
Decommissioning discipline Retired hosts removed from the entitlement record with evidence, not left on the asset register The saving that is most often lost, because nobody owns the paperwork
Growth headroom Headroom bought as a modelled quantity rather than as spare hardware Prevents an emergency purchase at list price in year three

We assemble the entitlement evidence as part of the programme rather than leaving it to renewal time: which cores are licensed, on which hosts, under which agreement, with the measurement that supports it. An organisation that cannot answer that question quickly is negotiating from a weak position every year, and the answer is a by-product of a properly run consolidation.

Operations After the Last Migration

A migration ends on the day the last physical host is switched off. An operating model starts the next day, and it is what determines whether the estate is still healthy in three years. The five disciplines below are the ones that lapse first, and each has a measurable signal that shows it has lapsed.

Discipline What good looks like The signal that it has lapsed
Patching and lifecycle A tested baseline rolled cluster by cluster on a published cadence, with evacuation capacity reserved for it Hosts running three different firmware baselines, and a cluster that cannot evacuate a host without overcommitting
Capacity management Weekly reporting in three units — CPU, memory and storage — with a forecast against the growth horizon Capacity discovered by a failed migration or a latency incident
Backup integration Every virtual machine in a backup policy by default, with restore tests scheduled and evidenced New machines provisioned outside the policy, and a restore test nobody can date
Monitoring thresholds Alerts derived from the design intent: admission control headroom, datastore latency, host memory pressure, snapshot age Alerts set at vendor defaults, so the estate runs red permanently and nobody reads it
Snapshot hygiene Snapshots treated as a maintenance tool with an expiry, not as a backup Snapshots older than the last quarterly review, consuming datastore capacity and degrading performance
Template and image control Gold images patched on a cadence and versioned New machines built from an image whose patch level nobody can state

The reporting we hand over is deliberately boring: host count, virtual machine count, utilisation against admission-control headroom, datastore free space, backup success, patch level per cluster, and the licensed core count. Six numbers, weekly, on one page. An estate whose health is described by six numbers that somebody reads is an estate that does not surprise anyone.

How This Differs From Our Neighbouring Pages

Our Computing, Server, Storage and Data Centre Engineering page covers the compute estate as a whole — servers, storage and the facility that houses them. This page is the virtualisation programme within it: the platform decision, the cluster design, the migration waves and the licensing exposure. If your question is what the estate should look like in three years, start there. If your question is which hypervisor to standardise on and how to move four hundred machines without stopping the business, this is the engagement.

Your question Right starting point What it produces
Which hypervisor should we standardise on, and what will it cost over five years? This page A measured platform recommendation with a costed migration programme
What should our whole compute estate look like in three years? Computing, Server, Storage and Data Centre Engineering An estate roadmap covering servers, storage, virtualisation and the facility
Our storage estate is the constraint, not the servers Computing, Server, Storage and Data Centre Engineering A storage architecture with tiering, replication and the capacity plan behind it
We want compute and storage converged into one platform Our hyperconverged infrastructure offering, built in this same programme A converged design with the failure-domain and capacity arithmetic modelled before purchase
Our recovery cannot meet the RTO the business agreed Backup, Disaster Recovery & Business Continuity Services A recovery architecture and the test regime that proves it
The workloads should leave our hosts entirely Cloud Services & Migration — AWS, Azure and Google Cloud A migration and landing-zone design, with the cost model compared against staying
Containers rather than virtual machines Kubernetes & Container Platform Engineering Services A container platform with the underlying virtual or bare-metal layer designed beneath it
Who runs the estate once it is built? Managed IT Infrastructure & IT AMC Services A managed service with published SLAs against the platform we designed

Virtualisation and container platforms are not alternatives at the level most enterprises operate at: containers run on something, and that something is usually a virtual machine on a cluster designed here. Getting the layer below right is what makes the layer above possible.

Evidence From Our Work

The engagements below are the closest evidence we can publish, and they show the two halves of a virtualisation programme that clients rarely buy together but always need: the consolidation and platform work, and the operations discipline that keeps it alive afterwards.

Frequently Asked Questions About Server Virtualisation

Should we stay on VMware or move to something else?

Neither answer is correct in the abstract, and any adviser who gives one without seeing your estate is selling something. The decision turns on four measurable things: the licensed core count your consolidated estate will actually carry, the features your team genuinely operates, the migration labour and risk of leaving, and whether your organisation can operate the alternative at three in the morning. We model all four over five years and give you a written recommendation with the arithmetic attached. Frequently the honest answer is to consolidate harder on the platform you already own, because the licensing saving comes from host count rather than from a change of vendor.

How many virtual machines can one host carry?

It is decided by memory and by failure tolerance, not by CPU. A host with 512 GB of memory and a measured average of 6 GB per machine could hold eighty, but the cluster must still absorb that host’s load when it fails, so the practical density is set by admission control and by the largest machines rather than the average. We size from measured demand including the worst week, and we publish the blast radius per cluster so the density decision is a funded choice rather than an accident.

What stays physical?

Usually fewer machines than expected, but the list is real: workloads with hardware dongles or licence terms tied to physical hardware, latency-sensitive systems that cannot tolerate hypervisor overhead, some high-frequency and real-time control applications, and anything whose vendor explicitly does not support virtualisation. We identify these in the assessment with a reason per machine, because “it has always been physical” is not a reason.

Can you migrate a clustered application without breaking quorum?

Yes, but not with a simple conversion. Clustered workloads are migrated as a unit with a documented sequence: the storage path is established first, nodes are moved one at a time with quorum preserved, replication is verified at each step, and failover is re-tested after the migration rather than assumed to still work. This is why clustered and stateful workloads are a separate wave with their own exit criteria.

What happens to our backups during and after migration?

Backup policy has to be re-attached deliberately, and this is where programmes leak. Machines migrated outside the normal provisioning path frequently arrive with no backup policy, or with a policy pointing at a datastore that no longer exists. We verify backup coverage per machine as a migration exit criterion, and we schedule an evidenced restore test on the new platform before the physical source is decommissioned.

How long does a consolidation programme take?

For a few hundred machines, three to nine months from assessment to decommission, with the assessment itself taking three to six weeks. The variables are dependency complexity, change control maturity and how many clustered workloads exist. Programmes that appear faster usually achieve it by skipping the assessment, which is where the surprises are stored.

Will virtualisation reduce our licensing cost?

It can, and it can also increase it. Under per-core subscription models, consolidation reduces cost only if hosts are genuinely retired — a migration that leaves the old hardware in service as spares reduces hardware cost and leaves licensing untouched. Operating system and application licensing often moves the other way, because virtual machine counts and affinity rules change the exposure. We model the whole position, not just the hypervisor line.

Do we need shared storage, or can we use local disks?

It depends on what you need live migration and HA to do. Shared storage or converged storage makes evacuation and restart straightforward; local storage with replication can work well for specific designs and reduces capital, but it constrains movement and changes the recovery model. We decide it from the workload’s latency requirement and recovery objective, and we document the consequence either way.

What utilisation should we target?

Typically 55 to 75 per cent of CPU across a consolidated cluster, with memory as the binding constraint and enough headroom to survive a host failure without overcommitting. Above about 80 per cent, a single failure produces latency rather than an outage, which is harder to diagnose and harder to explain to the business. Targets set from annual averages are meaningless; they have to be set against the measured peak week.

Can you virtualise our VDI estate as well?

VDI is a distinct design with its own storage and boot-storm characteristics, and we treat it as its own cluster rather than as more virtual machines. It is usually the largest cluster by machine count and the most latency-sensitive, so it gets its own failure domain, its own storage profile and its own capacity model.

What evidence do we get at the end?

The assessment report with measured utilisation and the dependency map, the platform recommendation with five-year cost model, the cluster design with published blast radius, the migration record per machine with verification and rollback evidence, the licensing position with entitlement evidence, the operating runbooks, and decommissioning certificates for every retired host including data sanitisation.

Is a hyperconverged platform a better answer than a traditional cluster?

Sometimes, and the decision is about operations rather than technology. Converged platforms simplify day-two work and reduce the number of vendors, at the cost of coupling compute and storage into one failure domain and one purchase. Where the estate is small or medium and the team is thin, converged usually wins. Where storage requirements are extreme or the estate is large and heterogeneous, a traditional cluster with separate storage often does. We model both and say which we would fund.

Start With a Measured Assessment, Not a Platform Brochure

A virtualisation decision made from a renewal notice is a decision made by somebody else. We will measure what you actually run, model the platform options against your core count and your team’s ability to operate them at three in the morning, and tell you whether the saving is in consolidation, in a platform change, or in both — including when the honest answer is that your current platform is fine and your host count is not.

Contact Unified Management Consulting to arrange a confidential first conversation.

Related Services

Virtualisation rarely stands alone. These are the engagements it most often runs alongside:

Leave a Reply

Your email address will not be published. Required fields are marked *