Solveo ConsultingSolveo Consulting

Optimisation and Eco-Design

Reducing the footprint of an information system at constant service rests on four levers : the volume of resources, their utilisation rate, hardware efficiency, and its lifespan. We activate them from your own data, measuring impact in euros and in kgCO₂e, to anchor them durably in your design standards.

Schedule a conversationSee the other GreenOps domains

CHALLENGES

Engagement framework and macro challenges

Engagement scope

Dimensions and stakes addressed in the field

Sizing and waste elimination

The first savings seam lies in a rigorous cleanup of ghost resources: orphaned volumes, superfluous backups, inactive IP addresses, or test environments idle at night and over weekends. Once this purge is complete, we move on to resizing by analysing the high percentiles of memory and processor utilisation, thereby avoiding averages that mask load peaks and maximums that freeze over-provisioning. Finally, our recommendations clearly separate immediate, risk-free deletions from safety-margin reductions, which are submitted to your teams for arbitration.

Density and utilisation rate

An idle server already consumes half of its maximum energy, while bearing the entire carbon weight of its manufacturing. The priority is therefore to maximise the utilisation rate of your equipment. To achieve this, we consolidate your environments: precise tuning of container orchestration nodes, pooling of non-production spaces, and use of interruptible instances for flexible workloads. Above all, we configure genuine automatic scaling, based on real measurements rather than declared principles. Too often, capacity is allocated by mere habit: the gap between reserved resources and those actually consumed then reaches a factor of three. Yet it is this blindly provisioned capacity, not your actual usage, that weighs down your bill and your environmental footprint.

Hardware efficiency and architecture choices

Moving to more efficient processors, such as ARM architectures in the Cloud (Graviton, Cobalt, Axion), cuts consumption and hourly cost by 30 to 40 % compared with x86, depending on the recompilation effort tied to the language. On the storage side, the consumption gap between disk and flash varies with the access profile. We run these migrations by workload family, weighing the switching cost against the double financial and environmental gain.

Data life cycle

Stored data consumes energy continuously, exposing many overlooked sources of waste: logs kept indefinitely, unlimited backups, and massive duplication for test environments. To address this, we introduce strict retention policies, automate the move to cold storage, and optimise compression or deduplication. These actions drastically cut both the footprint and the bill, once compliance issues have been settled with your legal teams.

Application eco-design and artificial intelligence workloads

On the application side, efficiency rests on eliminating unnecessary work: removing redundant queries, strategic caching, and lighter data flows, validated against the General Reference Framework for the Eco-design of Digital Services (RGESN). For artificial intelligence, the challenge is to avoid overreach by favouring, when warranted, the lightest model and optimising its execution through quantisation, distillation, or request caching.

Service life, reuse and hardware end of life

The heaviest item in the balance of a device estate is its manufacturing: extending service life is therefore the dominant lever, far ahead of any usage optimisation. In practice this means decoupling the renewal policy from the accounting depreciation cycle, establishing real rather than declared obsolescence, treating maintainability and spare-part availability as purchasing criteria, and organising an internal reuse channel before devices leave the estate. Refurbishment and donation remain preferable to recycling, which recovers only a fraction of the materials and comes as a last resort. We formalise that policy and the matching purchasing criteria, including the repairability requirements to be carried into tenders.

METHODOLOGY

A pragmatic approach, matched to your level of maturity

  1. 01

    Map the potential

    Measure actual utilisation to identify waste, the capacity gap and the age of the hardware. This hierarchy avoids launching a programme on an under-used estate.

  2. 02

    Price every lever

    Assessment of each action in euros and in kgCO₂e, factoring in implementation and engineering cost. We name the overall impact explicitly, even when it contradicts the bill.

  3. 03

    Execute in waves

    First wave on risk-free removals, then technical resizing with the teams. Long-horizon architecture migrations and eco-design run in parallel.

  4. 04

    Anchor in the rules

    Decisions turned into enforceable criteria: retention policies, sizing standards and tender requirements, to prevent any relapse.

WORKED EXAMPLE

The real ranking of levers on a typical estate

An estate runs the equivalent of 1,000 x86 virtual processors, with an average utilisation rate of 25 % and hardware renewed every 4 years. The 4 levers below apply to the same perimeter. They are complementary, each requiring a different level of effort.

The order of execution is not the order of effect. Waste is dealt with first because it is risk-free and immediate. Hardware service life produces the largest gain but is decided elsewhere and measured in years. A credible roadmap carries both horizons at once.

DELIVERABLES

Sample deliverables

  • Inventory of established waste, priced in euros and in kgCO₂e, with a service-risk level per category

  • Utilisation analysis by workload family and measurement of the gap between reserved and consumed capacity

  • Ranking of levers by measured effect, engineering effort and risk, with a recommended execution order

  • Processor architecture migration study by workload family, migration cost set against the double gain

  • Retention and storage tiering policy by data type, validated with the legal teams

  • Default sizing reference and guardrails to be embedded in infrastructure templates

  • Eco-design criteria grid for architecture reviews and project gates, aligned with the RGESN framework

  • Hardware service-life policy and repairability criteria to be carried into tenders

  • Costed 3-year trajectory, separating secured gains, conditional gains and their prerequisites

KPIS

Steering indicators

  • Average utilisation rate of reserved capacity, by workload family

  • Gap between requested and consumed resources

  • Share of the estate running on a high energy efficiency architecture

  • Average hardware service life, compared with the accounting depreciation period

  • Share of data in immediate-access storage beyond the justified retention period

  • Unit intensity, in kgCO₂e and in kWh per business functional unit

FAQ

Frequently asked questions

Continue reading