Virtual Machines & Compute

How it works

From image to running fleet

A healthy fleet starts with a golden image: a patched operating system with your agents baked in, rebuilt on a schedule rather than patched by hand. A launch template pins that image together with the machine size, disks and network, so every VM is identical.

A scaling group then keeps between a minimum and maximum number of VMs alive, spread across zones, and replaces any that fail health checks. An autoscaler adds or removes VMs from metrics such as CPU, request rate or queue depth, and a load balancer spreads traffic across whatever is running.

Paying less for the same capacity

The same VM has several prices. On demand is full flexibility at list price. Commitments (Savings Plans, Azure savings plans and reservations, Google committed use discounts) trade a one- or three-year promise for large discounts. Spot capacity is the cheapest of all, up to about 90%, but can be reclaimed at short notice.

The winning pattern combines them: commit to the baseline that always runs, autoscale on demand for predictable peaks, and put interruptible work on Spot. Arm processors (AWS Graviton, Azure Cobalt, Google Axion) often add further savings for Linux workloads.

AWS, Azure and Google Cloud compared

What makes each cloud different

Design decisions that matter

1. 01Bake images, do not patch servers

   Rebuild golden images on a schedule and roll them through the scaling group. Servers stay identical and patching becomes a deployment.

2. 02Spread every fleet across zones

   Set scaling groups to at least two zones, ideally three, before the first production deploy.

3. 03Size from data

   Start smaller than you think and use the provider's right-sizing tool after two weeks of real traffic.

4. 04Layer your purchasing

   Commit to the steady baseline, keep peaks on demand and move retryable work to Spot.

5. 05Test Arm early

   Build multi-architecture images so moving to Graviton, Cobalt or Axion is a configuration change, not a project.

Common mistakes we see

• Sizing from the old data center. On-premises servers were sized for five-year peaks; cloud VMs should be sized for today and scaled out.
• Committing before right-sizing. Discounts on oversized VMs lock in waste for one to three years.
• Spot for stateful work. Without checkpoints or a queue, a reclaim notice means lost work and failed jobs.
• Forgetting non-production. Dev and test fleets left running nights and weekends often cost as much as production.

A practical 30-day plan

1. Week 1

   Inventory running VMs, utilization and purchase options; tag owners.

2. Week 2

   Right-size the largest 20 workloads and schedule non-production shutdowns.

3. Week 3

   Move suitable workloads to scaling groups across zones; trial Arm and Spot.

4. Week 4

   Buy commitments for the proven baseline and set budgets and alerts.

How NTS helps

NTS reviews VM fleets on AWS, Azure and Google Cloud and turns the findings into action: right-sizing, golden-image pipelines, zone-spread scaling groups, Arm migrations and a commitment strategy that matches real usage. Most clients see the savings within the first billing cycle.

How we work with you

1. 1Assess

   A focused review of your compute and virtual machine estate across AWS, Azure and Google Cloud. We look for the gaps we find most often: sizing from the old data center, committing before right-sizing, Spot for stateful work and forgetting non-production.

2. 2Design

   A target architecture and roadmap built on the decisions that matter most: bake images, do not patch servers, spread every fleet across zones, size from data, layer your purchasing and test Arm early.

3. 3Build

   Hands-on delivery alongside your team. We inventory running VMs, utilization and purchase options; tag owners, right-size the largest 20 workloads and schedule non-production shutdowns, move suitable workloads to scaling groups across zones; trial Arm and Spot, and finally buy commitments for the proven baseline and set budgets and alerts.

4. 4Run and improve

   Quarterly right-sizing and commitment reviews keep your compute estate efficient as workloads and instance families change.

Why teams choose NTS for compute and virtual machine estate

• Performance and price together

  We size for real workload behavior, so you stop paying for idle capacity without risking slowdowns.

• Latest chip generations

  We test Graviton, Cobalt and Axion Arm processors against your workloads and move what benefits.

• Immutable by design

  Servers are rebuilt from images, not patched by hand, which makes every environment repeatable.