Save up to 30% CPU capacity

The IT market continues to experience rising prices due to global shortages of critical hardware components such as CPUs and RAM. At the same time, software vendors are increasing licensing fees to offset higher development and support costs. As a result, many organizations are paying significantly more today – despite the fact that their environments often run far below optimal efficiency.

Targeted CPU optimization helps to counter these market trends and unlocks measurable cost reductions.

Initial situation

A mid‑sized industrial company was preparing for an upcoming tech refresh. Their existing VMware environment had been in operation for five years and support extensions were becoming increasingly expensive. At the same time, the market confronted them with substantially higher prices for new servers and shipping times stretching over several months.

When multiple business units also complained about slow applications, pressure grew to accelerate host replacement. Yet internal monitoring showed an average CPU utilization of only ~40% - seemingly “within normal range.” Still, performance issues persisted.

CPU optimization in VMware ESXi is complex because multiple technical factors interact:

1. Lack of historical utilization insights
   vSphere offers only limited long‑term visibility across systems, making trend comparison difficult.
2. CPU overcommitment
   ESXi allows provisioning for far more vCPUs than physical cores – but overcommitment quickly results in CPU Ready spikes and performance degradation.
3. Missing NUMA awareness
   Multi‑socket VMs stretched across NUMA nodes introduce unnecessary latency.
4. Oversized VMs
   VMs equipped with excessive vCPUs consume resources without providing measurable benefits.
5. Low utilization despite large CPU allocations
   Many VMs operate consistently below 20% CPU load while blocking scheduling capacity.

Using BVQ°’s in‑depth analytics, clear and actionable recommendations were generated to realign the VMware environment for maximum efficiency. 
 

How BVQ° provides immediate clarity

• cross‑comparison of all relevant VMs and hosts at a glance
• identification of VMs running permanently below 5%, 20% and 40% CPU load
• detection of VMs misaligned with the current NUMA topology
• trend analysis of CPU Ready Time, Co‑STOP values, and scheduling inefficiency
• concrete insights into where vCPU reduction improves performance
• quantified savings potential and how to achieve it

BVQ° showed within just a few minutes that the problem was not missing hardware, but inefficient resource usage.

The following weaknesses were identified:

• 24 VMs with less than 5% actual CPU load but equipped with 8 vCPUs or more
• approx. 20 TiB of storage wasted in datastores because VMs are set to “thick”
• approx. 15 TiB of storage wasted on NetApp because volumes are “thick” and efficiency is not enabled
• around 1 TB of RAM overprovisioned
• high CPU Ready Time on several systems – caused by massive overprovisioning
• multiple NUMA violations significantly slowing down data‑intensive applications
• uneven workload distribution on a host, even though capacity was available in the cluster
• vCPU reservation configured although it is not being used

To avoid unnecessary hardware and licensing costs, the customer opted for targeted optimization rather than an early tech refresh. Measures included:

• right‑sizing oversized VMs to realistic vCPU allocations
• adjusting VM vSockets for proper NUMA alignment
• rebalancing host load through targeted VM migrations
• implementing clear guidelines for future vCPU sizing
• aligning vCPU reservations with actual usage patterns

Immediate Impact: Results After Just Two Days

The previously planned - and costly - tech refresh was postponed. Still, the customer achieved:

• 35% faster application response times
• significantly reduced CPU Ready Time
• more available resources without purchasing new hosts
• a more stable, fully transparent VMware environment

1. Automated detection of inefficient vCPU allocations
   → Identifies VMs with disproportionate CPU assignments and quantifies performance impact.
2. Visibility into overcommitment & bottlenecks
   → Clear overview of pCPU‑to‑vCPU ratios per host, including threshold analytics.
3. NUMA topology analysis
   → Highlights VMs spanning multiple NUMA nodes and the latency this introduces.
4. Actionable optimization guidance
   → Concrete recommendations for vCPU reduction, host load balancing and efficient VM sizing.

This case study showcases BVQ°’s core strength: within a short deep‑dive analysis, the tool uncovered that insufficient transparency – not insufficient hardware – was the real issue. BVQ° accurately identified oversized VMs, unnecessary Ready Time and NUMA misalignments. The targeted optimizations prevented costly infrastructure expansion and enabled the company to reclaim up to 30% CPU resources – delivering measurable improvements in performance, stability and cost efficiency.

• increased performance without new hardware
• more stable applications through reduced latency
• less troubleshooting thanks to transparent data
• optimal CPU allocation for every VM
• automated alerts for configuration issues
• better planning confidence and lower operational costs
• from installation to first results in just 2 hours