About Server Sizing

Server sizing is a critical skill for SRE work. Four parameters matter roughly equally when sizing virtualized servers.

Memory

Check memory first — it typically depletes before other resources.

• Install the maximum RAM the physical machine supports.
• Allocate minimal memory to virtual machines.
• Avoid memory overcommitment, which forces reliance on swap and degrades performance.

CPU

• Physical machines typically have 4–24 cores for servers.
• A single core supports roughly 4 VMs, ranging from 1–2 to 8–10 VMs per core depending on workload.
• Reserve at least one additional core beyond the maximum vCPU assigned to any single VM.
• Each core typically handles 2 threads (some specialized processors support 4–8).

As a rule of thumb: scaling from 4 vCPUs handling 2,000 requests/second to 5,000 requests/second requires roughly 10 vCPUs.

Network

Sizing depends on desired redundancy:

• Minimum of 4 network interface cards (NICs).
• Additional NICs for network storage scenarios.
• Manage Quality of Service (QoS) for bandwidth, throughput, jitter, packet loss, and latency.

Storage

• Virtual disks maintain their original formats.
• SSDs outperform HDDs for performance-critical applications.
• Allocate total VM storage plus 10–20% for system files and snapshots.
• Use at most 80% of physical HDD capacity, reserving 20% as buffer.