🧠 Cores, Threads, and vCPUs: The Untold Story of CPUs in Linux, Docker, and Kubernetes

🧠 Cores, Threads, and vCPUs: The Untold Story of CPUs in Linux, Docker, and Kubernetes

“I gave my Docker container 2 CPUs — why is it still slow?”

“Why does Kubernetes use millicores instead of cores?”

“Why does htop show 8 CPUs when my laptop has 4?”

Welcome to the chaos.

Most developers have a vague sense that CPUs are “doing work” — somewhere down there, deep in the machine. But when it comes time to set CPU limits on Docker containers or request compute in Kubernetes clusters, that vagueness becomes a real liability.

If you want to level up as a software engineer, platform engineer, or backend specialist — you need CPU literacy.

This blog will demystify CPUs from the bare metal up — from physical cores and hyperthreading, to virtual CPUs, millicores, and how Linux, Docker, and Kubernetes interpret them.

🟩 1. The Developer’s Dilemma: How Many CPUs Do You Really Have?

Let’s say you’re on a Linux system and you run:

$ lscpu | grep "^CPU(s):"CPU(s): 8

Looks like you’ve got 8 CPUs. But does that mean 8 physical processors? 8 cores? Or something else?

Let’s break this down before we start throwing around terms like vCPU, millicore, and CPU-quota.

🟨 2. The Physical Reality: CPU, Core, Thread

🔧 CPU (Central Processing Unit)

The brain of the computer. It executes machine instructions.

🧩 Core

A single compute unit inside a CPU. Think of it as a miniature processor within the processor.

• Adual-core CPUhas 2 cores.
• Aquad-core CPUhas 4 cores.
• Each core can (usually) run one instruction stream (thread) at a time.

🧵 Hyperthreading

Intel’s marketing name forSimultaneous Multithreading (SMT).

• Each core presentstwological execution contexts (called threads).
• One core = 2logical CPUsto the OS.

Thus, your 4-core laptop with hyperthreading shows up as:

8 CPUsin Linux.

That’s not a lie — it’s ascheduler trick.

🧠 Linux View:

$ nproc8$ lscpu | grep "Thread|Core|Socket"Thread(s) per core: 2Core(s) per socket: 4Socket(s): 1

So now you know:8 logical CPUs = 4 physical cores × 2 hyperthreads.

🧠 Visual Example: Single Socket System

Example: Intel Xeon with 1 socket, 4 cores, Hyper-Threading enabled

Socket 0 ├── Core 0 │ ├── Logical CPU 0 │ └── Logical CPU 1 ├── Core 1 │ ├── Logical CPU 2 │ └── Logical CPU 3 ├── Core 2 │ ├── Logical CPU 4 │ └── Logical CPU 5 └── Core 3 ├── Logical CPU 6 └── Logical CPU 7

🟦 3. vCPU: The Illusion of Compute

☁️ What is a vCPU?

AvCPU (virtual CPU)is the compute unit exposed by virtualization technologies like:

• VMware
• Hyper-V
• KVM (used by cloud providers like AWS/GCP)

Rule of thumb:

1 vCPU = 1 hyperthread = ½ physical core

In cloud platforms like AWS EC2 or GCP, when they say “2 vCPUs”, you’re usually getting2 hyperthreads, possibly on the same physical core.

That means:

• You’re sharing the silicon with another user
• You’re time-sliced and isolated by the hypervisor

🟧 4. Docker CPU Limits: The Cgroups Game

Let’s say you run:

docker run --cpus=2 my-app

What happens?

You’re not “getting” 2 CPUs. You’re telling Linux to usecgroupstothrottlethis container’s CPUtime slice.

Under the hood:

cpu.cfs_quota_us = 200000cpu.cfs_period_us = 100000

→ This tells the Linux scheduler:

This container can use up to 200ms of CPU time per 100ms period — i.e., 2 cores worth.

Want to pin containers to specific cores?

docker run --cpuset-cpus="0,1" ...

That’s hard binding — the container can only run on CPU 0 and 1.

💡Key Insight: CPU quotas limithow much, notwherethe process runs.

🟪 5. Kubernetes: Millicores and Scheduling Realities

In Kubernetes, CPU limits and requests are expressed inmillicores.

1000m = 1 vCPU = 1 logical core

So this:

resources: requests: cpu: "500m"

Means:

“Please schedule me on a node where I can get at least50% of one CPU’s time.”

Kubernetes doesn’t slice physical cores. It usescgroupsand theLinux Completely Fair Scheduler (CFS)tothrottle CPU usagebased on time.

That’s why:

• CPU limits protect the node from noisy neighbors
• CPUrequestshelp the scheduler make bin-packing decisions

But — and this is important —there’s no hard CPU isolation unless you use CPU pinning(via static CPU Manager policies).

🎯 Millicore ≠ Microprocessor

It’s aunit of time, not hardware. Think in terms ofquota, notchip.

🟥 6. How CPUs Actually Run Threads

Most developers think threads are executed by cores directly.

Wrong.

Here’s the reality:

• Your Java app creates 10 threads.
• The OS scheduler decides which thread gets time on which core.
• Context switching occurs at nanosecond precision.
• Two threads might beinterleavedon the same core, ormigratedacross cores.

🔄 Concurrency vs Parallelism

• Concurrency: Multiple tasksin progress(via context switching)
• Parallelism: Multiple tasksexecuting simultaneously(requires multiple cores)

Example:

• Your 4-core machine runs 40 threads → concurrent, not fully parallel.
• 4 cores can onlytrulyrun 4 threads at a time.

Everything else isillusion + scheduler magic.

🟫 7. Real-World Scenarios: Now You Know

🧪Your container is slow, even with 2 CPUs?

→ You’re CPU throttled. Check cpu-quota. You’re likely using 2 cores’ worth oftime, not dedicated compute.

🧪Your K8s pod is crashing under load?

→ You hit your CPUlimit, and the OS started throttling or evicting your pod.

🧪top shows 100% CPU for your process?

→ That’s 1 full logical core’s worth of time. On a 4-core, 8-thread system, that’s12.5% of total machine capacity.

🟩 8. Closing Thoughts: CPU Literacy Is Dev Maturity

Whether you’re shipping containerized microservices or debugging JVM threads in production —understanding CPUs is leverage.

It helps you:

• Avoid overprovisioning or underprovisioning compute
• Diagnose performance issues accurately
• Speak confidently in system design interviews

Next time someone says “give the container 1 CPU”, ask:

“Do you mean onephysical core, onehyperthread, or1 vCPU’s worth of time?”

Because now, you know the difference.

🔖 Further Reading & Tools

• lscpu— Check system CPU info
• htop— Interactive CPU and process monitor
• taskset— Pin process to CPU cores
• cgroups— Resource limits in Linux
• kubectl top— View CPU usage in Kubernetes

🛠️ Show Your Support

If this post brought you clarity, saved you hours of Googling, or challenged the way you think:

• 👏Clapto support the effort (you can hit it up to 50 times on Medium).
• 🔁Shareit with a fellow engineer or curious mind.
• 💬Commentwith questions, feedback, or requests — I read every one.
• 📩Requesta topic you’d like covered next.
• ⭐Followto stay ahead as new deep-dive posts drop.

Let’s build real engineering wisdom — not trivia.