🧩 At Scale, Time Is Your Strongest (or Weakest) Link

When your system hits 1 million TPS, milliseconds aren’t enough.

Every microservice, log, and transaction wants to know: "What happened, exactly when?"

And if you're still timestamping in milliseconds — congrats — you just grouped 1,000 transactions under the same time.
That’s a challenge for:

• 🔁 Ordering

• 🔍 Auditing

• 🧩 Debugging

• 📦 Replication

• 🧠 Observability

Let’s talk timestamp granularity, how to implement it right, and where the real cost of precision lives. Spoiler: some platforms make this a lot easier than others — and System Z might already have it built in.

🧠 How Precise Is Precise Enough?

At 1M TPS:

🎯 Goal: Microsecond granularity (µs) — minimum for 1M TPS
🛠 Bonus: Nanosecond (ns) granularity — ideal in multi-threaded or distributed systems

⚙️ How You Implement It Depends on Your Platform

On IBM Z (z/OS / LinuxONE):

• Hardware-level time: STCK, STCKE, ETOD

• Precision: Micro or nano, monotonic across CPUs and LPARs

• Available natively in COBOL, PL/I, Java, Db2, CICS, SMF

• No additional tooling required

On x86 / Kubernetes:

• OS/syscall based: clock_gettime() or app frameworks

• Consistency depends on time sync (NTP, Chrony)

• High precision often means custom tooling, sidecars, or observability layers

💸 Let’s Talk Cost

Precision has trade-offs — in storage, performance, and operational overhead.

Here’s a side-by-side look:

📌 Worth noting: human-friendly formats like ISO 8601 are great for parsing — but can be surprisingly large in memory and storage. More bytes, more parsing effort, more overhead.

🧾 TL;DR: In many platforms, precision is something you plan for.
On IBM Z, it’s something you inherit.

🧠 So What?

• If you're building for scale — microservices, APIs, financial apps — precision isn’t optional.

• If you're running hybrid or cloud-native workloads, time precision can be a challenge.

• If you're on System Z? You may already have what others are still engineering toward.

Not everything in the mainframe world is "legacy." Some of it is just already solved.

✍️ Final Thought

Whether you're building systems of record or systems of engagement, time is a shared truth. It’s how we order events, trace causality, and ensure consistency.

So ask yourself:
📉 Is your platform helping or hindering precision?
🛠️ Are you building around time — or building on top of it?

Written by Henri Kuiper