Please hold while we process the request.

Five Things Nobody Tells You About Building a Carrier-Grade O-RAN Testbed

Five Things Nobody Tells You About Building a Carrier-Grade O-RAN Testbed

Validating O-RAN at real carrier density is hard. Not just technically — but in ways that don’t show up in architecture diagrams or spec sheets.

Simnovus together with Dell Technologies, recently published a whitepaper documenting what it actually took to simultaneously emulate 8 cells and 1,024 UEs on a single commercial off-the-shelf Dell PowerEdge R7725 — achieving zero-drop 25G fronthaul O-RAN 7.2 split emulation across 240 fully virtualized, isolated cores. Here are five things we achieved.

1. Your CPU governor is working against you — and you won’t know it.

Standard DPDK tuning advice tells you to disable CPU sleep states. On modern AMD Zen architectures, that single decision can quietly throttle your compute to less than a third of its rated speed. The fix exists, but it’s not obvious — and the whitepaper walks through exactly what to change and why.

2. An 8-core boundary can make or break your slot timing.

EPYC Gen5 processors partition L3 cache into fixed 8-core blocks. Cross that boundary with a workload, and you introduce micro-stalls that are individually tiny but collectively enough to miss OFDM symbol windows. VM topology isn’t just a resource question — it’s a precision question.

3. 1 TB/s of memory bandwidth doesn’t mean 1 TB/s of throughput.

We measured roughly 307 GB/s at full saturation on hardware theoretically capable of far more. The bottleneck wasn’t the DRAM — it was the memory controllers, saturated by millions of fragmented micro-transactions. Getting past it required simultaneous changes at the BIOS, DPDK, and application layer.

4. NTP will silently destroy your slot timing.

Under the interrupt load of hundreds of cores polling at full speed, standard virtual clocks drift in ways that only become visible when fronthaul frames start dropping. Clock architecture needs to be a hardware-first decision, not an afterthought.

5. All of this can be solved on COTS hardware.

That’s perhaps the most important finding. Carrier-grade O-RAN 7.2 determinism — including sub-100 nanosecond synchronization stability — is achievable on a standard server, without proprietary test appliances, when software design is aligned with the physical realities of modern silicon.

 

The whitepaper documents the exact configurations, tuning parameters, and architectural decisions behind each of these findings — a validated blueprint for teams building high-density O-RAN emulation environments.

Download the whitepaper →

Simnovus builds high-scale 5G network testing and simulation tools. Our UE Simulators, O-RU Simulators, and Network Emulators are used by telecom operators, equipment vendors, and research organizations worldwide.