Payments survive peak load only if the network slice is reallocated while the surge is happening.

The problem

UPI traffic is not one workload. A secure transfer, an ATM withdrawal and a merchant payment tolerate different delays and fail with different consequences. A network that treats them the same will fail the one that matters most first.

What we built

We modelled all three transaction classes across the three 5G slice types (ultra-reliable low latency, enhanced broadband and massive machine-type) and pushed half a million concurrent transactions through to find the failure point.

Transaction class by 5G slice type: all 9 combinations modelled

Transaction class ↓ · 5G slice type →	URLLC	eMBB	mMTC
Secure transfer	Modelled	Modelled	Modelled
ATM withdrawal	Modelled	Modelled	Modelled
Merchant payment	Modelled	Modelled	Modelled

500,000+ concurrent transactions. Dynamic allocation held packet loss on the critical slice below 0.01% through peak.

Source: COEP TU course project, MATLAB, 2025

Static allocation collapses under surge. Weighting resource blocks dynamically toward the critical slice held packet loss below a hundredth of a percent through peak. We layered transmission, processing and waiting delay into one queuing model, so latency could be attributed to a cause and not just observed.

It was the first time my telecommunications work and my banking minor pointed at the same problem.