Network brain — choosing the
best route for packet data
Every time your phone loads something, a decision gets made hundreds of times a second: which way should this packet go? I spent twenty years in telecom — a lot of it at Nokia — building the systems that answer it. Below is the real journey that packet data takes across a modern 5G/LTE network: UE → RU → DU/CU → cell-site router → backbone → UPF → CDN. The "network brain" is what scores every route by latency, throughput, jitter and loss and forwards traffic down the best one. Pick a scenario, watch the telemetry, then break something and find the fault.
A phone streaming video. The brain routes packet data across the RAN, over IP/MPLS backhaul, through the UPF and out to a CDN — picking the lowest-cost backbone path in real time.
best route —·~0 ms end-to-end
The topology, node counts and telemetry values are an accurate but illustrative model — a legible stand-in sized for a browser, not a capture of any production network. What's real is the routing: the "best route" is a live shortest-path computation over current link costs (latency, load, loss), and it genuinely recomputes the instant you inject a fault or apply a fix.
The real signal path
Not a random mesh — the actual O-RAN split (RU/DU/CU, the disaggregated BBU), fronthaul over eCPRI, midhaul over F1, IP/MPLS backhaul on virtualized routers (NFV), and the 5G Core. The UPF forwards the user-plane packets; AMF/SMF are control plane — they set the session up and steer it, but never carry your data.
Observability — see every hop
Each hop reports the KPIs that matter for its layer — RAN: RSRP/SINR/PRB; transport: latency/jitter/loss/link-load; core: UPF load and sessions; CDN: cache-hit ratio. Flip Telemetry OFF and a failing node looks exactly like a healthy one. That blind spot is the whole argument for observability.
Find the fault, then fix it
Inject a fault — fronthaul down, backbone congestion, UPF overload, satellite rain-fade — and the KPIs degrade at and after the faulty hop, so you can pinpoint where the break is. Then remediate: reroute onto a redundant path, fail over to a standby UPF, or lean on local breakout. Watch the brain choose the new best route.
Three journeys, one core
A phone streaming video. Packets ride the RAN, cross IP/MPLS backhaul on virtual routers, anchor at the UPF and reach a CDN. When the primary backbone congests, IP/MPLS reroutes onto the redundant path in real time.
A device with no terrestrial signal reaches a LEO satellite — direct-to-cell access and a Ka-band feeder into a ground gateway, then satellite backhaul into the same core. Honestly higher latency, and exposed to rain fade you can watch it ride out.
A power utility runs its own RAN and an on-prem UPF that does local breakout — SCADA control stays on-site at single-digit-ms latency and keeps running even when the WAN to the cloud is cut.
Why call it a brain
It's a light analogy, not biomimicry. A brain and a telecom network face the same problem — keep full visibility of a huge distributed system and route signals along the best available path, with no one part in charge of the whole thing. Four properties they genuinely share are exactly what network engineers design for.
BrainA brain constantly senses its own state — it can't act on what it can't perceive.
NetworkObservability instruments every hop (RSRP, latency, loss, UPF load) so a failing node is visible, not guessed at.
BrainSignals take the path that works, weighing cost and delay — not a single hard-wired track.
NetworkThe brain scores every route by latency, throughput, jitter and loss, and forwards packet data along the cheapest one.
BrainNo single neuron is in charge; control is spread across the network.
NetworkNo master switch. Control-plane functions (AMF/SMF) set sessions up; forwarding decisions live in the routers and UPF.
BrainLose a pathway and signals re-route; function degrades gracefully instead of stopping.
NetworkCut a link or overload a node and the brain reroutes / fails over onto a redundant path — an outage becomes a detour.
A network that really does route without a boss
The interactive above is my own diagram of a telecom network. This next panel is the system that inspired the framing: the real, unedited FAFB electron-microscopy volume of a fruit-fly brain — streamed straight from public storage into Neuroglancer, the open-source viewer Google built for connectomics. A distributed network with no central controller, no master switch, and graceful degradation baked in. Sound familiar?
This panel is not a drawing — it is the real connectome substrate rendered in Neuroglancer, Google's open-source WebGL viewer for petabyte-scale connectomics. Scroll to move through the tissue; drag to pan.
Open full-screen ↗What's real here — and what's illustrative
- AccurateThe architecture: the node types and segment names (RU/DU/CU/BBU, fronthaul/midhaul/backhaul, CSR, vRouter/NFV, UPF with AMF/SMF as control plane, NTN and private-LTE breakout) reflect how real 5G/LTE networks are actually built.
- RealThe routing: the "best route" is a live shortest-path computation over current link costs, and it genuinely recomputes the instant you inject a fault or apply a fix.
- IllustrativeThe topology size, node counts and telemetry values are a legible stand-in sized for a browser — not a capture of any production network or a specific vendor's numbers.
Connectome panel & open tools: the FAFB fly-brain EM volume in Neuroglancer; FlyWire whole-brain connectome (Dorkenwald et al., Nature, 2024) · nature.com.
The same decision layer at planetary scale: real cities, submarine-cable routes and a LEO/NTN constellation, with live Dijkstra best-path selection you can steer.
The signal-cascade interactive this network piece grew out of — watch a smell become a memory, neuron by neuron.
My AI/tech consultancy takes these ideas out of the demo and into production telco, satellite/NTN, and enterprise networks.
I wrote about the whole signal journey — and how observability turns a five-hour outage into a five-minute fix — in this post. For the science that started it, we can map a whole brain now.