# Tips for Network Engineers
Colorful prose is fun, but colorful terminals save production.
## Paint Your CLI
Human vision > muscle memory. Assign a distinct background per saved session (black for backbone, gray prod, beige lab, whatever). Reloading the wrong box rn? Less likely when the prompt screams “LAB.”
## Capture **Everything**
Turn on full input/output logging in your terminal client. It’s your alibi when finger-pointing starts and a gold mine when future-you wonders, “how did I fix BGP flaps last March?”
## OSI Layer 1 Is Forever
AI, 400 Gbps, yada yada – copper still rots, optics still cough up CRCs. Start troubleshooting at Layer 1, then scoot up. Afaict most “mysteries” are bad cables with better PR teams.
## Fear The MTU Goblin
Mismatched MTU causes “works for email but not for $randomApp” syndrome. Ping end-to-end with DF bit set before crying in the ticket comments. Brutal, effective.
## Loopy Layer 2
Spanning-Tree meltdowns and rogue docks are Friday’s gift to network engineers. Debug by yanking uplinks until the storm calms, then autopsy the corpse. Elegance is overrated when the floor is dark.
## Trust Logs, Not Users
“Oh, nobody touched WAN1.” Sure, Jan. Parse the logs, find the unplug event, enjoy the awkward silence.
## No Friday Changes
Unless you adore panic Teams at 02:00. Schedule big hits when you’re awake tomorrow.
## Leverage Vendor Rollbacks
Cisco reload in, Juniper commit confirmed, FortiGate cfg-save: set them before poking prod devices remotely. Fat-finger safety nets are free uptime.
## External Brain
If you Google the same syntax twice, you’re the bug. Dump snippets into OneNote, Obsidian, plain text – idc – just version it and sync it.
## Docs Or Die
Store PDFs/HTML of every active device’s manuals somewhere searchable. Vendors vanish; your audit doesn’t care.
## Packet Literacy
Wireshark isn’t magic if you’ve never read TCP RFCs. Learn to read a three-way handshake; future outages will thank you.
## Ubiquitous Telemetry
NetFlow (or IPFIX) pointed at a collector keeps “the network is slow” arguments short. You can’t SSH into 900 switches fast enough.
## Backup Configs Religiously
Automated, versioned, diff-able. Whether rancid, Oxidized, or $CloudVendor, just get the running-config in Git every night and on every commit.
## Time Estimates Lie
Whatever downtime you predict, triple it. Management will still ask why it took “so long,” but at least you’ll be right.
Be real: none of this is novel. But applying even half will shave hours off your mean-time-to-facepalm.