Networking Interview Questions & Answers

An accidental physical loop lets a broadcast circle endlessly between two paths, multiplying itself into a storm that can bring a whole segment down. Spanning Tree Protocol exists specifically to detect that kind of redundant path and deliberately block one of the looping ports before it ever becomes a problem.

EIGRP's DUAL algorithm won't just accept any alternate path as a usable backup, it specifically requires that path's reported distance to be strictly better than the current best path's own feasible distance. That rule is what guarantees backups are always loop-free, but it also means a redundant physical path can exist and still not qualify as a usable backup route.

HSRP and VRRP both work the same fundamental way: only ONE router in the group is actively forwarding traffic at any moment, the other is just standing by in case it fails. GLBP was built specifically to break that pattern, letting multiple routers all forward traffic simultaneously by handing different clients different virtual MAC addresses, giving you real load sharing instead of just a backup.

Ping only needs one single reply, from the final destination, to succeed. Traceroute needs a reply from EVERY router along the way to build its full picture, and if any of those routers are configured to ignore or rate-limit that specific kind of reply, traceroute shows gaps or timeouts even though the destination itself is perfectly reachable.

Connection draining, also called a deregistration delay, gives an instance being removed from the pool a grace period to finish requests already in progress before it is forcibly cut off, instead of killing its connections the instant it leaves rotation. Passing health checks only confirms an instance is currently healthy, they say nothing about how its removal is handled. A shorter health check interval would only detect an unhealthy instance faster, it wouldn't protect requests already in flight when a healthy instance is deliberately taken out of rotation.

STUN lets each client discover its own public IP and port as seen from outside its NAT, which both sides can then exchange, often via a signaling server, to attempt hole punching, where both send packets simultaneously so each NAT sees outbound traffic and permits the matching return path. When NAT behavior is too restrictive for hole punching to work, such as symmetric NAT, the connection falls back to relaying all traffic through a TURN server instead. MASQUERADE is just the mechanism a NAT itself uses to rewrite outbound source addresses, it doesn't help two separate NATs punch through to each other.

Ping only needs one single reply, from the final destination, to succeed. Traceroute needs a reply from EVERY router along the way to build its full picture, and if any of those routers are configured to ignore or rate-limit that specific kind of reply, traceroute shows gaps or timeouts even though the destination itself is perfectly reachable.

HTTP/2 multiplexing allows many independent request and response streams to be interleaved over ONE TCP connection at the same time, removing the need to open multiple connections, and wait for a free slot, the way HTTP/1.1 does. Header compression via HPACK is a real HTTP/2 feature too and does reduce overhead, but it addresses request size, not the connection-per-host bottleneck. Server push is also real but is about proactively sending resources, unrelated to this queuing problem.

Marking traffic with DSCP EF is like putting a 'priority' sticker on a package, it doesn't actually do anything by itself. Something on the congested link still has to actually READ that sticker and act on it by prioritizing that traffic in its queue, and without a queuing policy configured to do exactly that, correctly-marked traffic gets treated no differently than anything else once the link is busy.

An IP address alone doesn't tell you which part identifies the network and which part identifies the specific device on it, that's exactly the job of the subnet mask. It draws the line between 'network' and 'host' within the address, which is how two devices figure out if they're on the same local network or need a router in between.