SRE interview prep · rebuilt from the ground up
Don’t memorize answers.
Derive them.
SRE interviews reward people who can reason from fundamentals — queueing, probability, the kernel, the network, the hardware. Every page here starts with one principle, lets you play with it, and then shows you exactly how it shows up in the interview room.
5 pillars · 33 principles · interactive visuals, no paywall
The method
Every topic, four moves.
Understanding first, interview performance second — because the second falls out of the first.
One sentence
The idea stated plainly enough to say out loud in an interview.
Play with it
An interactive model you can push until it breaks — that’s where intuition comes from.
Where it shows up
The prompts, follow-ups, and signals interviewers are actually listening for.
A strong answer, dissected
A sample question with a strong answer broken into beats — and why each beat lands.
Start here
Two flagship principles, fully built.
Queueing & Little’s Law
Why a service at 90% utilization is not “10% away from trouble” — it’s already in it.
Open the interactive →AI hardware · flagshipRoofline & failure at scale
Why GPUs sit idle waiting on memory — and why a 4,096-GPU job is only as healthy as its worst node.
Open the interactive →The five pillars
The whole interview loop, decomposed.
Networking
Every request is a chain of round trips.
Packets & layers · DNS resolution · TCP handshake & congestion · TLS · L4 vs L7 load balancing · Latency budgets
02 / SYSSystems
The kernel is the platform you’re actually running on.
Processes vs threads · Scheduling · Virtual memory · File descriptors · Syscalls · The I/O stack · cgroups & namespaces
03 / DISTDistributed Systems
Everything fails, and it fails partially.
Queueing & Little’s Law · Availability math · Replication & quorums · Consensus (Raft) · Consistency models · Partitions · Retries & cascades
04 / CODECoding
Practical code, under real constraints.
Complexity from first principles · Core data structures · Concurrency & races · Log parsing · Rate limiters · LRU cache
05 / AI/HWAI Hardware
Reliability at the scale of thousands of GPUs.
The roofline model · GPU vs CPU · HBM & the memory wall · NVLink & InfiniBand · Training vs inference · Data, tensor & pipeline parallelism · Failure at cluster scale & checkpointing