What to Expect in the Hudson River Trading System Design Interview

Hudson River Trading does not run a classic web-scale system design interview. Its design questions concern the machine itself: operating systems, networking, memory, and concurrency, all in service of latency. Candidates report questions about lock-free data structures, CPU cache effects, and kernel bypass networking, especially for core developer roles. Latency means the time from an input arriving to the output leaving, and at HRT it is counted in microseconds. Prepare for systems depth, not for drawing load balancers.

How This Interview Differs

A typical product company asks you to design a service for millions of users. HRT asks how one machine, or a few, can react to a market event as fast as physics allows. The building blocks change accordingly:

• Threads and cores. Which thread does what, and which core it is pinned to. Pinning means fixing a thread to one core so it never migrates.
• Memory behavior. CPU caches serve data far faster than main memory. Designs that keep hot data small and contiguous win.
• Lock-free structures. A lock-free queue lets threads pass data without waiting on locks. The single-writer ring buffer is the classic example.
• Kernel bypass. Normal networking passes through the operating system kernel. Kernel bypass lets a program read packets directly from the network card, saving microseconds.

The Question Types

• Design a fast pipeline. For example, a service that consumes a high-rate data feed and reacts within a strict time budget.
• Diagnose a latency problem. Given a slow system, name your measurements in order and the likely causes.
• Concurrency correctness. Design a structure several threads share without corruption, and explain the trade-offs of your approach.
• Classic fundamentals. Virtual memory, context switches, and TCP behavior, asked directly. These support the design questions. A context switch is the operating system pausing one thread to run another, and its cost is a favorite topic.

A Signature Question: Design a Market Data Feed Handler

Here is a high-level pass at the most HRT-shaped design.

The problem. An exchange sends a stream of order book updates as network packets. An order book is the live list of buy and sell offers for an instrument. Your system must keep an accurate book and hand updates to a trading strategy, fast.

The path. Packets arrive at the network card and are read by kernel bypass, avoiding the operating system's slower path. One thread parses packets into normalized updates. It writes them to a single-writer ring buffer. The strategy thread reads from that buffer, pinned to its own core, polling continuously instead of sleeping.

Correctness. Feeds drop and reorder packets, so every message carries a sequence number. A gap in sequence numbers triggers a recovery request, and the book is marked stale until the gap closes. Never trade on a stale book.

Measurement. Timestamp at the wire, after parsing, and at the strategy. Report the 99th percentile, not the average, because in trading the worst case is what costs money. State this in the interview without being asked.

The follow-up questions. Expect "what happens at market open when volume spikes?" Answer: size buffers for the burst, and measure queue depth. Also expect "why not use locks?" Answer: a lock can make a thread wait for an unpredictable time. Removing that unpredictability is the whole point of the design. Say the cost too: lock-free code is harder to write and to prove correct.

How to Prepare

• Relearn your systems fundamentals. Threads, caches, and networking decide this interview. Grokking the System Design Interview gives the shared design vocabulary, and Grokking the Advanced System Design Interview treats concurrency and replication in more depth.
• Learn one lock-free structure well. The single-writer ring buffer is small enough to code from memory and rich enough to discuss for an hour.
• Practice thinking in percentiles. For any system you have built, learn its 50th, 99th, and worst-case latency. Then practice explaining the differences.
• Connect design to the rest of the loop. The same depth is tested in stories and fundamentals rounds. See What is the Hudson River Trading interview process like? and Top Hudson River Trading behavioral interview questions.