← Optiver Interview Insights

Optiver·Software Engineer·Onsite - System Design / Architecture·Senior

SeniorPrefer not to say
Apr 2026

Summary

Optiver system design round that went way deeper than I expected. Started with 'design the network for a trading firm' and two hours later I was explaining where individual microseconds disappear in the order-flow path. Brutal but genuinely interesting if you're into low-latency infrastructure.

Questions Asked (3)

Q1

Design the full network and infrastructure for a trading firm that has servers co-located next to an exchange and traders working from a remote office. Walk through physical wiring choices, bandwidth and latency requirements, server hardware and OS tuning, kernel-bypass networking, time synchronization, redundancy and failover, and the complete order-flow path end-to-end.

System DesignTechnical Trade-offs
Author's notes

This was the whole interview basically.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Structure your answer around the critical path from trader to exchange, emphasizing latency minimization at every layer. Start with the physical and network topology, then dive into server hardware/OS tuning and kernel-bypass, and finish with redundancy and time synchronization. Quantify latency budgets and justify trade-offs (e.g., cost vs. speed) to show engineering maturity.

Pro tip: Always tie design choices back to business impact: e.g., 'A 1µs reduction in order latency can translate to millions in alpha capture.' This shows you understand the domain, not just the tech.

1. Map the end-to-end order flow

Describe the path: trader's order entry → remote office network → WAN → colocation → trading server → exchange gateway. Identify where latency is introduced and which segments are most critical.

2. Design physical and network topology

Choose fiber vs. copper, switch hierarchy (leaf-spine), and WAN options (dedicated fiber, microwave). Specify bandwidth (e.g., 10/40/100GbE) and latency targets (e.g., sub-100µs colo, sub-1ms WAN).

3. Optimize server hardware and OS

Select low-latency CPUs (high clock, large cache), NVMe storage, and NICs with kernel-bypass (e.g., Solarflare, DPDK). Tune OS: CPU pinning, hugepages, busy-polling, disable interrupts, and real-time kernel patches.

4. Implement redundancy and failover

Design for high availability: redundant switches, dual-homed servers, failover trading servers, and backup WAN links. Ensure failover is automatic and tested, with minimal disruption to order flow.

5. Ensure precise time synchronization

Use PTP (IEEE 1588) with hardware timestamping for sub-microsecond accuracy, GPS clocks in colo, and NTP as fallback. Explain how time sync affects order sequencing and regulatory compliance.

Key Points to Mention

  • Kernel-bypass networking (e.g., DPDK, Solarflare Onload) to avoid OS overhead and reduce latency.
  • CPU pinning, NUMA awareness, and disabling power-saving features to ensure deterministic performance.
  • Use of FPGA or ASIC for market data parsing and order generation to achieve ultra-low latency.
  • Redundant paths and automatic failover with BGP or proprietary protocols to minimize downtime.
  • PTP with hardware timestamping for nanosecond-level time synchronization across servers.
  • Latency budget breakdown: e.g., 10µs for NIC, 20µs for OS, 50µs for application, etc., to justify design choices.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q2

For TCP/IP specifics: explain Nagle's algorithm, how congestion control behaves in a low-latency trading context, when retransmits become a problem, and where you would choose UDP over TCP.

System DesignTechnical Trade-offsAPI & Integrations
Author's notes

Nagle I could explain fine.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Structure your answer around the four sub-questions, showing deep understanding of TCP/IP internals and their impact on low-latency trading. Use concrete examples from trading systems to illustrate trade-offs and your decision-making process.

Pro tip: Emphasize that in low-latency trading, you often disable Nagle's algorithm and use TCP_NODELAY, but also consider that even with that, TCP's congestion control and retransmission mechanisms can introduce unacceptable latency, leading to UDP with application-level reliability.

1. Explain Nagle's Algorithm

Describe how Nagle's algorithm coalesces small packets to reduce network overhead, and why this is problematic for latency-sensitive applications. Mention TCP_NODELAY to disable it.

2. Congestion Control in Low-Latency Trading

Discuss how TCP congestion control (e.g., slow start, congestion avoidance) can cause latency spikes and is often unsuitable for trading. Explain why traders might use UDP or custom protocols to avoid these delays.

3. Retransmits as a Problem

Explain that retransmits occur on packet loss and introduce delays due to round-trip time and head-of-line blocking. In trading, even milliseconds matter, so retransmits can lead to missed opportunities.

4. Choosing UDP over TCP

List scenarios where UDP is preferred: when latency is critical, when occasional packet loss is acceptable, or when implementing custom reliability (e.g., for market data multicast). Mention that UDP avoids handshakes, congestion control, and retransmits.

5. Summarize Trade-offs

Conclude by summarizing the trade-offs between TCP and UDP in trading, emphasizing that the choice depends on specific requirements like latency, reliability, and message ordering.

Key Points to Mention

  • Nagle's algorithm: coalescing small packets, TCP_NODELAY to disable.
  • TCP congestion control: slow start, congestion avoidance, and its impact on latency.
  • Retransmits: triggered by packet loss, cause delays and head-of-line blocking.
  • UDP advantages: no handshake, no congestion control, lower latency.
  • Use cases for UDP: market data multicast, order entry where speed is critical.
  • Application-level reliability: sequence numbers, acknowledgments, and retransmit logic in UDP-based protocols.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q3

Walk through the complete order-flow path from a trader clicking 'send' at the remote office to the order hitting the exchange matching engine. Account for where each microsecond is spent.

System DesignRoot Cause Analysis
Author's notes

I probably should have structured this more carefully from the start.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Structure your answer as a chronological journey through each component, from the trader's click to the matching engine, quantifying latency at each stage. Emphasize the critical path and where optimizations (hardware, software, network) can shave off microseconds. Conclude by discussing trade-offs and monitoring.

Pro tip: Demonstrate awareness that in ultra-low-latency trading, every microsecond counts, so mention specific technologies like kernel bypass, FPGA, and precise time synchronization (PTP) to show depth. Also, acknowledge that the exact breakdown varies by firm and setup, but the principles remain.

1. Client-Side Processing

Start with the trader's action: click event handling, order validation, and serialization into a message (e.g., FIX or binary). Mention the time spent in the application and OS (system calls, context switches).

2. Network Transmission

Cover the path from the trader's machine to the firm's gateway: NIC, switch, router, and any WAN links. Highlight propagation delay, serialization delay, and queuing. Discuss optimizations like kernel bypass (e.g., Solarflare) and multicast.

3. Gateway and Risk Checks

Explain the firm's gateway: order parsing, risk checks (e.g., credit, position limits), and forwarding to the exchange. Note that risk checks can add latency but are necessary; mention parallelization or FPGA-based checks.

4. Exchange Ingress and Matching

Describe the exchange's entry point: order normalization, validation, and queuing to the matching engine. Then, the matching engine's processing: order book lookup, matching algorithm, and trade generation. Emphasize the exchange's internal latency.

5. Quantify and Optimize

Summarize typical microsecond budgets for each segment (e.g., client 10-50µs, network 100-500µs, gateway 5-20µs, exchange 10-100µs). Discuss how to measure (hardware timestamps, PTP) and optimize (colocation, FPGA, low-latency NICs).

Key Points to Mention

  • Kernel bypass techniques (e.g., DPDK, Solarflare Onload) to reduce OS overhead
  • Network protocols: TCP vs. UDP, multicast for market data, and the impact of retransmissions
  • Colocation and proximity to exchange to minimize propagation delay
  • Hardware acceleration: FPGAs for risk checks and order parsing
  • Time synchronization (PTP) for accurate latency measurement
  • Risk checks and their trade-off between safety and latency

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.