HFT & Low-Latency C++ Interview Guide

What HFT Interviewers Look For

Cache Optimization

• Data-oriented design: struct of arrays vs array of structs
• Hot/cold data separation keep critical path data in L1
• Prefetching (__builtin_prefetch), cache line alignment (alignas(64))
• Avoid pointer chasing; prefer contiguous memory (vector > list)

Lock-Free Structures

• CAS loops (compare_exchange_weak/strong)
• Lock-free queue → src/concurrency/lock_free_queue.cpp
• ABA problem: use tagged pointers or hazard pointers
• Memory ordering: acquire/release for producer-consumer

Custom Allocators

• Arena/pool allocators for deterministic latency → src/modern_cpp/custom_allocator.cpp
• Memory pool → src/stl_like/memory_pool.cpp
• Avoid malloc on hot path (pre-allocate everything)
• Huge pages (mmap with MAP_HUGETLB) to reduce TLB misses

SIMD & Vectorization

• SSE/AVX intrinsics for batch processing → src/systems/hpc_gpu/simd_vectorization.cpp
• Auto-vectorization hints: restrict, loop alignment, no branches in loop body
• Use SIMD for checksum, parsing, search in market data

Kernel Bypass

• DPDK: user-space networking, poll-mode drivers
• Solarflare OpenOnload: kernel bypass via ef_vi
• Bypass eliminates syscall overhead (~1μs → ~100ns for packet)
• Busy-polling vs interrupt-driven

Hot/Cold Path Separation

• [[likely]] / [[unlikely]] attributes
• Place error handling in cold functions (won't pollute icache)
• Inline hot path; __attribute__((noinline)) cold path
• Profile with perf stat look at icache misses

Branch Prediction

• __builtin_expect (or C++20 [[likely]])
• Branchless programming: x = (cond) * a + (!cond) * b
• Sort-by-likelihood in switch statements
• Measure with perf stat -e branch-misses

Memory Layout

• alignas(64) to avoid false sharing
• Pack structs on hot path; pad between thread-local data
• Use std::hardware_destructive_interference_size

Repo HFT Examples

• Order book (sorted price levels, O(1) best bid/ask) → src/systems/hft/order_book.cpp
• Market data handler (fast parsing) → src/systems/hft/market_data_handler.cpp
• Matching engine → src/systems/hft/matching_engine.cpp

Common HFT Interview Questions

Question	Key Points
Design an order book	Sorted map for price levels, deque per level, O(1) top-of-book
Lock-free SPSC queue	Ring buffer + atomic head/tail, acquire-release ordering
Reduce tick-to-trade latency	Kernel bypass, pre-allocated buffers, branch-free parsing
False sharing explain & fix	Two threads writing adjacent cache lines; pad with alignas(64)
When would you use relaxed ordering?	Counters, statistics no inter-thread data dependency
Custom allocator for trading	Pool allocator: fixed-size blocks, free-list, O(1) alloc/dealloc
How to measure latency?	rdtsc, clock_gettime(CLOCK_MONOTONIC), histogram percentiles
What is cache warming?	Touch data structures before market open to load into cache

Target Companies & Focus Areas

Company	Special Focus
Citadel Securities	System design, low-latency networking, atomics
Jump Trading	FPGA awareness, kernel bypass, lock-free
Optiver	Mental math + cache optimization + branchless code
Two Sigma	Concurrency, system design, Python+C++ interop
Tower Research	Template metaprogramming, SIMD, memory model