vCryptoNetby Perkunas LLC Book a call
C++ / DPDK market data engine

Crypto market data in 2.6 µs

vCryptoNet replaces Linux networking, OpenSSL and WebSocket libraries with one purpose-built stack on DPDK. Decrypted, arbitrated market data lands in shared memory about 30 times sooner.

Book a free setup call See the benchmarks
20 exchange feeds shared memory
2.6µsmedian, NIC hardware timestamp to decrypted data in shared memory
30×faster median than every WebSocket library we tested
5,000concurrent feeds on a single CPU
42Bof memory per connection, against kilobytes in OpenSSL
Internal latency · same trade, six stacks

Watch the same trade arrive.

Six implementations read Binance btcusdt@trade over 20 connections each, side by side on one AWS machine. This replays their measured latency, from the NIC's hardware timestamp to decrypted data, in slow motion.

Slowed down about 26,000 times AWS c7i.metal-24xl · Tokyo (apne1-az4) · full method
Under the hood

We took the kernel out of the path.

    Smart arbitration

    20 connections. The first copy wins.

    Exchange TLS endpoints drift in load and latency by more than 200 µs over time. vCryptoNet listens on many redundant connections, keeps the first copy of every trade and scores every link as it goes.

    • Late copies are dropped 10x to 100x faster than generic TLS/TCP stacks can process them.
    • Slow links get recycled. Each connection is scored and replaced by an IP predicted to be faster.
    • The gains compound. Across up to 5,000 channels at peak volume, standard libraries queue up. vCryptoNet stays real time.
    20 redundant connectionsshared memory →
    0trades kept
    0late copies dropped
    0slow links recycled

    Simulation of the method. Not live market data.

    Why it is faster

    Built around how crypto market data behaves.

    Generic network stacks are written for every kind of traffic. vCryptoNet exploits properties of exchange market data that they can't.

    Ultra-lean memory and CPU
    42bytes

    per connection. OpenSSL needs several kilobytes.

    • Zero-copyData flows straight across every network layer without buffering.
    • Zero heap allocationsOpenSSL and WolfSSL pause to allocate memory in the hot path. vCryptoNet never does.
    • Hot CPU cacheWhile idle, threads loop synthetic packets so code and data stay hot in L1 and L2.
    Custom TLS network stack
    0kernel

    checks, retransmit buffers or congestion windows.

    • Custom ENA driverA heavily customized DPDK poll-mode driver, tuned for market data arbitration.
    • No server delaysThe TCP stack never slows server pacing. It can even ACK a lost packet already received on a redundant connection.
    • Forces low latencyExchanges behave as if TCP_NODELAY were on, which tightens tail latency in heavy spikes.
    Smart connection management
    5,000feeds

    per CPU, scored and arbitrated in real time.

    • Connection recyclingEvery link is scored. Slow ones are replaced by IPs predicted to be faster.
    • Fast arbitrationLosing packets are processed 10x to 100x faster than in generic TLS/TCP stacks.
    • Compounded effectMicrosecond advantages add up across thousands of channels while standard libraries lag.
    Benchmarks

    Measured on AWS, published in full.

    Same machine, same feed, same moment. Every implementation stamps the first time it sees each trade ID, and we compare the best timestamps. Counted from Binance's own clock, vCryptoNet's slowest 1% still arrives sooner than the slowest 5% on either WebSocket library.

    Open the full benchmarks

    Binance timestamp to decrypted data

    µs · lower is better

    Measured from the exchange's own "E" timestamp, so this includes the network path from Binance to the AWS machine.

    Services and pricing

    The engine, the feed, and the engineers behind them.

    vCryptoNet license
    from$4,800/ month

    License the engine and plug it into your existing trading platform. We help you integrate it and provide technical support.

    • Custom DPDK drivers and TLS/TCP stack
    • Integration help with your platform
    • Technical support included
    Book a free setup call
    Market data feed
    from$1.40/ hour

    Fast crypto price information delivered to your systems over UDP, billed by the hour.

    • UDP delivery into your systems
    • Pay only for the hours you use
    Ask about the feed
    Engineering
    from$200/ hour
    • C++ development. Over 15 years in HFT and high-performance computing. Ultra low latency software in C++11, 14, 17 or 20.
    • Development consulting. Technical team management for firms without a senior person in financial markets or this technology.
    • Hiring assistance. Shortlisting, evaluation, and C++, architecture, networking and Linux tests for candidates.
    Discuss a project
    Built for Proprietary trading firms and hedge funds running algorithmic HFT on crypto exchanges hosted on AWS.
    BinanceCoinbaseBybitHuobiKuCoin
    Past customers include Liquid Capital Tower Research Crédit Agricole Chopper Tradingnow DRW
    About Perkunas

    “We treat any application we write as an avionics package, where our clients are our passengers.”

    Perkunas LLC builds fintech software with no tradeoffs on quality or robustness. Quick-to-market strategies may win in bull times, but we believe the market eventually rewards quality.

    Trading technology is mission critical: any failure can, and likely will, lead to large financial losses. That is why we work in a mix of feature-driven development and classic waterfall, which we found gives the best results in a mission-critical environment.

    FoundedOctober 2025
    FlagshipvCryptoNet
    FocusCrypto HFT on AWS
    StackC++ / DPDK
    Fabio Fernandes, director of Perkunas LLC, arms crossed
    Director

    Fabio Fernandes

    Senior C++ developer specializing in low-latency network infrastructure. He holds an M.Sc. in Computer Science (High-Performance Computing).

    Since 2011, Fabio has held senior engineering and contract roles at trad-fi and crypto traders, hedge funds and market makers across Dubai, Singapore, London and Chicago, where he built DPDK-based TLS network stacks achieving sub-90 ns latency.

    DubaiSingaporeLondonChicago
    LinkedIn
    Contact

    Talk to the engineer who built it.

    Book a free 30-minute setup call, or tell us about your platform and we will get back to you.

    Book a free setup call
    Emailinfo@vcryptonet.com
    Phone+370 643 77688
    LinkedInlinkedin.com/in/fabfer