18.3 Market Data Systems: Ticker Plants, Data Broadcast/Fan-Out, Multicast Feeds, FIX Protocol

What it is

Market data systems distribute the exchange’s authoritative event stream to consumers with different freshness, bandwidth, and recovery needs. A ticker plant is the low-latency normalization and publication path; broadcast and fan-out distribute the resulting messages; multicast and FIX are delivery or protocol choices, not interchangeable guarantees.

How it works

A market-data handler receives sequenced updates from the exchange, validates sequence numbers, normalizes fields, and updates an in-memory top-of-book or depth view. The ticker plant publishes updates to local consumers first. Downstream distribution may use UDP multicast for a one-to-many feed, a message bus for durable replay, or a specialized binary protocol for bandwidth-sensitive consumers.

A consumer must distinguish an update from a snapshot. If sequence numbers skip, the consumer requests a snapshot or resynchronizes rather than applying a gap blindly. A feed can be fast while the decision built from it is stale, so timestamps, receive times, and processing times should be retained separately.

    flowchart LR
    Exchange[Exchange event stream] --> Normalize[Sequence validation and normalization]
    Normalize --> Snapshot[Snapshot and book builder]
    Normalize --> Publish[Ticker plant]
    Publish --> Local[Local strategy consumer]
    Publish --> Multicast[UDP multicast distribution]
    Publish --> Bus[Durable event bus]
    Multicast --> Consumers[Remote consumers]
    Bus --> Recovery[Replay and recovery]

FIX is a session-oriented message standard commonly used for order routing and execution reports. It has message types, tags, checksums, sequence numbers, and recovery semantics. A market-data session can use FIX, but high-rate consumers often use a more compact binary feed because FIX framing and text encoding add bandwidth and parsing overhead. The protocol choice does not remove the need for sequence and timestamp validation.

8=FIX.4.4|9=138|35=W|34=1842|49=CLIENT|56=VENUE|52=20260101-12:00:00.123456|115=3|48=BTC-USD|22=8|31=42000.00|32=4.0|14=42001.00|11=ORDER-17|21=3|6=42000.00|39=2|150=F|151=4.0

The line is a tagged FIX-like artifact for inspecting fields, not a complete session. A production parser must not infer missing tags, reset sequence numbers without the prescribed recovery process, or treat a quote update as an executable guarantee.

Tradeoffs

Design choice	Gain	Cost or risk
UDP multicast	Low per-consumer fan-out cost	Loss, reordering, and receiver tuning require controls
Durable message bus	Replay and recovery are straightforward	Adds latency and operational storage
Binary market-data protocol	High message density and fast parsing	More implementation effort and less interoperability
FIX session	Broad counterparty interoperability	Text parsing and session recovery add overhead
Snapshot plus deltas	Efficient steady-state updates	Consumers need versioned snapshots and gap handling
One feed per consumer type	Tailored bandwidth and latency	More contracts and more cross-feed consistency work

When to use

You need a fast view of an authoritative exchange event stream.
Consumers have different latency, replay, and bandwidth requirements.
You can version snapshots and detect sequence gaps.
You can measure feed staleness separately from strategy decision time.
You need an interoperable order or execution protocol.

Alternatives

Polling a REST snapshot — simple and recoverable, but too slow and expensive for active market making.
A single normalized internal bus — easy to govern, but can hide the latency differences between local and remote consumers.
Kafka-style durable streams — excellent replay and consumer isolation, but usually not the lowest-latency direct feed.
Vendor-specific binary feeds — efficient and feature-rich, but create portability and versioning work.

18.2 Order Types & Execution: Market/Limit/Stop Orders, Smart Order Routing 18.4 Low-Latency Engineering & High-Frequency Trading: Kernel Bypass, Co-Location, Hardware Timestamping, LMAX Disruptor Pattern, and FPGA Accelerator Offloading