Trade Demo Architecture

Producer: MarketDataSimulator

MarketDataSimulator is a BackgroundService optimized for very high synthetic throughput. It continuously generates TradeSignal records across equities, crypto, and futures-style instruments using random-walk prices, burst intensity, preallocated symbol/exchange strings, batched timestamps, and monotonic sequence IDs.

Each generated signal is published once:

MarketDataSimulator
    ↓ IMarketDataBus.Publish(signal)
InMemoryMarketDataBus
    ├── LosslessTickStore.OnMarketData(signal)
    └── TradeQueueProcessor.OnMarketData(signal)

This split is important:

• The authoritative/replay path accepts ticks into LosslessTickStore for sequence-aware recent replay and optional durable journal writes.
• The UI path accepts ticks into TradeQueueProcessor for latest-by-symbol coalescing and SignalR broadcast.

In-Process Fan-Out: InMemoryMarketDataBus

InMemoryMarketDataBus is a small in-process pub/sub component. It is registered as IMarketDataBus and receives the fixed subscriber set from dependency injection:

IMarketDataSubscriber[]
├── LosslessTickStore
└── TradeQueueProcessor

Subscribers own their own channel and backpressure semantics, so the producer path does not need to know whether a consumer is replay-oriented or UI-oriented.

UI Stream: TradeQueueProcessor

TradeQueueProcessor is a BackgroundService and market-data subscriber. It owns a bounded Channel<TradeSignal> sized for bursty demo traffic:

The channel uses DropOldest because stale quotes are less useful than the newest quote in a real-time terminal.

MarketDataSimulator
    ↓ Publish
InMemoryMarketDataBus
    ↓ OnMarketData / TryEnqueue
Bounded Channel<TradeSignal>
    ↓ drain on timer tick
Latest-by-symbol dictionary
    ↓ SignalR SendAsync("TradeSignals", snapshot)
Browser and WPF clients

Key mechanics:

• bounded channel backpressure with stale-tick discard
• single-reader drain loop
• multiple producer support
• latest-by-symbol coalescing
• snapshot broadcast near monitor refresh rates
• Stats SignalR messages with processed/dropped/coalesced/broadcast metrics
• interlocked counters on the hot path
• latency and byte metrics recorded through PerformanceMetrics

The clients do not receive every raw tick. They receive compact snapshots that keep the UI responsive while the server continues processing high raw event rates.

Lossless Recent Store + Optional Journal

LosslessTickStore is also a BackgroundService and market-data subscriber. It owns its own bounded channel and keeps a large recent replay window.

It tracks accepted count, dropped count, sequence gaps, last sequence ID, queue depth, and recent-window size. The word “lossless” describes the intent of the authoritative path, but the current implementation can count drops if the in-memory channel cannot immediately accept a tick under extreme saturation.

Supported journal interfaces:

ITickJournalWriter
ITickJournalReader
ITickJournal

Implemented providers:

• NullTickJournal when journaling is disabled or provider is None
• local segment files
• Azure Blob-backed journal
• Azure Event Hubs-backed journal adapter

The default appsettings.json has journaling disabled:

                    JSON
"TickJournal": {
  "Enabled": false,
  "Provider": "None"
}