SELLBYX
Menu
WorkServicesBlogAboutCareers
ENUA
+380 98 151 72 78

Trading · Client delivery

Scaling market-data delivery toward 500,000 concurrent connections

The Sellbyx team reduced per-connection overhead and redesigned real-time topic fan-out so each market update could be encoded once and shared across subscribers.

500Kproject target connections10–50reported updates/s per connection~90%reported CPU reduction
Case context

Delivered by the Sellbyx engineering team. The client is confidential. Performance figures come from project records and team reporting; no independent public audit is available.

01 / Challenge

The constraint behind the architecture.

The naïve model performed serialization, allocation and routing work per connection. At high fan-out, cost grew with both event volume and subscriber count, making conventional framework abstractions the bottleneck.

02 / Architecture

The decisions that changed the system.

  1. Move connection handling to a lean event loop rather than one heavyweight execution path per socket.
  2. Fan out shared immutable buffers by topic so one market update is encoded once.
  3. Use a compact binary protocol, ring buffers and explicit backpressure instead of repeated JSON allocation.
  4. Tune the network path and model capacity from measured memory and CPU per connection.

03 / Outcome

What the work established.

The architecture was designed to support 500,000 concurrent connections on a small cluster while preserving operational headroom. The capacity figure is the documented project target; the client has not published an independent audit.

Direct access to the delivery team

Discuss a similar product or systems challenge.

Discuss a project