分布式通讯架构引擎套件 Distributed Communication Engine Suite

CEDA

为严苛场景而设计的四项基石Four pillars for the most demanding workloads

从协议层到业务层,CEDA 在每一个细节上为金融级可靠性与性能而优化。 From protocol to application layer, every detail is engineered for financial-grade reliability and performance.

高并发High Concurrency

单机 50 万 TPS 异步处理能力,支持 26 万并发连接,承载行情洪峰。500K TPS per node with 260K concurrent connections, absorbing peak market data bursts.

低延迟Low Latency

接近网卡流量极限的数据通道,为高频交易与实时风控提供稳定响应。Throughput that approaches the NIC hardware limit, stable for HFT and real-time risk.

跨语言Cross-Language

原生支持 C++/Java/Python/C# 等多语言 SDK,一套架构统一异构系统。Native SDKs for C++, Java, Python, C# — one architecture unifies heterogeneous systems.

协议适配Protocol Adaptation

内建 FIX、Binary、自定义协议适配层,一次接入,万路互通。Built-in FIX, Binary and custom protocol adapters — connect once, talk to everything.

分层解耦,组件化组合Layered, decoupled, composable

CEDA 由多个可独立部署的组件构成,通过统一的消息总线协同工作,支持按需组合以匹配不同业务形态。 CEDA is built from independently deployable components connected by a unified message bus — composed on demand to match any business topology.

APPLICATION Market Data E-Trading Quant Strategy Risk / Clearing CEDA CORE Message Bus Protocol Gateway Session Manager Monitor Async Scheduler / Thread Pool Serialization / Codec Multi-Language SDK TRANSPORT TCP / UDP / Multicast Shared Memory / RDMA FIX / Binary Protocols INFRASTRUCTURE Linux · x86 / ARM · Kernel Bypass · Container / Bare Metal

被数字验证的极致表现Performance, measured

以下数据来自真实压测环境。在千兆网络环境下,CEDA 的数据吞吐已逼近网卡物理极限。 Results from real benchmarks. On a 1 GbE network, CEDA's data throughput approaches the physical NIC limit.

500,000TPS
单机峰值消息吞吐Peak message throughput
接近千兆网卡流量极限Near 1 GbE hardware limit
260,000+
最大并发连接数Max concurrent connections
单节点支撑Per single node
117.76MB/s
实测数据吞吐Measured data throughput
1024B 消息 · 50 订阅者1024B msg · 50 subscribers
14,000+
单源扇出订阅能力Single-source fan-out
行情分发场景实测Market data scenario

吞吐量压测结果Throughput Benchmark

不同订阅者规模下的 TPS 表现TPS across different subscriber scales
500K 400K 300K 200K 100K 95K 110K 115K 500K ★ 350K 1 sub 10 subs 50 subs 2,000 subs 14,000 subs 1024 B 1024 B 1024 B 125 B 125 B
Pub / Sub / TopicPub / Sub / Topic
消息大小Msg Size
TPS
MB/s
1 / 1 / 1
1024 B
95,000
97.28
1 / 10 / 1
1024 B
110,000
112.64
1 / 50 / 1
1024 B
115,000
117.76 ★
1 / 2,000 / 1
125 B
500,000
62.5
1 / 14,000 / 1
125 B
350,000
43.75
★ 网卡极限: 千兆网卡流量极限约 125 MB/s,扣除包头后数据流量上限约 115 MB/s。在 1024 字节消息、50 订阅者的配置下,CEDA 实测吞吐达到 117.76 MB/s,已触及硬件物理极限。测试环境:Non-persistent / Non-durable。 ★ Hardware ceiling: 1 GbE has a theoretical limit of ~125 MB/s; data payload ceiling after headers is ~115 MB/s. With 1024 B messages and 50 subscribers, CEDA measured 117.76 MB/s — hitting the physical NIC limit. Test: Non-persistent / Non-durable.

— 01

面向多源行情的高扇出分发,支持组播、过滤、聚合与重放。High fan-out distribution for multi-source feeds with multicast, filtering, aggregation and replay.

  • 多交易所接入Multi-exchange connectivity
  • Tick 级数据回放Tick-level replay
  • 订阅级过滤Subscription filters
— 02

订单管理与撮合前置的高可靠通道,具备会话级顺序保证与幂等语义。Reliable channels for OMS and pre-matching — with session-level ordering and idempotent semantics.

  • FIX / Binary 双栈FIX / Binary dual-stack
  • 严格顺序投递Strict ordered delivery
  • 毫秒级故障切换Millisecond failover
— 03

为策略引擎提供超低延迟数据通道和信号总线,稳定支撑高频回测与实盘。Ultra-low-latency data pipes and signal bus for strategy engines — stable for both HFT backtesting and live trading.

  • 共享内存通道Shared-memory channel
  • C++ / Python SDKC++ / Python SDK
  • 确定性时序Deterministic timing

15+

8+

12+

20+

5+

10+

6+

10+

与我们的工程师聊聊Talk to our engineers

告诉我们你的场景,我们将安排专属的技术对接、定制化 POC 与部署方案。 Tell us your scenario — we'll arrange a dedicated technical session, a tailored POC and deployment plan.

邮箱 · EmailEmail
support@meridian.com.cn
电话 · PhonePhone
021-5080 5159
微信客服 · WeChatWeChat Support
扫描右侧二维码添加Scan the QR on the right
添加微信客服Add WeChat Support 微信扫一扫 · 即刻咨询 Scan with WeChat · Get instant help