High Performance Messaging for Web-Based Trading Systems

High Performance Messaging for Web-based Trading Systems

Frank Greco - @frankgreco Director of Technology, Kaazing Outline

§ Background

§ Trading Systems and the Web

§ Requirements

§ Web Messaging for Trading Apps

§ Layered Approach to Web Protocols

§ New Requirements

§ The Web beyond the Browser

§ Financial Services – Capital Markets

§ Desktop trading systems – Swing/Java/TIBCO, et al

§ Customer use cases – FX, Commodity desks

§ Director of Technology @ Kaazing § Chairman NYJavaSIG (javasig.com) § Largest JUG in North America 7,000+ members § [email protected], @frankgreco

FX Trader Application – front office demo.kaazing.com/forex

§ Trading System – A set of rules governing buying/selling

§ An application that expedites a trading system

§ Modern Web Trading requirements

- HTML5 – no more plugins

- UI/UX – Sencha/extJS, GWT, jQuery, GPU acceleration, etc

- Usable application cache

- Fast, reliable, secure

- Minimal impact to back-end services

- Works with pub/sub brokers and data feeds

- High message rate, small payload – equity ticks, watchlist, currency pairs, etc.

§ Designed for document transfer – HTTP

- Short-lived Request / Response interaction

§ Bidirectional, but half-duplex

- Traffic flows in only one direction at a time

§ Stateless

- Large amounts of metadata resent for each request

Web was not designed for “real-time”, event-based services…

HTTP

TCP

Oldies Hits - AJAX and the Comet Pollers - 2006

iOS/Android

Pricing Feeds Mobile Authentication Authorization Browser/Native Client

Make

Net me Msg News/Alerts Web Broker friendly Browser JMS/AMQP etc… Trade Exec

What everyone does… Desktop Java/.NET Reports

iOS/Android

Pricing Feeds Mobile Authentication Authorization Browser/Native Client

Make

Net me Msg News/Alerts Web Broker friendly Browser JMS/AMQP etc… Trade Exec Developers want a secure version of this Desktop architecture so they Java/.NET can focus on the app Reports

Instead, we try this traditional solution

Trading App Broker

Just have to open some ports…

Merely deploy Java/.NET on Only need to traverse several client… proxies between the endpoints…

Easy, right?...

Or we try to proxy things over the web

ticker plant

Scalability Issues… Protocol mismatches… Ineﬃcient… Latency Issues…

§ Users are Demanding more from Apps § UI/UX Requirements are more sophisticated § API Explosion § Browser Enhancements § Web no longer just about Documents but the existing infrastructure reflects this legacy

The New Web is a Foundation for Enterprise-grade Applications

• “Real-Time”, bi-directional connectivity • IETF Protocol - RFC 6455 – Dec 2011 • W3C API • Easily add event-based (trading) data to web apps • Avoids polling • Avoids HTTP meta-data overhead • Shares port with HTTP (80/443) • Peer protocol to HTTP (both use TCP) • Most Important API in HTML5 • Facilitates other protocols • Puts the web in a better place…

§ W3C API – Candidate Recommendation - http://www.w3.org/TR/websockets/ § IETF Formal Protocol (RFC 6455) - Event-driven JavaScript API - Full-duplex communication protocol § Integrates HTTP addressing - ws://yourcompany.com/collaboration_svc - wss://anothercompany.com/marketdata_svc § Traverses firewalls, proxies, routers securely § Text and Binary § Leverages Cross-Origin Resource Sharing (CORS)

Legacy HTTP vs WebSocket

Seems like a no-brainer for trading apps…

§ Creation of WebSocket Java components to handle bi-directional WebSocket conversations

§ Handling WebSocket events

§ Creation and consumption of WebSocket text and binary messages

§ Allows for WebSocket protocols and content models for an application

§ Configuration and management of WebSocket sessions, like timeouts, retries, cookies, connection pooling

§ Specification of how WebSocket application will work within the Java EE security model!

§ Official Java SE WebSocket API in the works…

What does this really have to do with high- performance messaging?

Half Duplex - HTTP Full duplex

Web

Browser Web Tier Middleware Msg Broker

Live Web

Full duplex - WebSocket Full duplex

Web

WebSocket Msg Broker Server

Complexity

More Traders

More Services

More…More…

Growth Greco’s Law! Scalability = Growth / Complexity Simple things scale…

But… Dealing with WebSocket is like dealing with TCP.

It’s a streams-based model. You need to understand how to handle streams-based data over the wire.

So how do I do publish/subscribe for my trading system?

Where is the Application-level Protocol?

• Who handles retries? • How do we handle publish/subscribe semantics? • How do we handle market data? • Is guaranteed delivery possible (trades)? • What if the client is not active? • How do we handle [fill in with favorite semantics] • What about partials? • Who’s responsible for entitlements? How do I manage that? • etc…

Whoa… Its just like TCP! Huzzah!

Browser and Native Applications

JMS XMPP AMQP B2B FTP VNC mktdata etc

WebSocket

TCP

Internet

WebSocket WebSocket Gateway Gateway

Other Considerations…

• Need to handle multiple WS versions • Need to handle multiple (and legacy) browser versions • Can’t have business logic in the DMZ • Have to work in multiple DMZs • AuthN/AuthZ has to work multiple times • High-availability topology • Concerns about open ports with back-end service • Services architecture needs to be consistent • Native, HTML5 and hybrid environments • Integrate easily with non-messaging services • XaaS integration – the Web beyond the browser • Bandwidth management • etc… All things you need for a real enterprise app

Publish/Subscribe over the Web for Trading Systems

Java Message Service (JMS) over WebSocket

A 60-second Tutorial

In case you haven’t heard of JMS…

• Java EE Message Oriented Middleware • JMS 1.0 2001, JMS 1.1 2002, JMS 2.0 (Feb 26, 2013) • Asynchronous Messaging vs. RPC • Loosely coupled vs. Tightly coupled • Pub/Sub, Topics, Queues • Transactions, Reliable

conn = createConnection(); sess = conn.createSession(); topic = jndiContext.lookup(topic); pub = sess.createProducer(topic);

pub.send(“hey Frank”);

conn = createConnection(); sessconn = conn.createSession();= createConnection(); sessconn = conn.createSession();= createConnection(); topicsess = jndiContext.lookup(topic); = conn.createSession(); subtopic = sess.createConsumer(topic);= jndiContext.lookup(topic); subtopic = sess.createConsumer(topic);= jndiContext.lookup(topic); sub.setMessageListener(this);sub = sess.createConsumer(topic); …sub.setMessageListener(this); …sub.setMessageListener(this); public… void onMessage(Message m) { publicString voids = m.getText(); onMessage(Message m) { msg publicString voids = m.getText(); onMessage(Message m) { … doString stuff s… = m.getText(); broker } … do stuff… } … do stuff… }

1. Create connection from ConnectionFactory see tutorial.kaazing.com 2. Create session 3. Create topic or queue 4. Create producer 5. Create consumer 6. Create message listeners (function to handle messages)

2 session = connection.createSession(false, Session.AUTO_ACKNOWLEDGE); 3 var myTopic = session.createTopic("/topic/myTopic"); 4 topicProducer = session.createProducer(myTopic); 5 topicConsumer = session.createConsumer(myTopic); 6 topicConsumer.setMessageListener(handleTopicMessage);

7. Send messages

var doSend = function(message) { message.setStringProperty(MESSAGE_PROPERTIES.userId, userId); topicProducer.send(null, message, DeliveryMode.NON_PERSISTENT, 3, 1, function() sendFromQueue(); }); };

8. Process messages: the message listener function: handleTopicMessage()

var handleTopicMessage = function(message) { if (message.getStringProperty(MESSAGE_PROPERTIES.userId) != userId) { $("#slider").val(message.getText()); $("#pic").width(message.getText()); } };

• Kaazing • Node.js/socket.io • ActiveMQ • Tomcat • Jetty • Oracle Glassfish • Java EE • Play Framework • Rabbit MQ • JBoss • IIS/ASP .NET 4.5 • PHP, Objective-C, Ruby, Python, C/C++, JVM-langs… • Many more…

Futures

What’s next for Trading Systems and the Web?

Clouds

Storage Regulatory hypothetical Email Docs Cloud Cloud Clearance

websocket websocket Collaboration

Cloud websocket websocket Wealth Management websocket

Enterprise Notifications

Transactions Cloud FX Cloud Cloud Sentiment Analytics Risk Analysis Management 36

Clearance as Compliance a Service as a Service External Cloud

Service Bus

websocket service service service

Risk Management Regulatory as a Service iPaaS Auditing as a Service websocket

Internet Service Bus Enterprise

Enterprise Service Bus

service service service

Integration as a Service Monitoring as a Service Governance as a Service Identity as a Service Testing as a WAN Optimization as a Service Risk Analytics Service as a Service Data Center as a Service Windows Desktop Enterprise Messaging as a Service as a Service CDN as a Service Network as a Service Sentiment Analysis as a Service Trade Clearance as a Service Database as a Service Notification as a Service

Security as a Service Backup Analytics as a Service as a Service EAI as a Service Telephony as a Auditing as a Service Service 38

[email protected]

Extends Browser as IETF RFC 24x7 Global full enterprise client Full JMS API for Compliance Support JavaScript, .NET, Flash, iOS, Android No business logic in Can Connect to (soon) Gateway - can be any TCP or UDP installed securely in DMZ Granular Client data source Timeouts for mobile JMS Topics, Queues, to detect offline Emulation when Durables, Acks for intermediaries block Guaranteed msgs iOS-APNS & native WS Android/GCM JMS Edition Works with any Integration JMS provider - no broker lock-in AuthN/AuthZ Bandwidth Kerberos and SSO Control Secure Services – can over WS integration close all inbound ports No Server-Side API Emulates WebSocket End-to-End for messaging needed protocol for old encrypted messages for legacy integration browsers w/ no polling for multiple DMZs