ApacheConUS2007 ZeroLatencyHTTP

Zero Latency HTTPthe comet technique : Zero Latency HTTP the comet technique


And Comet is…. : And Comet is…. Comet is the Serial Port of Web 2.0


Who am I – bla bla : Who am I – bla bla fhanik@apache.org Tomcat Committer / ASF member Co-designed the Comet implementation Implemented NIO connector in 6 Responsible for session replication and clustering Been involved with ASF since 2001


What we will cover : What we will cover Brief History of HTTP How HTTP is used today Introduction to AJAX Beyond AJAX, The Comet Technique Comet Implementation – Under the hood Uni- vs. Bidirectional Comet Problems and their solutions Demo Q & A


The History of HTTP : The History of HTTP 1.0 (7th release) RFC 1945 in 1996 Co-authored by ASF’s Roy T Fielding Current RFC 2616 Lead by W3C Activity on specs are closed Standard has been achieved http://www.w3.org/Protocols/History.html


What is HTTP : What is HTTP HyperText Transfer Protocol Text based protocol Request/Response semantics Stateless Most commonly run over TCP/IP networks Protocol used for much more than just hypermedia information sharing


What is HTTP : What is HTTP Text based protocol – CRLN delimited GET /Protocols/History.html HTTP/1.1 Host: www.w3.org User-Agent: Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.8.1.2) Gecko/20070219 Firefox/2.0.0.2 Accept: text/xml,application/xml,application/xhtml+xml,text/html;q=0.9,text/plain;q=0.8,image/png,*/*;q=0.5 Accept-Language: en-us,en;q=0.5 Accept-Encoding: gzip,deflate Accept-Charset: ISO-8859-1,utf-8;q=0.7,*;q=0.7 Keep-Alive: 300 Connection: keep-alive


What is HTTP : What is HTTP Request/Response Semantics GET /Protocols/History.html HTTP/1.1 Host: www.w3.org Keep-Alive: 300 Connection: keep-alive HTTP/1.x 200 OK Date: Mon, 19 Mar 2007 15:46:17 GMT Server: Apache/1.3.37 (Unix) PHP/4.4.5 Keep-Alive: timeout=2, max=100 Content-Length: 19575 Connection: Keep-Alive Request Response


What is HTTP : What is HTTP Stateless TCP session can end after each request State based on cookies TCP/IP transport No limitations for transporting with other protocols Delivers both text and binary data


HTTP Limitations : HTTP Limitations Request/Response Change one field on a page requires a reload of the entire page Client always has to initiate the request Stateless Server required to keep state State is timed out, if client is not cancelling it Finer grained communication is needed AJAX Rich Clients (FLEX, OpenLaszlo, …)


AJAX : AJAX Asynchronous JavaScript +(and) XML Uses HTTP Asynchronous data processing Able to request data from server and update a page in the browser Common examples maps.google.com


Benefits of AJAX : Benefits of AJAX Performance More can be accomplished in less amount of transactions Less user interaction Program can make intelligent decision about when a request needs to happen and what data it needs to fetch Sample Application for view of benefits http://www.ajaxinfo.com/ucst/index.html


Beyond AJAX : Beyond AJAX What did AJAX not accomplish? Still client poll based Server push can be accomplished by a client poll followed by a delayed response Traditional web/servlet containers are thread-per-request based Server resource tied up for the duration of the request


Introducing Comet : Introducing Comet The answer to our problems Q: What problems? A: The fact that port HTTP/80 has been replacing every decent TCP protocol over the years. Q: What do you mean? A: I’ll explain…it all boils down to port 80


Slide15 : Picture borrowed from: dojotoolkit.org


How it Works : 3. Register with background thread How it Works Timeline Server Client 1. HTTP request 2. BEGIN Event 5. HTTP Response - 200 OK – Tx Enc: Chunked 4. Event Complete Webapp Async Thread 6. Server data push 7. Encode and transfer 8. Another server push 9. Client Push 10. READ Event 11. One more client push Webapp Worker Thread 12. And on it goes…


Under the hood : Under the hood Open HTTP request, leave it open ended Instant or delayed response(s), leave it open ended Tomcat processes just like servlet CometProcessor extends HttpServlet(or should) CometFilter extends Filter Connection remains open and is writable


Under the Hood : Under the Hood Open ended request POST /load/echo HTTP/1.1 User-Agent: Filips Comet Client/1.0 Host: 127.0.0.1:8080 Transfer-Encoding: chunked 10 test data test 1 Open ended allows for new data without the overhead for a new request Zero Latency HTTP


Under the Hood : Under the Hood Open ended response HTTP/1.1 200 OK Server: Apache-Coyote/1.1 Transfer-Encoding: chunked Content-Type: text/plain;charset=UTF-8 Transfer-Encoding: chunked Date: Mon, 19 Mar 2007 22:25:36 GMT 12 test data test 1 Open ended allows for new data without the overhead for a new request Zero Latency HTTP


Under the Hood : Under the Hood CometProcessor Extends HttpServlet to provide a fall back option should the connector not support Comet void event(CometEvent event)… Types of Events: BEGIN READ END ERROR A request is a Comet request if the URL maps to a servlet that implements the CometProcessor interface


Under the Hood : Under the Hood CometFilter Extends javax.servlet.Filter to provide fall back option should the Connector not support Comet void doFilterEvent(CometEvent event, CometFilterChain chain) … Same mapping rules as CometProcessor


Under the Hood : Under the Hood BEGIN New connection created Request headers fully parsed and mapped to a servlet Partial body may or may not have been read Processed by a worker thread May proceed directly to a READ event


Under the Hood : Under the Hood READ New data available on the connection Processed by worker thread Read now to avoid repeated “read” events Read using servlet input stream event.getHttpServletRequest.getInputStream()


Under the Hood : Under the Hood END The request has naturally come to an end Server may shutdown, and needs to close connections Only if Connector shutdown prior to app, piggy back on HttpServlet.destroy for most apps instead ERROR When the connection times out or an error has occurred during invokation


Under the Hood : Under the Hood Events are “connection” centric Thrown based on IO events Reading should only be done when an event is thrown Writing can be done async Synchronize your write methods


Gotcha’s : Gotcha’s NIO – client disconnect is signaled by a READ with a -1 or EOFException by InputStream.read() The event lifecycle is not always what it appears to be While the API seems simple, it is pretty fragile Easy to break if misused


Bidirectional Comet : Bidirectional Comet Tomcat supports bidirectional comet communication Server can push data to the client Client can push data to the server Timeout values can be set on a per-connection basis Unidirectional Comet means that only the server can push data to client


Scalability? : Scalability? No more thread per connection or thread per request One asynch thread can handle writes for thousands of Comet connections No overhead of request/response headers for each request No overhead for TCP setup/breakdown Memory overhead for open connections


Examples of use : Examples of use Rich Clients Mail Maps Online conferences Clients with need for server push Stock tickers Auction sites Chat


Future Improvements : Future Improvements Non blocking reads Non blocking writes Bayeux protocol implementation End Request – AJAX over Comet Keepalive Connections Comet through HTTP proxy(?)


Feedback : Feedback Tomcat Dev List fhanik@apache.org Help and ideas are wanted