+1, it should reduce greatly overhead and let the OS stack handle the traffic directly, it will also certainly reduce overhead in reader part (ie web server).
- Henri Gomez ___[_]____ EMAIL : [EMAIL PROTECTED] (. .) PGP KEY : 697ECEDD ...oOOo..(_)..oOOo... PGP Fingerprint : 9DF8 1EA8 ED53 2F39 DC9B 904A 364F 80E6 >-----Original Message----- >From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]] >Sent: Thursday, May 30, 2002 7:55 PM >To: List Tomcat-Dev >Subject: PROPOSAL: clientFlush() method > > >Hi, > >I would like to add a new method to Response. > >clientFlush() will be called by ServletOutputStream.flush() to >notify the >container ( and the low level connector ) that the user has >explicitely >asked for the stream to be flushed. > >The default impl. will be empty, and in coyote adapter we'll generate >an Action.CLIENT_FLUSH that may be interpreted by the connector. > >This would allow the connector to increase the granularity, which in >turns result in about 20% reduction in the overhead ( for ajp/socket ). > >This happens by allowing the use of a BufferedOutputStream on the >socket - the packets to send the header and the first chunk and >the end response messages will all be sent at once, in one >write() operation instead of 3. > >The servlet spec allows us to do that, but we must honor the flush() >requests. > >I'll leave the option as disabled in jk2 ( since 4.1 and the current >3.3 doesn't support this ), but an option will enable the buffered >writing. > >Comments ? Larry, Bill ? > >( I tested this with 3.3 - Remy, I suppose something similar >will be done >in a future release of 4.1.x, I don't want to delay the release ) > >Costin > > >-- >To unsubscribe, e-mail: <mailto:[EMAIL PROTECTED]> For additional commands, e-mail: <mailto:[EMAIL PROTECTED]> -- To unsubscribe, e-mail: <mailto:[EMAIL PROTECTED]> For additional commands, e-mail: <mailto:[EMAIL PROTECTED]>
