- From: Chin Chee-Kai <cheekai@ncb.gov.sg>
- Date: Tue, 7 Mar 95 14:35:47 SST
- To: www-talk@www10.w3.org, yhuubert@cherry.viabalt.ee
I'd like to clarify (or try to do it) some of the comments given earlier. I'd consider HTTP a "monodirectional" in the sense that only the browser (client) may initiate a communication to the server; the server cannot initiate a transmission to a specific browser until that browser opens a TCP connection to the server. HTTP specs states clearly that it is a request-response protocol; a protocol to transmit a request sent by the browser and to return a response generated by the server, within one TCP session (ie. prior to the connected socket being closed by either party). While TCP is a full duplex channel, the fact that only the server and not the browser provides a listening port means that no one can ever initiate a connection to a browser. So without making the browser install a listening port, it's hard to expect a browser to receive unsolicited information. On the "interactivity" issue, it is not HTML that needs to be interactive, but more on the system that comprises the HTTP server, the browser and the network. Interactivity, as I see it from the perspective of my problem on hand, allows a person to initiate a talk to another person via the server. Since a person runs a client (browser) to talk to a server which will in turn initiate a channel to another person's client browser, it is inherent in this set up that mechanisms are needed to identify any specific instance of browser, and for the browser to have the ability to "receive" incoming calls via the provision of a browser's listening port. On the comment about "I'm busy reading a document when a server decides to interact with my browser", I prefer to look at the issues separately. That a browser equipped with a listening socket and so may be sent with unsolicited data does not imply that the browser cannot decide how best it may present the data to the user. It may prompt the user, or allow the user to turn off the line, just like you may pull your phone off the wall to stop annoying calls. But the presence of an underlying mechanism that allows incoming calls to the browser is important to interactive applications. On the CCI issue, the web page on CCI seems to be uncertain about how a browser should select its port for applications to communicate with it. The service offered by the browser in this case is also slightly different. The applications manipulate the browser and makes use of its HTML rendering capability and HTTP retrieval capability to perform additional services. This is unlike allowing a remote server from sending in unsolicited information to the browser (A user would not want his/her browser to be manipulated by an external server without control). Also, if a browser or CCI-application uses a fixed port, no other browser may run with CCI capability. Scripts and CCI-based services address a different requirement, which is to provide "value-added" or intelligent processing and making use of browser's display, URL and HTTP-aware capabilities to achieve those services. It still does not allow a server to inform a particular browser its intended messages. On Netscape's PullPush mechanism, the server basically holds on to the TCP channel which a connecting browser has established in expectation of more data to be sent to the browser. The channel is held open as long as the server thinks there is more data that it may soon send, regardless of whether there is anything to be sent through the channel at the moment. Once again, it is not "interactive" in that the server may not initiate a communication to the browser once it allows that channel to be closed. It is an attempt, I'd consider, to do-away with the "interactivity" issue by exploiting TCP's duplex channel. The ability for a browser to be interactive does not also imply that the browser should be conference-capable. Using the example I gave in the previous mail, a long server search may first acknowledge the browser's request with a short document, and subsequently, when the search is concluded, initiation transfer of the results to the browser. This is not conferencing, but interactivity. It is a phenonmenon for WWW to have come this far. Interactive browsers need not always be involved in interactive transactions; it may disable its incoming port to behave like the current browsers. But if browsers can listen, it will mean that information can seek users instead of making users seek information. As usual, this is purely discussion on the points made and no personal feelings should be attached. Please share if you have a different opinion so others, myself included, may also benefit from learning. Afterall, that's why we're all here :) I've made MOLTIP-UID and R-HTTP specifications available at http://www.ncb.gov.sg/staff/cheekai/html/moltip-techspec.html Chin Chee-Kai http://www.ncb.gov.sg/staff/cheekai/
Received on Tuesday, 7 March 1995 01:35:46 UTC