Author: Venkata N. Padmanabhan (supervised by Jeffrey C. Mogul)
---------------------------------------------------------------

Salient features
----------------
* support for persistent connections with clients (with client-side mods.)

* 2 new HTTP methods for pipelining requests:
  GETLIST: given a list of URLs, get all the correspoding documents
  GETALL:  given the URL for an HTML document, get that document and all
           its inlined images that reside on that server.

* fully interoperable with unmodified clients
  

Implementation Details
----------------------

 The first and most important task is for each of the client and server
to inform each other that it is 'new', i.e. supports persistent connections.
This is done in an unobtrusive fashion so that it interoperates with 
unmodified implementations as well.

* Persistent connnections:

   The client sends its request with the additional field "Client-type: NEW"
in the MIME headers (see get_mime_headers() in http_mime.c) to indicate that
it supports (and desires) a persistent connection. In response to such a 
request, the server process handling the request continues to wait for 
further requests, and the client-server connection becomes persistent. Also
in its response to the client, the server includes the header line "NEW"
(see begin_http_header() in http_log.c) to tell the client that it supports
persistent connections as well.

The connection persists until one of the following happens:
   - a server-side script is executed in response to a client request
   - this particular server process is chosen as the victim when the server
     wishes to shed load
   - the connection remains idle beyond a threshold

   In each of these cases, the connection to the client is closed.

* Pipelining requests (GETLIST/GETALL):

   Both these methods are implemented as ordinary GETs with an additional
field in the MIME headers. This field is either "Command: GETLIST" or
"Command: GETALL (see get_mime_headers() in http_mime.c). According to
the primitive, different actions are taken:

  - GETLIST: Since in the current implementation GETLIST is implemented as
        just a bunch of GETs sent back-to-back, the server on receiving a
        GETLIST request does little more than making note of it. This 
        information could be used to guide other decisions. For instance,
        a GETLIST could treated as a hint that a burst of requests is
        going to arrive, so the server could defer activities such as 
        prefetching.

  - GETALL: If the GETALL is for an HTML document, the server parses the
        document to determine URLs of all inlined images that reside on 
        the server and sends back corresponding image data as well (see
        sendall_fd() in http_request.c which invokes code in HTMLparse.c
        to do the parsing). 


Organization of the code etc.
----------------------------

./src: contains all the source files and the makefile for httpd.
       httpd.c: contains the main server code that forks off a new process
           in response to a new connection request from a client
       http_request.c: parses the request and invokes the appropriate function
           depending on the type of request (GET, PUT etc.)
       http_mime.c: parses the MIME headers in the request.
       HTMLparse.c: Mosaic code that parses HTML, adapted to support GETALL

./conf: the configuration files. 
       access.conf, httpd.conf, srm.conf:  allow swetting of several parameters
           including port # for the server, the root of the server's document tree,
           etc. The files with the ".conf-dist" extensions are the original NCSA
           configuration files with comments describing the various parameters. 

./logs: the server logs.

./icons: a few commonly used icons

./cgi-bin: a few CGI scripts


Building the server
-------------------

 First set AUX_FLAGS and EXTRA_LIBS according to the system you are compiling
on.   
 Then in ./src, type "make <arch-name>" where <arch-name> is the type of the 
architecture. I have tested this only for DEC MIPS running ULTRIX and DEC
Alpha runing OSF.

 The executable "httpd" is placed in ./

Running the server
------------------
  Type  ./httpd [-d server_root] [-f server_confname] [-v]

   -d server_root: root of the server source tree, i.e. ./
   -f server_confname: name of the server configuration file; the default is conf/httpd.conf
   -v : just prints the version number
   
Common errors
-------------
1. The server port (in conf/httpd.conf) is set to 80 but the server is not being run
   as root. Use a port number larger than 1023.

2. The server is run a second time, while an older invocation is still active. Either kill
   the old invocation or use a new port number.
 
Changed or added source modules:
--------------------------------
src/HTML.h
src/HTMLamp.h
src/HTMLparse.c
src/http_access.c
src/http_delete.c
src/http_dir.c
src/http_get.c
src/http_log.c
src/http_mime.c
src/http_post.c
src/http_put.c
src/http_request.c
src/http_script.c
src/httpd.c
src/httpd.h
src/util.c