Plasma GitLab Archive
Projects Blog Knowledge



A few words about this and that...

Organization of Ocamlnet

Ocamlnet has grown a lot, and it may look a bit confusing to beginners. Fortunately, Ocamlnet follows a few organizational principles:

String routines are often factored out

String routines like URL parsing, buffer handling, message decoding etc. are often not directly included in the protocol interpreter, but defined in separate modules. Actually, there is a whole library, Netstring, only consisting of string functions.

The advantage is obvious: String functions can be reused, for example from within clients and servers, but also from user code wishing to do certain string manipulations directly.

Common types ease plugging parts together

It is encouraged that one can plug together various parts of Ocamlnet to create new functionality. Ocamlnet tries to use the same types for the same kind of values throughout the whole library to make this easier. Examples:

  • Netmime defines mail messages, but as this data format is also used in HTTP, Http_client represents requests and responses with the same classes
  • Netchannels defines object-oriented I/O channels. These are used almost everywhere in Ocamlnet.
Common infrastructure for logging, exception printing, and signal handling

All Ocamlnet modules use this infrastructure:

  • Netlog is a simple logging module and redirector
  • Netexn provides extended support for exception printing
  • Netsys_signal is a little framework for coordinating signal handlers
These modules are as simple as possible to do their task. It is up to the user to extend them, e.g. by plugging in fancier logging modules.

Preference of asynchronous protocol interpreters

A number of protocol interpreters are written in asynchronous style. This is generally preferred because of better timeout and error handling, but also because it makes it possible to run several instances of these protocol interpreters in the same thread concurrently (i.e. it is a kind of micro-threading). Ocamlnet includes the Equeue library as support library for asynchronous network programming (see Equeue_intro for more information).

Nevertheless, nobody is forced to use these protocol interpreters asynchronously. There is always also a simple way for synchronous calls (i.e. where the execution of the caller blocks until the network function is completely done).

Parallelization is done at a single place

In other network environments, often every protocol interpreter defines functions for parallelizing its use, e.g. by using thread pools. In Ocamlnet we avoid that. For parallelization there is only the Netplex library which supports both multi-threading and multi-processing.

It is easy to use Netplex with the available server implementations:

  • Nethttpd_plex allows it to run a webserver inside Netplex
  • Rpc_netplex can run SunRPC servers inside Netplex
  • Netcgi_plex can run FCGI, SCGI, and AJP connectors inside Netplex
There are a number of advantages. First, one can easily create compound servers running a number of protocol interpreters at the same time. All interpreters are configured from a single place. All interpreters use the same logging mechanism. All interpreters share the same workload mechanism. All interpreters can be administered with the same command, netplex-admin.

Second, one does not need to develop the parallelization code anew for each protocol implementation. It is difficult enough.

Third, this approach makes it first possible to provide a serious implementation of multi-processing. For this kind of parallelism, more coordination between the interacting components is required than for multi-threading. For example, when child processes are forked off, it must be clear which protocol interpreter are run within them.

At Mylife, we have developed a server program that integrates dozens of different protocols under the same roof. The Netplex approach is scalable!

Pick what you like, and omit things you don't understand

Ocamlnet is modular enough that one can cherry-pick the parts one needs. After all, it is only a library, i.e. a collection of Ocaml modules, and by using one module there is no strict requirement to use another module.

Of course, there is also a "use hierarchy" in Ocamlnet - the HTTP code uses Equeue, for instance. However, the user of HTTP needs not to know this or care about it. Equeue is then simply an internally used helper library.

Future Directions

Ocamlnet is already focused on Internet protocols. Current activities especially try to

  • focus even more on datacenters as computing environments (affecting scalability, selection of protocols, etc.)
  • provide more support to get high availability (e.g. utilities to detect failed services quickly, or a common directory service)
  • improve multi-core support (especially speed up interprocess communication)

Libraries Using Ocamlnet

There a number of libraries using Ocamlnet:

This web site is published by Informatikbüro Gerd Stolpmann
Powered by Caml