Experience report from a TINA experimentation using ORBs and the Web

F. Dupuy and S. Pensivy
France Telecom CNET

An emerging need is to provide mediation services to end-users in order to help them to find, in a consistent way, the information services they are looking for. By mediation services, it is meant services like identification services, service directories, global charging services, end-user profile management, etc. As the Internet free information superhighway is now turning into a business-driven toll-based superhighway, there is a need for standardizing these mediation services that will otherwise be as different as Internet Access Providers.

We are currently looking at this issue, trying to assess the extent to which standards based on open distributed processing techniques can solve them. The TINA-C Consortium provided us with the candidate software architecture, named TINA, that we decided to go with. TINA is based on the use of OMG CORBA environments, upon which libraries of (telecom) service object-based components are specified, tested and agreed upon by around 40 company members. The telecom service domains the architecture is deemed to be applicable to are the intelligent services, the telecom network management services, and multimedia information services.

When we built our mediation service infrastructure, we made the following assumptions or remarks. The end-user terminal software (Web browsers) is generally not running on a CORBA object request broker. The network mediation services, which ought to be open, easily deployed, should be developed on the other hand on an ORB. Lastly, we assumed that information service providers will not shift straightaway from current http-based Internet servers to IIOP-based object-oriented servers. Therefore, around one year ago, we embarked upon developping mediation services on CORBA (following the TINA specifications given in OMG IDL), with HTTP-IIOP gateways. We are now at a stage where we can assess the interest of using both a CORBA environment and the HTTP protocol.

Thanks to the ANSA-sponsored APM efforts in the domain, we could overpass a first interworking issue between HTTP and CORBA by using the CGI scripts generated by ANSAweb that enables to invoke IDL methods. The next step (hopefully considered as an extension of the W3C architecture) would be to transmit the end-user requests to the mediation platform using the IIOP protocol.

We definitely deem that HTTP 1.0 lacks the concept of access and service sessions, as defined in TINA. Currently, each end-user HTML request consists of establishing one or more TCP connections, requesting the HTML data, and releasing the TCP ports. So currently there is no notion of access session to the mediation services (that would enable to keep the user identification and to customize the service) and service session with a particular information server. Therefore tricks had to be found in order to associate a single access session identifier to all HTML requests that a particular end-user emits under the same 'virtual' session : we propose for example to add an access session identifier in the PATH_INFO of the HTML requests and to extend the HTTP server with kernel processes that route the coming requests towards the appropriate context-dependent application process in charge of this particular access session.

Another point is that the correspondence between the Uniform Resource Identifiers (URI) selected by the end-users in his/her HTML pages and the identifiers of the actually invoked CORBA/TINA components is not yet as proper as it should. The URI mixes information related to the physical location of the server and semantical information related to the invoked service. The use of IDL interface references in URIs and that of a (possibly CORBA-compliant) name service in the terminal would offer much more flexibility in the requests.

Lastly, another issue is the assessment of the CORBA platform performance under the work load that seems to be characteristic of the most sollicitated Internet servers (more than 200 transactions per second).

In conclusion, the joint use of the W3C and the TINA/OMG architectures is inevitable and adds a lot of value to both types of information networks. It is timely to start pushing the information technology industry towards provisioning products that facilitate the use of the two. There is also some standardization work to be done (i.e. CORBA interface references in URI) in order to channel all the current efforts on the subject.

last modified on April 15, 1996
Inquiries to Stephane Pensivy.