Report About Real Time Systems' Conferece

Report about real time systems. Conference by FranÐ*ois Dubuc.

Introduction.

If we think about the ISS (International Space Station), we can see that it's a complex system. If we see a specific system, the navigation system, we can see how the altitude is regulated depending on the actual altitude, so the altude must be always between 300 km. and 400 km. If the altitude is not between these values, then the propulsion system is activated. If we think that the reaction must be in real time and the fact that the size of the ISS is the same of a football field then we can imagine how complex all these systems are. But there is also another interesting real time systems like the systems of: cars, aeroplanes, rockets, etc... For each one of these systems there is a development process, which depends on costs, time of development and characteristics. It will be different the process for a car (it's going to be made thousands of them) or an aeroplane that the process for a space shuttle (only one space shuttle); this thing is going to be essential for the test process. The conference is going to be focused in space vehicles.

What is a system?

A system is composed by other subsystems. Each subsystem has a particular objective and all subsystems together must work alright. A subsystem is going to be composed by simple or complex components. A standard system is going to have a user, who is the person/people that uses the system; it's going also to have interfaces for communication with other systems or objects; an environment (which is those things and physical phenomenon around the system) and a concrete behaviour.

Real time systems and tests.

Real time systems are systems that must have an immediate response to intern or extern stimulus. To test these systems can be difficult because it's necessary to simulate all conditions. If the product for testing is something like a space shuttle it's impossible to make "real tests" because it could be very expensive; the test are limited to those "on ground", not flying, so it's very important to simulate all conditions and to study the behaviour of the system. Then it's important to act in two ways: sensors and simulators. In the way of simulators it's very important to test all possible conditions. To test all possible conditions is supposed to test not only software but also physical conditions. For testing software it's very important to execute all lines of code and to make the software to evolve in different versions; each version must be completely tested. For that aim it's very interesting to test automatically and it will save time and money.

In the way of physical conditions it's important to simulate all meteorological conditions: brusque changes of temperature, rain, ice, etc... Also physical aspects: electromagnetism, vibrations, gamma rays, noise tests and others. The reason for those tests is because in the space the conditions are very special and many electrical and computer parts can crash in those abnormal conditions. One of the consequences is not to buy the best components but to buy the strongest components (the objective is a long life component); and also to buy many of them for having spare parts (for 20 or 30 years). For example: to change the processor cost a lot and only the avionics spend hundreds of million euros. So the most important is not only the calculation capability of processors.

There is a loop formed between real equipments and simulators for representing with more or less fidelity the reality.

Why is it important to test always the software? For example: let's think about ariane 5. It exploited some seconds after takeoff; it was because one of the switchboards (or something like this, a main electronic component) was used for the ariane 4, so they thought that it was not necessary to test it because it has been tested enough for the fourth version of ariane. The fact is that the way of flying was different for ariane 5 as that of ariane 4; when this switchboard began to work after launching it made many flight mistakes so it had to be destroyed. That was because they should have tested all components (for ariane 5 and those incorporated form the 4th) together.

For the tests there are many methods: the Monte Carlo method which tries to test all possible entries and checking all corresponding outputs. Each test is adapted in accordance with "sensibility" categories, from A (the more critical components) to D (the less critical components). The A tests must be done over the real machine where they are going to be working. For these last tests it is necessary another company (different from that which made the software or the component) does the testing; that's because of security reasons. The other company can test things that the first one didn't test. In the next section we'll see why the test automation is very important.

Life cycle.

At the beginning of the project we have a specification, that's mean, a list of requirements for each system. These requirements must be tested so it has to exist a way of testing them. The specification is very important because it can be like a contract; that's means that it is necessary to have many meetings with the clients for defining each specification