Team Phoenix

Overall Objectives
Scientific Foundations
Application Domains
New Results
Contracts and Grants with Industry
Other Grants and Activities

Section: Scientific Foundations

Adaptation in Systems Software

DSLs in Operating Systems

Integrating our adaptation methodologies and tools into the development process of real-size software systems was achieved by proposing a new development process. Specifically, we proposed a new approach to designing and structuring operating systems (OSes) [34] . This approach was based on DSLs and enables rapid development of robust OSes. Such approach is critically needed in application domain, like appliances, where new products appear at a rapid pace and needs are unpredictable.

Devil - a DSL for device drivers

Our approach to developing systems software applied to the domain of device drivers. Indeed, peripheral devices come out at a frantic pace, and the development of drivers is very intricate and error prone. The Compose group developed a DSL, named Devil (DEvice Interface Language), to solve these problems; it was dedicated to the basic communication with a device. Devil allowed the programmer to easily map device documentation into a formal device description that can be verified and compiled into executable code.

From a software engineering viewpoint, Devil captures domain expertise and systematizes re-use because it offers suitable built-in abstractions [35] . A Devil description formally specifies the access mechanisms, the type and layout of data, as well as behavioral properties involved in operating the device. Once compiled, a Devil description implements an interface to an idealized device and abstracts the hardware intricacies.

From an operating systems viewpoint, Devil can be seen as an interface definition language for hardware functionalities. To validate the approach, Devil was put to practice [33] : its expressiveness was demonstrated by the wide variety of devices that have been specified in Devil. No loss in performance was found for the compiled Devil description compared to an equivalent C code.

From a dependable system viewpoint, Devil improves safety by enabling descriptions to be statically checked for consistency and generating stubs including additional run-time checks [36] . Mutation analysis were used to evaluate the improvement in driver robustness offered by Devil. Based on our experiments, Devil specifications were found up to 6 times less prone to errors than writing C code.

Devil was the continuation of a study of graphic display adaptors for a X11 server. We developed a DSL, called GAL (Graphics Adaptor Language), aimed to specify device drivers in this context [42] . Although covering a very restricted domain, this language was a very successful proof of concept.

Plan-P - a DSL for programmable routers

Besides device drivers, the Compose group also explored the area of networking in the context of DSLs. More specifically, we developed a language, named Plan-P, that enables the network to be programmable and thus to offer extensibility [41] . As such, Plan-P enables protocols to be defined for specific applications. Plan-P extends a language, named Plan, developed by the University of Pennsylvania and devoted to network diagnostics. Plan-P enables routers to be programmed in a safe and secure way without any loss in bandwidth. To achieve safety and security, the language is restricted, and programs are downloaded into the routers as DSL source code to enable thorough verifications. For efficiency, a light Just-In-Time compiler is generated from the Plan-P interpreter via program specialization. This compiler is installed on routers to compile uploaded Plan-P source code.


Logo Inria