Section: New Results

Experimenting with Clouds

Simulating Distributed IT Systems

Participants : Toufik Boubehziz, Benjamin Camus, Anne-Cécile Orgerie, Millian Poquet, Martin Quinson.

Our team plays a major role in the advance of the SimGrid simulator of IT systems. This framework has a major impact on the community. Cited by over 900 papers, it was used as a scientific instrument by more than 300 publications over the years.

This year, we pursued our effort to ensure that SimGrid becomes a de facto standard for the simulation of distributed IT platforms. We further polished the new interface to ensure that it correctly captures the concepts needed by the experimenters. To that extend, we also added several complex applications to our Continuous Integration (CI) testing framework, to ensure that we correctly cover the needs of our existing users. We also worked toward our potential users by reworking the documentation, and by proposing new pedagogical resources. Making SimGrid usable in the classroom should greatly increase its impact. A publication on this effort was recognized as Best Paper in the Workshop on Education for High-Performance Computing [17].

The work on SimGrid is fully integrated to the other research efforts of the Myriads team. This year, we added the ability to co-simulate IT systems with SimGrid and physical systems modeled with equational systems [16]. This work, developed to study the co-evolution of thermal systems or of the electic grid with the IT system, is now distributed as an official plugin of the SimGrid framework.

Formal Methods for IT Systems

Participants : The Anh Pham, Martin Quinson.

The SimGrid framework also provide a state of the art Model-Checker for MPI applications. This can be used to formally verify whether the application entails synchronization issues such as deadlocks or livelocks [7].

This year, we pursued our effort (in collaboration with Thierry Jéron, EPI SUMO) to improve the reduction techniques proposed to mitigate the state space explosion issue. We are leveraging event folding structures to improve the performance and accuracy of dynamic partial ordering reduction techniques. We plan to submit a publication on this work by the beginning of 2019.

Executing Epidemic Simulation Applications in the Cloud

Participants : Christine Morin, Nikos Parlavantzas, Manh Linh Pham.

In the context of the DiFFuSE ADT and in collaboration with INRA researchers, we transformed a legacy application for simulating the spread of Mycobacterium avium subsp. paratuberculosis (MAP) to a cloud-enabled application based on the DiFFuSE framework (Distributed framework for cloud-based epidemic simulations). This is the second application to which the DiFFuSE framework is applied. The first application was a simulator of the spread of the bovine viral diarrhea virus, developed within the MIHMES project (2012-2017). Using both the MAP and BVDV applications, we performed extensive experiments showing the advantages of the DiFFuSE framework. Specifically, we showed that DiFFuSE enhances application performance and allows exploring different cost-performance trade-offs while supporting automatic failure handling and elastic resource acquisition from multiple clouds. These results are described in a journal article under submission. In 2018, we also released the fist major version of the DiFFuSE software (v1.0) under the CeCILL-B licence.

Implicit locality awareness of Remote Procedure Calls evaluation

Participants : Javier Rojas Balderrama, Matthieu Simonin.

Cloud computing depends on communication mechanisms implying location transparency. Transparency is tied to the cost of ensuring scalability and an acceptable request responses associated to the locality. Current implementations, as in the case of OpenStack, mostly follow a centralized paradigm but they lack the required service agility that can be obtained in decentralized approaches. In an edge scenario, the communicating entities of an application can be dispersed. In this context, we focus our study on the inter-process communication of OpenStack when its agents are geo-distributed regarding two key metrics: scalability and locality. Scalability refers to the ability of the communication middleware to handle a massive number of clients while consuming a reasonnable amount of resources. Locality refers to the ability of the communication middleware to serve requests as localy as possible while mitigating long-haul data transfers.

Results show that scalability and locality are very limited when considering the traditionnal broker-based approaches [28]. Novel solution such as router-based communication middleware offers better scalability and a good level of implicit locality. This work is an initial step towards building locality-aware geo-distributed systems.

Tools for the experimentation

Participant : Matthieu Simonin.

In collaboration with the STACK team and in the context of the Discovery IPL, novel experimentation tools have been developed. In this context experimenting with large software stacks (OpenStack, Kubernetes) was required. These stacks are often tedious to handle. However, practitioners need a right abstraction level to express the moving nature of experimental targets. This includes being able to easily change the experimental conditions (e.g underlying hardware and network) but also the software configuration of the targeted system (e.g service placement, fined-grained configuration tuning) and the scale of the experiment (e.g migrate the experiment from one small testbed to another bigger testbed).

In this spirit we discuss in [19] a possible solution to the above desiderata. We illustrate its use in a real world use case study which has been completed in [28]. We show that an experimenter can express their experimental workflow and execute it in a safe manner (side effects are controlled) which increases the repeatability of the experiments.