SafeScale projectC. CERIN, J.L. PAZAT, J.L. ROCH, R. KERYELL

https://www-lipn.univ-paris13.fr/safescale/

SafeScale overview

Funded by ANR (Agence Nationale de la Recherche)

3 years term project (started on January 1st, 2006)

Partners LIPN (Paris XIII) (coordinator) christophe.cerin@lipn.univ-paris13.fr IMAG/INRIA (Grenoble), Jean-Louis.Roch@imag.fr IRISA/INRIA (Rennes), Jean-Louis.Pazat@irisa.fr ENSTB (Brest) Ronan.Keryell@enst-bretagne.fr Université Joseph Fourier (Grenoble). Roland.Gillard@ujf-grenoble.fr

Research focus Security and safety in global ambient computing systems

computational grid peer-to-peer environments.

Environment Middleware provides strong authentication, secure communications and

resource management. Computational nodes operate in an unbounded environment subjected to

a wide range of attacks.

• Fail-stop failures: connection/disconnections of heterogeneous resources

• Malicious failures: forgery of results on resources (Trojan horses, …) Challenge

Develop applications with guarantees on correctness of computed results.

Two kinds of security issues (1/2)

Internet

1. Node failures “fail stop” model

User

Two kinds of security issues (2/2)

Internet

1. Task forgery– “massive attacks”

Userworm,virus

bad result

Works objective

• study and evaluate a methodology and tools to obtain certified results. our approach

• Adaptability to support addition / resilience of resources– dynamic task scheduling and mapping– adaptability of software components

• Probabilistic certification of results to support forgeries– Verifications of few randomly chosen tasks– Performed on trusted resources (hardware crypto-processor)

• Validation on two class of large scale computational applications– data merging and sorting– finite field computations

experiments • on the grid'5000 architecture

Running an application: from this...

To that...

How to detect: random faults (type 1) or malicious (systematic) faults (type 2)?

Send certified codes on ill machines to detect the natureof faults? (testing?)

Known/Unknown tools, results Platforms: Kaapi, XtremWeb, Grid5000 Fault-tolerance and adaptive programming (IRISA, IMAG)

Key point : application adaptability• Fine-grain work stealing [IMAG] (provable

performances on processors with changing speeds)• Adaptation of components [IRISA]

Probabilistic certification (ENSTB, IMAG) Detection of massive attacks [IMAG] Crypto-processor (ENSTB)

Applications: Sorting on an heterogeneous cluster (LIPN) Classification of provably secure cryptographic boxes (UJF)

Conclusionhttps://www-lipn.univ-paris13.fr/safescale/pôle « Grille Système Parallélisme »

(http://gsp.asr.cnrs.fr/)action sécurité et sûreté