[pso] Seminar 'Reliable Software Systems' - Luni, 10 decembrie, 14:00, EC101

[Razvan Deaconescu]

Salut!

Luni 10 Decembrie sala EC 101 orele 14-15 va avea loc un seminar
prezentat de George Candea, sef de lucrari la EFPL (Elvetia). Aria de
interes a lui George o reprezinta "software systems" (sisteme de
operare, retele de calculatoare, limbaje de programare).

Recomand acest seminar persoanelor cu interes in domeniile de mai sus.
Scopul lui este, totodata, stabilirea de perspective de colaborare
(internship-uri, proiecte de diploma, burse de doctorat), lucru care va
poate fi util la anul.

Mai multe detalii aveti mai jos.

Razvan

----
An Overview of EPFL's Research in Reliable Software Systems
  Prof. George Candea
  http://people.epfl.ch/george.candea
  EPFL (Lausanne, Switzerland)


In aceasta expunere voi prezenta cateva proiecte de cercetare de
la EPFL, voi furniza informatii despre facultatea noastra si societatea
EuroSys, precum si oportunitati pt. colaborare, schimb de studenti,
proiecte de diploma si studii de doctorat.  Din motive de vocabular,
expunerea tehnica va fi in limba engleza.


I will give an overview of the research on reliable software systems
being done at EPFL, including failure immunity, fault injection,
automated testing, and practical formal verification.  As a concrete
example, I will treat in depth the topic of deadlock immunity -- a
property  by which programs, once afflicted by a given deadlock pattern,
automatically develop the ability to avoid all future occurrences of
similar  deadlocks.

I will present D-Immunix, a technique that enables large software
systems to automatically gain such immunity without any assistance from
programmers or users.  I will describe an implementation  of D-Immunix
for Java and demonstrate its effectiveness against bugs in  real
systems, like MySQL JDBC and the Java JDK runtime.

You will see that, in the worst case, our prototype introduces only
modest performance overhead (e.g., <2% in JBoss, a widely used
application server wit >350 thousand lines of code), that is independent
of code size.

In most practical cases, deadlock immunity is as powerful as complete
freedom from deadlocks, so D-Immunix resolves a long-standing challenge
of ridding large-scale real systems of elusive deadlocks. This is joint
work with my doctoral student Horatiu Jula.
---