Palestra: A COMPOSITION APPROACH TO MUTUAL EXCLUSION ALGORITHMS FOR GRID APPLICATIONS.
Profa. Luciana Arantes Laboratoire d'Informatique de Paris 6 (LIP6) University of Pierre et Marie Currie (Paris 6) Paris, France, na Série de Seminários 2009 da Pós-Graduação, dia 21/08/2009, às 14:00 h, Auditório do IC , Sala 85 - IC 2.
| What | Palestra |
|---|---|
| When |
21/08/2009 from 14:00 to 15:00 |
| Where | Auditório do IC - Sala 85 - IC 2 |
| Add event to calendar |
|
ABSTRACT
A Grid comprises of a large number of nodes grouped into clusters.
Nodes within a cluster are often linked by local networks (LAN)
while clusters are linked by a wide area network (WAN). Therefore,
Grids inherently present a hierarchy of communication delays: the
cost of sending a message between nodes of different clusters is
much higher than that of sending the same message between nodes
within the same cluster.
Distributed or parallel applications that run on top of a Grid
usually require that their processes get exclusive access to some
shared resource (critical region). A mutual exclusion (MUTEX)
algorithm then ensures that exactly one process can access a
critical region at a given time. We are particularly interested in
token-based distributed mutual exclusion algorithms where a unique
token is shared among all nodes (processes), and its possession
gives a process the exclusive right to access the critical region.
Several distributed token-based MUTEX algorithms exist in the
literature. However, they do not take into account the
above-mentioned hierarchy of communication latency. Therefore, we
have proposed a composition approach which allows the combination
of any two distributed token-based MUTEX algorithms: one at
intra-cluster level and a second one at inter-cluster level. By
using our composition mechanism different algorithms can be easily
"plugged in" on both levels. Furthermore, performance evaluation
tests have shown that the good choice for an inter cluster mutual
exclusion algorithm depends on the application behaviour i.e., the
frequency with which the application processes request the shared
resource.
OBS: A palestra será ministrada em português.
