Fault-Tolerance
-
Raphael Yokoingawa de Camargo (Doctoral student at U. of São Paulo)
Implementation of a portable checkpointing-based rollback recovery
mechanism for sequential and parallel applications. Data is stored on a
distributed storage system composed of non-dedicated repositories. [
more...]
-
Stanley Araújo de Sousa (Masters student at U. F. do Maranhão)
Definition of an infrastructure to provide dependability to Grid applications and components.
Identification of User Access Patterns
-
Danilo Conde (Masters student at U. of São Paulo)
This work intends to develop a mechanism of
usage pattern recognition based on unsupervised machine learning
techniques. Our goal is to establish data clusters on the gathered
resources usage information. These clusters will define categories of
operation of each node of the grid. In a given time, the identification
of the current operating category will make it possible to make a
systematic decision about the possibility of use of a specific shared
resource.
-
Germano Capistrano Bezerra (Masters student at U. of São Paulo)
In
this work, cluster analysis will be applied to perform resource usage
pattern recognition. Prototypic behaviors of resource usage will be
identified. Based on the recurrence of these behaviors, a pattern
recognition framework will estimate the sharing availability of a
resource. Some cluster analysis concepts will be discussed and applied
to the computer resources usage pattern recognition framework. A software component was developed in order to perform the proposed
techniques validation through simulations. [more...]
Inter-Cluster Resource Management
-
Vladimir Emiliano Moreira Rocha (Masters student at U. of São Paulo)
The current version of InteGrade sees all Grid nodes as being in the
same pool of machines. Our goal is to develop protocols and algorithms
to allow InteGrade to organize its nodes in a structured collection of
clusters, each one composed of several individual nodes. Initially, we
will focus on protocols for linking
the different clusters one to
another, for disseminating resource availability information across
clusters, and for executing applications across clusters.
Distributed Execution of Parallel Algorithms
-
Ulisses Kendi Hayashida (Doctoral student at U. of São Paulo)
I'm studying techniques of loop transformations for extracting
parallelism from nested loop structures. Nested loops can be scheduled
to run in parallel so that execution time is minimized. The study of
"dependence vectors" and "systolic arrays" are fundamental parts for
the base of my work. Most commom literature known are for fine grained
algorithms, because initially the objective was to implement them in
high circuit density VLSI technology, designing special purpose VLSI
chips. Nowadays, with the advent of Grid Computing, it's good to
minimize the amount of communication to improve performance, improving
the quantity of data in a message send, for example, and so, coarse
grained algorithms are required. So, I intend to adapt known techniques
of loop transformations to coarse grained algorithms, specially those
related with nested loops. Fault tolerance techniques will be studied
too. I also intend to
implement some examples of algorithms using MPI.
Mobile Agents
-
Vinicius Gama Pinheiro (Masters student at U. of São Paulo)
This work aims at investigating and implementing
fault-tolerant agent-based mechanisms to
meet the requirements of executing parametric and long running applications in
opportunistic environments. To accomplish this in a dynamic environment
several parameters like number of replicas, checkpointing frequency
have to be tuned to meet different scenarios of resource availability.
Security
-
José de Ribamar Braga Pinheiro Junior (Doctoral student at U. of São Paulo)
Security
is of critical importance to the sucess of Grid Computing. In these
systems, the fundamental problem is to provide security policies that
are expressive and flexible enough to satisfy the specific needs of a
usual variety of aplications. In the InteGrade Project, it would be
interesting to support multiple, dynamic security policies. In this
work, we will investigate the development of a framework to provide
security policies to Grid Computing in general and to InteGrade in
particular.
Development Environment for Grid Applications
-
Eduardo Leal Guerra (Masters student at U. of SãoPaulo)
Currently,
much of the work done by Integrade applications developers is done by
hand. Moreover, on Integrade there is a lack of a unified environment
to provide support for all phases of application development. Our goal
is to help the development of Grid applications, being aware of the
specificities of this kind of application. Initially, we will focus on
Integrade, providing to the application developer an Eclipse-based tool
that unify in a single environment tasks such as design, coding, test,
deployment, and monitoring of Grid applications.
This tool will be an Eclipse plug-in that, among other things, would
both generate code based on high-level specifications and check the
consistency of user code. A list of current functional requirements is
located at
http://www.ime.usp.br/~eguerra/toolForInteGradeApplications.html.
Acessing Computers Behind NATs and Firewalls
-
Antonio Carlos Theophilo Costa Junior (Masters student at PUC-Rio)
Nowadays
is very common to find computers that are connected in LANs (Local Area
Networks) protected by firewalls. Similarly these machines are often in
a environment where NAT (Network Address Translation) is used. These
two aspects (firewalls and NAT) represent a problem for the the
development of P2P (peer-to-peer) and Grid applications. The problem,
also faced in project
Integrade, essentially is due to the fact that
firewalls forbid certain kinds of conections and NAT works with IP
address not uniquely identified in the Internet.
This work aims
at investigating means for enabling communication between machines
shield by firewalls and/or NAT and others machines that form the
InteGrade network. With this work we intend to allow the admission of a
very large number of machines in the production phase of the Project.
Because
InteGrade Project uses CORBA as communication layer, our
intended way to solve this problem is to modify the CORBA's transport
layer to use JXTA (
www.jxta.org). The last uses HTTP protocol and
proxies to solve the problem of firewalls while the NAT question is
addressed with a new way of identification broader than IP address.
Flexible and Easy to Use Middleware
-
Renato Maia (Post-Doctoral researcher at PUC-Rio)
This work focus on the development middleware for distributed applications using the specific features of the dynamic language Lua to provide an implementation that is both easy to use and adapt to different uses. One of the practical results of this work is the OiL middleware, a simple and efficient ORB implementation in Lua that provides interoperability with CORBA as well as support for other RMI technologies.
Constraining resource usage at user level
-
Valéria Quadros (PhD student at PUC-Rio)
General-purpose Operating Systems do not provide effective mechanisms for application resource reservation. To overcome this limitation, some initiatives aim at guaranteeing processing by instrumenting kernels or by isolating the performance through the creation of virtual machines. However, these practices may be infeasible if the kernel code is not available or high overheads should be avoided. Taking a different approach from modified kernels and virtualization, this work focus on developing tools to constrain the usage of application resources (CPU and disk, for instance) at user level. Using low-level primitives provided by ordinary operating systems, these tools are able to change processes' priorities and to intercept system calls in order to enforce a usage pattern. An example of this kind of tool is CPUReserve, a service-based mechanism for isolating performance in modern computational scenarios such as opportunistic grids, utility-computing environments, and multi-processed architectures.
Ricardo Couto Antunes da Rocha Ph.D. Thesis
-
A middleware architecture for ubiquitous provision of distributed and heterogeneous context information (January 2009)
This thesis proposes a new architecture
for the management of distributed, dynamic and heterogeneous context. The
approach underlying the architecture is based on the concepts of context
domains and context interest of variable scope. Based on such a
middleware, mobile applications that require context-awareness can keep
continuous access to the context information, even when the client moves
between different context domains, each with its specific context model
and sources.
Gustavo Baptista M.Sc. Dissertation
-
MD-ECI: A Publish/Subscribe system supporting mobility and disconnection of terminals and NAT/firewalls traversal. (February 2009)
The MD-ECI is an extension of ECI - the original Pub/Sub interface of the
MoCA middleware - which uses the Session Inititiation Protocol (SIP) for
transparently handling dynamic changes of IP addresses of terminals. It
also minimizes the loss of notifications during periods of Pub/Sub client
disconnections. Through the optional use of a Media Proxy, the MD-ECI also
allows Pub/Sub clients - both event producers or consumers - to execute
behind firewalls with NAT.
URL:
http://www.lac.inf.puc-rio.br/moca/mdeci/mdeci.htm