Parallel and Distributed Computing: BOINC Grid Implementation Paper

1

Parallel and Distributed Computing
BOINC Grid Implementation
˜
Rodrigo Neves, Nuno Mestre, Francisco Machado, and Joao Lopes

Abstract—With the development of communications and Internet, distributed computing became an everyday reality for everyone
rather than just for a limited group of IT specialists and investors. This development allowed the emerging of several new computational
concepts, some even at the cost of non-consensus. This paper intends to make a brief approach at some of the current paradigms
like cloud and grid computing, peer-to-peer and client-server methods. Afterwards, it will go deeper into a detailed review of the Public
Resource Computing concept and the BOINC software implementation. Finally, a case study on the Extended BOINC System created
by Søttrup and Pederson [18] will present a simple solution to integrate the concepts of PRC and Grid in order to provide simple,
scalable, volunteer based computer power to process QoS dependent jobs.

Index Terms—parallel computing, distributed computing, grid computing, cloud computing, client server, peer to peer, boinc, extended
boinc, public resource computing, quality of service

!

1 I NTRODUCTION work hardware, operating systems and programming
languages, the term middleware has been created. It is
D ISTRIBUTED systems have been given many defini-
tions throughout the years but none of these has
been consistent with each other.
a software layer that provides abstraction, setting up a
uniform computational model for software developers to
As Andrew S. Tanenbaum and Maarten van Steen work on. One of the most widespread middleware soft-
suggested [1]: wares available is the Common Object Request Broker
(CORBA) [2].
A distributed system is a collection of independent
computers that appears to its users as a single
coherent system. 2 I MPORTANT C ONCEPTS
The first part of this definition deals with hardware In order to better understand the remaining of this
while the second focus specially on software. paper, it might be interesting to clear up some technical
Distributed systems must present a transparent work- nomenclature in the Parallel and Distributed Systems
flow to the user regardless of the differences in hardware environment.
and communication methods throughout the connected • Servers: Typically high-powered workstations,
computers. Other important characteristics of these sys- minicomputers or mainframes that hold the infor-
tems are the ease of scalability and the high availability. mation and provide it to clients through request
Such systems should easily connect users to resources handling;
while hiding the fact that these may be distributed across • Clients: Computers or mainframes that request
a network. It should also be open and respect the avail- services from the servers;
able standards in order to facilitate future development • API: An Application Programming Interface is a
and scalability and ensure the data and communication tool used to provide the end-user with an abstract
security. All these specific issues shall be addressed in set of operations. Its main attraction is the ability
further detail later on this paper. to hide from the user the implementation of such
In order to create the illusion of a single system operations;
both at high-level, for users, and at low-level, for net- • SDK: The Software Development Kit is typically
a set of development tools that allows a software
• R. Neves is with MIEET, Departmento de Engenharia Electr´ nica e o
Inform´ tica, Faculdade de Ciˆncia e Tecnologia, Universidade do Algarve.
a e
engineer to create applications for a certain software
E-mail: a25067@ualg.pt package. Usually a SDK implements an API;
• N. Mestre is with LEI, Departmento de Engenharia Electr´ nica e In-
o • Middleware: A software layer that conceals all
form´ tica, Faculdade de Ciˆncia e Tecnologia, Universidade do Algarve.
a e
E-mail: a28997@ualg.pt
the heterogeneity in a system in order for software
• F. Machado is with LEI, Departmento de Engenharia Electr´ nica e In-
o developers to easily work on;
form´ tica, Faculdade de Ciˆncia e Tecnologia, Universidade do Algarve.
a e • Query Languages: Techniques and protocols to get
E-mail: a28994@ualg.pt
• J. Lopes is with LEI, Departmento de Engenharia Electr´ nica e Inform´ tica,
o a
the sough information being the most famous the
Faculdade de Ciˆncia e Tecnologia, Universidade do Algarve.
e Structured Query Language (SQL);
E-mail: a27981@ualg.pt • Virtual Organization: A dynamic set of individ-
uals and/or institutions defined by a batch of

2

resource-sharing rules. Those rules need to make else users will eventually evade the security in favor
perfectly clear just what is shared, who is allowed of productivity.
to share, and the conditions under which sharing
occurs [9]; 3.3 Scalability
• Public Resource Computing: Usually known as
Scalability is one of the most important design goals for
PRC, this concept describes the idea of getting any-
developers in distributed systems. According to Clifford
body with an Internet connection and spare com-
Neuman [5], a system’s scalability may be weighted
pute power to donate CPU cycles on their computer
along three main dimensions:
for a greater project.
• Size: Whether it is simple to add more users and
resources to the system or not;
3 D ISTRIBUTED S YSTEMS C HALLENGES
• Geographical: Regarding the location of users and
3.1 Connecting Users and Resources / Concurrency resources;
Resource sharing and distribution in any given multi- • Administration: Ease to manage the system even
user system is always a main concern. In the case of dis- when it is shared by multiple administrative orga-
tributed systems this task becomes of utter importance. nizations.
In a distributed system, both applications and services As a system scales up, in any of these three dimen-
provide resources that can be shared among clients. sions, it may exhibit problems that could affect perfor-
These usually allow multiple client requests to be ac- mance.
cepted even though they may be processed one at a time. Size scalability problems are related with the increased
If we consider each resource being encapsulated as amount of users and their demands. Such situation
an object and the requests being treated as concurrent occurring on centralized services, data and algorithms
threads, it becomes clear that any application or service will eventually create a bottleneck at the server.
must be carefully managing this concurrency in order to Geographical scaling problems are usually related to
avoid inconsistent results and/or deadlocks. the connectivity limitations of the communications. In
Though at a first glance this situation may require wide-area networks, access to information is usually
careful analysis and programming, it’s cost effectiveness made through unreliable connections and virtually al-
may become obvious when sharing expensive resources ways a point-to-point connection.
like high performance processing and data structures. Administrative scalability issues often relate to mul-
tiple domain trustworthy security certificates. As more
3.2 Transparency / Heterogeneity organizations join the network, strict administration and
One of the major concepts and advantages of distributed management protocols must be defined as all included
systems is the transparency. It allows a user to access parts should play an active role in this processes.
all resources and data that may be scattered around
a network, without having to worry where these are 3.4 Security
actually located. This may prove to be a problem when Usually, the information resources available and main-
an virtually infinite combination of hardware resources tained in a distributed system have a high intrinsic
and operating systems are involved. This ability to pro- value to the users, therefore the security of that data is
vide transparency within all the heterogeneity of the dis- considerably important.
tributed system is usually implemented by a middleware There are three main security components that should
software layer as described on the introduction. be taken into consideration:
The concept of transparency may be applied to many
• Confidentiality: Protection against unauthorized
aspects of distributed systems [1]:
accesses;
• Access: Hide differences in data representation and
• Integrity: Protection against unauthorized alter-
how a resource is accessed; ation or corruption;
• Location: Hide where a resource is located;
• Availability: Protection against disruption of the
• Migration: Hide that a resource may move to communication with the resources.
another location;
Preventive actions should be taken in order to ensure
• Relocation: Hide that a resource may be moved to
this security parameters, like the use of a firewall and
another location while in use;
careful code development.
• Replication: Hide that a resource is replicated;
• Concurrency: Hide that a resource may be shared
by several competitive users; 3.5 Openness
• Failure: Hide the failure and recovery of a resource; According to Kazi Farooqui, Luigi Logrippo and Jan de
• Persistence: Hide whether a (software) resource is Meer [3], openness is the combination of the previous
in memory or on disk; stated characteristics. Sometimes however, it is generi-
• Security: Negotiation of secure access to resources cally referred to as the level of respect for standards that
must require a minimum of user intervention, or define the syntax and semantics of the provided services.

3

time. Unlike the Client-Server approach, P2P systems
rely on an agglomeration of applications and systems al-
together in order to have access to distributed resources
on a decentralized way. In terms of maintenance costs,
P2P has the edge over Client-Server since the contents
and systems are maintained individually by each peer.
P2P systems can be categorized into three main
groups: centralized, decentralized and hybrid implemen-
tations.
The centralized P2P concept bases its implementation
on a central server that executes simple generic functions
for the system, like work load scheduling and result
validation.
Decentralized P2P systems rely completely on the
peers themselves to perform all the functions without
Fig. 1. Generic client-server environment intervention of centralized servers. In these cases, all the
result validation and communication must be handled
by each node and coordinated with its peers.
This important concept, despite of the interpretation, Hybrid systems are a bit of a mix between the pre-
provides the developers with the needed flexibility to vious two implementations. These bring up the concept
add, configure and integrate new components and ser- of super-nodes, made of several regular nodes, which
vices. Also, this flexibility ensures that the addition and are responsible of the centralized work while it’s con-
removal of components can be made without affecting stituents can still work on the regular decentralized way.
the stability of the overall system.

4.3 Cloud Computing
4 D ISTRIBUTED S YSTEM M ODELS
A Distributed System is a set of loosely coupled re- Nowadays, data storage and programs are being swept
sources interconnected by a communication network. from the desktop computers and corporate server rooms
[4] and installed in the computer cloud. Cloud computing
emerges from the lesser need of the users to have
applications installed on their machines due to increased
4.1 Client-Server communications speed and availability.
Client-Server is a particular type of distributed systems Every operating system update cascades into a batch
design that clearly distinguishes the relationship be- of time and resource consuming software revisions. Out-
tween two computers. The Server provides some kind of sourcing computation through Internet based services
service, such as processing database queries or sending significantly reduces these costs while offering a whole
out current stock prices. The client uses the services that new set of advantages like mobility and cooperation.
are provided by the server, either displaying database The amount of services and applications provided in
query results to the user or making stock purchase the cloud is growing every day and cannot be considered
recommendations to an investor. as a bunch of simple tools anymore. Companies are
The communication that occurs between the client and starting to acquire cloud based services for every kind
the server must be reliable. That is, no data can be of managerial and business oriented tasks.
dropped and it must arrive on the client side in the Growing voices of worry have been expressing their
same order in which the server sent it. In order to ensure concerns about data privacy and confidentiality not tak-
the reliability between Server and Client the communi- ing the service’s privacy policies as creditable enough.
cation uses the TCP/IP protocols. The Internet Protocol One famous scenario used as argument is cited by Hayes
(IP) suite is a set of communication tools that regulate [17]:
communication on the Internet and most commercial
(...) a government agency presents a subpoena or
networks. The Transmission Control Protocol (TCP) is
search warrant to the third party that has posses-
one of the core protocols of this suite. Using TCP, clients
sion of your data. If you had retained the physical
and servers can create connections to one another, over
custody, you might still have been compelled to
which they can exchange data in packets.
surrender the information, but at least you would
have been able to decide for yourself whether or
4.2 Peer-To-Peer not to contest the order. The third-party service
The concept of Peer-To-Peer communication, also known is presumably less likely to go to court on your
as P2P, is based on the idea that each individual node behalf. In some circumstances you might not even
(peer) in the network is both client and server at the same be informed that your documents have been released.

4

These kind of issues will probably never be solved 4.4.1.1 Applying the three point checklist: Follow-
in time to stop or control the growth of the cloud as ing there are some practical examples to help us make
we see major software developers and investors (Apache the concept of grid clear, according to Ian Foster’s three
Foundation, Amazon, Adobe, Google, IBM, etc.) trying point checklist.
to keep up with the evolutionary pace. • Sun Grid Engine:
The Grid Engine project is an open source commu-
4.4 Grid Computing nity effort to facilitate the adoption of distributed
4.4.1 Definition computing solutions. [12]
The term grid was first used as a metaphor to the This system delivers quality of service when in-
electric power grid. The intended idea was that access stalled on a parallel computer or local area network.
to computation and data should be as easy, pervasive However, its complete knowledge of system states
and standard as plugging in an appliance into an outlet. and user requests, as well as control over individual
Nowadays it is hard to find a consensual definition. Here components, implements a centralized management
are some from the most referred authors: system that makes this system fail the first point of
Foster’s checklist.
A computational grid is a hardware and software
• The Web: The Web is open and its general-purpose
infrastructure that provides dependable, consistent,
protocols support access to distributed resources,
pervasive and inexpensive access to high-end
however it fails to coordinate those resources to
computational capabilities. [6]
deliver interesting quality of services.
• TeraGrid:
Computational grid is the technology that enables
TeraGrid is an open scientific discovery infrastruc-
resource virtualization, on-demand provisioning and
ture combining leadership class resources at eleven
service (resource) sharing between organizations. [7]
partner sites to create an integrated, persistent com-
putational resource. [13]
Grid computing has the ability, using a set of
This system integrates resources from multiple insti-
open standards and protocols, to gain access to
tutions, each with their own policies, uses open and
applications and data, processing power, storage
general-purpose protocols to negotiate and manage
capacity and a vast array of other computing
sharing and addresses multiple quality of service
resources over the Internet. A grid is a type of
dimensions, therefore fully fits Ian Foster’s three-
parallel and distributed system that enables the
point checklist.
sharing, selection and aggregation of resources
distributed across ”multiple” administrative 4.4.2 Usual Features
domains based on their (resources) availability,
In this section we thrive to describe in detail some of
capacity, performance, cost and users’ quality-of-
the most important features usually associated with the
service requirements. [8]
grid computing method.
4.4.2.1 Volunteer Computing: Most grids use vol-
The problem that underlines the Grid concept is unteer resources, that is, resources contributed to the
coordinated resource sharing and problem solving in grid by anonymous individuals or organizations with
dynamic, multi-institutional virtual organizations. no profit intended.
[9] 4.4.2.2 Geographically dispersed: Due to its archi-
tecture and volunteer characteristic some grid resources
Although we can find some common ground, like re- can be spread throughout the globe.
source sharing and processing power, these are features 4.4.2.3 Idle Resources: One of the benefits of using
not only of a grid computing system but of any kind grid computing is that you can exploit resources with
of distributed system. Due to the lack of consensus and low usage rates because in most organizations, there are
the use of the term ”grid” as a marketing slogan (science large amounts of under utilized computing resources. Most
grid, access grid, knowledge grid, bio grid, campus grid, desktop machines are busy less than 5% of the time over a
commodity grid, etc.), Ian Foster suggested a checklist to business day [14].
define what is and is not a grid in his paper ”What is 4.4.2.4 Inexpensive: Either through volunteering
the grid? A three point checklist” [11]: or by using idle resources within a company, usually
A grid is a system that: it’s possible to reach a considerable computational per-
1) coordinates resources that are not subject to formance without major investments in supercomputers
centralized control (...) or clusters.
2) (...) using standard, open, general-purpose pro-
tocols and interfaces (...) 4.4.3 Architecture
3) (...) to deliver nontrivial quality of service. In an effort to standardize grid architectures, I. Foster, C.
Kesselman and S. Tuecke [9] presented an open five-layer

5

4.4.3.4 Collective Layer: The Collective layer con-
tains protocols and services, like APIs and SDKs, which
are not associated with specific resources but global
in nature, and capture interactions across collections of
resources. Meaning that, at this layer, individual resource
architectures and functionalities are abstracted in order
to provide collective functions that can be implemented
as persistent services with associated protocols, or as
SDKs and APIs, designed to be linked to applications.
4.4.3.5 Application layer: This is the final layer,
therefore it is responsible to provide the end user with an
Fig. 2. Layered Grid Architecture abstract interface that includes the user applications and
functionalities that operate within a Virtual Organization
environment. These applications are constructed using
services defined at any layer.
structure (Application, Collective, Resource, Connectiv-
ity and Fabric layer) based on the ”hourglass model”
[10]. 4.5 Quality of Service
In this architecture the narrow neck, represented by In the multimedia communities, Quality of Service (QoS)
the Resource and Connectivity layers, defines a small set issues are geared to provide a client with an acceptable
of protocols onto which many high-level behaviors, used level of presentation quality when accessing content.
by the Application and Collective layer, can be mapped Network QoS deals specifically with providing certain
(the top of the hourglass). On the other hand, the ”neck” quality levels for network link characteristics between
protocols can themselves be mapped onto many different two points. These characteristics are expressed in terms
underlying technologies (the base of the hourglass, the of delay, jitter, packet loss rate and throughput.
Fabric layer) (Figure 2). Unlike multimedia and network QoS, Grid QoS re-
4.4.3.1 Fabric Layer: This layer provides the requires a central information service for up-to-date infor-
sources to which shared accesses are mediated by Grid mation on resources available for use by others. Such
protocols. Fabric layer works with resource-specific op- information can be interrogated by an application user
erations, there is no abstraction at this level. These to determine which resources can be used to execute an
operations are usually a result of sharing operations at operation. In Grid computing, QoS management focus
higher layers. There’s interdependency between func- on providing assurance on resource access while main-
tions implemented in this layer and sharing operations taining the security level between domains [22].
supported by the Grid. A rich and more complex Fabric
set of functionalities enables more sophisticated sharing 4.5.1 QoS in Grid Computing
operations. As opposed, fewer functionalities and de- Once the Grid applications submit their requirements to
mands at this layer imply a simplified Grid structure. the management services that schedule jobs as resources
4.4.3.2 Connectivity Layer: The connectivity layer become available, these must support a resource man-
defines the core communication and authentication pro- ager or scheduler that can receive requests from external
tocols. Communication protocols enable the exchange of applications. Nevertheless, there are several applications
data between Fabric layer resources, requiring transport, that need to get results for their tasks within strict
routing and naming mechanisms. Authentication proto- deadlines. Consequently they cannot wait for resources
cols are used to ensure the security and identity of users to become available so, it is necessary to reserve Grid
and resources. Due to complex security problems and resource and services in a particular time. In order
wide usage, existing protocols and standards should be to handle complex scientific and business applications,
used to implement this layer. other features are highly desirable, sometimes even re-
4.4.3.3 Resource Layer: Resource layer provides quired.
the means to share single resources using information A Grid resource management system tries to address
and management protocols. Information protocols are the following QoS issues [22]:
used to obtain details about the structure and state of • Advanced Resource Reservation: It’s important
a single resource. Management protocols are used to when dealing with scarce resources, as is often the
negotiate access to the shared resource, specifying its case with end resources made available on the Grid.
requirements and the operations to be performed. These Should support mechanisms for advance, immedi-
protocols should also ensure that requested operations ate, or on-demand resource reservation.
respect individual policies of each resource. This layer is • Reservation Policy: The system should have mech-
only concerned with individual resources. Global state anisms that provide the Grid resource owners ways
and atomic actions over multiple resources are issues of of enforcing their policies by governing when, how,
the next layer. and who can use their resource.

6

• Agreement Protocol: The system should inform
the clients of their advance reservation status, and
the resource quality they should expect during the
service session.
• Security: The system should prevent malicious
users penetrating or altering data repositories that
hold information about reservations, policies and
agreement protocols.
• Simplicity: The QoS enhancement should be rea-
sonable and simplistic in design so that it requires
minimal changes to be made to existing computa-
tion, storage or network infrastructure.
• Scalability: The approach should be scalable to a
large number of entities, since the Grid is a global-
scale infrastructure.

5 BOINC Fig. 3. BOINC Infrastructure
Berkeley Open Interface for Network Computing
(BOINC) is an open middleware platform that supplies
the scientific research by placing their confidence in temporary input and output files on the server. Only
using resources donated by simple personal computers the files are deleted, the entries on the database are
around the world (core clients). The objective is to gather kept and therefore it is possible to find information
all this energy and make a supercomputer which helps even after the project is completed.
the researchers (clients) in several projects. [18] • The ”feeder” is used to enhance the schedulers per-
Using BOINC allows user to achieve high processing formance and to reduce the queries to the database.
performance with low costs. For example, to have 100 It does so by placing WUs, from the database, into
TFlops available for one year, Amazon’s Elastic Com- a shared memory.
puting Cloud costs 175 million dollars, to build a cluster • The ”database purger” removes work-related
you need 12.4 million dollars, but with BOINC, clients database entries when they are no longer needed
only need, in average, 125,000 dollars [19]. On the other in order to keep the database from growing into an
hand, looking at the top500 supercomputers list (Novem- unpractical size.
ber/2008) [21], roadrunner achieves 1105 TFlops, where
BOINC has a daily average of 1,700 TFlops and the most 5.1.2 Scheduler
relevant project (SETI@Home) has a daily average of 615 The scheduler is a CGI software that runs every time
TFlops. a client connects to a project and asks for work. It has
to compare the available WU’s needs with the clients’
5.1 Infrastructure shared resources in order to match them.
BOINC provides a set of tools, daemons, scheduler and 5.1.3 Database
database (Figure 3).
The BOINC database is a MySQL database that stores in-
5.1.1 Daemons formation about registered users and hosts, applications
and their versions, WU’s and their results, and other
BOINC servers use daemons to manage and keep track relevant information.
of their jobs or, in BOINC terms, work units (WU).
• The ”work generator” has to generate WU and
correspondent input files. 5.2 PRC vs. Grid
• The ”transitioner” has to control and change the According to David P. Anderson [23], both PRC and Grid
states of each WU. Computing methodologies share a common goal: to use
• The ”validator” has to validate the results of each the existing resources in the best possible way. There are
WU and its redundant copies. however some important differences between the two.
• The ”assimilator” daemon regroups the final results While a Grid is usually managed and controlled by
and processes them according to the administrator’s a single organization, a PRC network relies on separate
specification. It could zip and e-mail the results or individuals to share their resources. While this particu-
automatically do post processing and store those larity may allow a huge growth in terms of connected
results on a magnetic tape. nodes, it brings out other liabilities like the unreliability
• The ”file deleter”, as the name indicates, checks for of the processed results and uncertain processor time.
completed and assimilated WUs and then deletes Since each user is volunteer and therefore allowed to

7

Manage their states;
•
Pull the results from the grid resource broker.
•

These results don’t need validation because the re-
sources from the Grid are considered trustworthy.

6 C ONCLUSION
Throughout the development of this paper, we scanned
the currently available paradigms on distributed com-
puting, their main advantages and limitations. This
process has taken us through part of the history of
computing as we start from the traditional client-server
model and develop it until nowadays cloud and grid
computing concepts.
The evolution of communications and commodity per-
sonal computers brought the distributed computation
to a whole new level while approaching people and
Fig. 4. Extended BOINC Infrastructure science through Public Resource Computing. This new
area of computational resource sharing brought the need
to rethink the Quality of Service requirements in dis-
control the amount of work done, nothing ensures the
tributed networks. PRC proved that a huge ammount
project management that this user will be cooperating
of volunteers can supply an unbeatable system-wide
for a long time or keeping a steady work-flow.
throughput without the need of strict QoS policies. For
Another important difference between the two meth-
instance, the major BOINC based project, and also the
ods relies on the Quality of Service. It is virtually impos-
one that encouraged it’s development, SETI@Home has
sible to ensure a strong QoS on a PRC network due to
produced so far 3 Million+ years of processing time. [20]
slow connections and low availability.
For the reduced ammount of research projects and
processing jobs that require strict deadlines, high com-
5.3 Extended BOINC System munication speed and permanent connectivity there was
As we have seen BOINC only implements two out of the a need to merge the benefits of PRC globalization and
three points in Foster’s checklist. It has a decentralized the QoS that a Grid system could provide. In order to
control over resources and it uses open protocols and address this need, the bridge connector model [18] was
interfaces, but it fails to deliver non-trivial quality of developped to extend the regular BOINC System.
service, because it does not fulfill the information ac-
cessibility and connectivity requirements. R EFERENCES
With this in mind, Søttrup and Pederson [18], sug-
[1] A. S. Tannenbaum and M. van Steen, Distributed Systems - Principles
gested a bridge between BOINC and a private Grid, and Paradigms, International Edition, Pearson, U.S.A.: Prentice Hall,
where instead of clients pulling the jobs from BOINC 2002.
server by connecting to it, the server connects to a [2] G. Coulouris, J. Dollimore and T. Kindberg, Distributed Systems
- Concepts and Design, Fourth Edition, Pearson, U.K.: Addison-
resource broker, responsible for scheduling, submitting Wesley, 2005.
jobs to remote machines, transferring files and logging, [3] K. Farooqui, L. Logrippo, J. de Meer, The ISO Reference Model for
on the Grid and pushes jobs into it. This Grid will be re- Open Distributed Processing - An Introduction, February 14, 1996
[4] A. Silberschatz, P. B. Galvin, and G. Gagne, Operating System
sponsible for providing quality of service and therefore, Concepts, Seventh Edition, Wiley, U.S.A.: John Wiley & Sons, 2005,
together they would build a Foster’s Grid (Figure 4). Page 611.
[5] B. C. Neuman, Scale in Distributed Systems, Readings in Distributed
Computing Systems, IEEE Computer Society Press, 1994
5.3.1 BOINC to Grid architecture [6] I. Foster, C. Kesselman, The Grid: Blueprint for a New Computing
Infrastructure, University of Michigan, U.S.A.: Morgan Kaufmann
In order to process the WU’s directly and for the speci- Publishers, 1999
fications of the Grid, a new daemon has to be created [7] P. Plaszczak, R. Wellner, Grid computing: The Savvy Manager’s Guide,
(bridge daemon) and some of the other have to be U.S.A.: Elsevier/Morgan Kaufmann, 2005
modified. The transitioner daemon must be adapted so [8] IBM Solutions Grid for Business Partners: Helping IBM Business
Partners to Grid-enable applications for the next phase of e-business on
that it wont change the state of the WUs sent into demand, U.S.A.:IBM, 2002
the Grid. Manipulating and controlling the several WU [9] I. Foster, C. Kesselman, S. Tuecke, The Anatomy of the Grid: En-
states is now the bridge daemon’s responsibility. The abling Scalable Virtual Organizations, International J. Supercomputer
Applications, 2001
feeder daemon also needs to be modified so that it wont [10] L. Kleinrock, Realizing the Information Future: The Internet and
load the WUs, intended to the Grid, into the shared Beyond, National Research Council, U.S.A.: National Academy
memory. This way, the bridge daemon has to: Press, 1994
[11] I. Foster, What is the Grid? A Three Point Checklist, GRIDToday, July
• Push WUs into the Grid; 20, 2002

8

[12] Grid Engine Project, http://gridengine.sunsource.net,
SunSource.net, Sun, June 6, 2009
[13] About TeraGrid, http://www.teragrid.org/about, TeraGrid, Na-
tional Science Foundation, June 6, 2009
[14] B. Jacob, M. Brown, K. Fukui, N. Trivedi, Introduction to Grid Com-
puting, First Edition, International Business Machines Corporation,
U.S.A.: IBM, International Technical Support Organization, 2005,
Page 8
[15] P. Roy, Operating Systems: Internals and Design Principles, 6/E
William Stallings, Manatee Community College, U.S.A.: Prentice
Hall, 2008
[16] Introduction to Distributed System Design,
http://code.google.com/intl/pt-PT/edu/parallel/dsd-
tutorial.html, Google Code University, 2009
[17] B.Hayes, Cloud Computing, Communications of the ACM, ACM,
July 2008, Pages 9-11
[18] C. U. Søttrup, J. G. Pedersen, Developing Distrubited Computing
Solutions: Combining Grid Computing and Public Computing, M. Sc.
Thesis, Department of Computer Science, University of Copen-
hagen, March 1, 2005
[19] BOINC Documentation Project, Why Use BOINC?,
http://boinc.berkeley.edu/trac/wiki/WhyUseBoinc, University
of California, June 11, 2009
[20] J. Koulouris, The Big BOINC ! Projects and Chronology Page,
http://www.angelﬁre.com/jkoulouris-boinc, June 11, 2009
[21] Top500.org, Top500 List - November 2008,
http://www.top500.org/list/2008/11/100, Top500 Supercom-
puting Sites, Top500.org, June 11, 2009
[22] R. J. Al-Ali, K. Amin, G. von Laszewski, O. F. Rana, D. W. Walker,
M. Hategan, N. Zaluzec Analysis and Provision of QoS for Distributed
Grid Applications, Kluwer Academic Publishers, 2004
[23] D. P. Anderson, BOINC: A System for Public-Resource Computing
and Storage, Proceedings of the Fifth IEEE/ACM International
Workshop on Grid Computing, 2004

Parallel and Distributed Computing: BOINC Grid Implementation Paper

More Related Content

What's hot (18)

Viewers also liked (15)

Similar to Parallel and Distributed Computing: BOINC Grid Implementation Paper (20)

Recently uploaded (20)

Parallel and Distributed Computing: BOINC Grid Implementation Paper