Condensed Graph applications executed by WebCom-G receive all the efits of the WebCom-G system including transparent support for fault toler-ance, load balancing, scheduling and security
Trang 1Program Execution in WebCom-G
Figure 1 Annotation and Extraction.
Extraction is a process of translating higher level specifications into densed Graph representation This process is suitable for specification lan-guages such as the Globus Resource Specification Language (RSL)[12], forexample RSL specifies the list of tasks to be executed and their associatedconfigurations During extraction, tasks specified in the RSL will be expressed
Con-as nodes in a Condensed Graph In addition, the tCon-ask sequencing constraintsspecified in the RSL script are represented as arcs in the resulting CondensedGraph For extraction, this Condensed Graph can be specified as an XMLdocument WebCom-G can dynamically load and execute Condensed Graphsspecified in XML
Condensed Graph applications executed by WebCom-G receive all the efits of the WebCom-G system including transparent support for fault toler-ance, load balancing, scheduling and security Hence, tasks extracted fromRSL scripts also benefit from these WebCom-G features For example, if the
ben-2.
Within WebCom-G task sequencing is expressed as a Condensed Graph[13].Task execution is carried out using appropriate Execution Engine modules.Support for legacy applications is provided by compiling or translating existingcode into an intermediate representation expressed as a Condensed Graph This
support is provided by two methodologies called Extraction and Annotation.
Extraction
Trang 2Figure 2 The process of generating a Condensed Graph from sequential code.
job should fail to execute, WebCom-G will reschedule the job for execution at
a later time
A different approach has to be adopted for traditional high level languageslike C, C++ and Java These languages are typically not optimised for execu-tion in distributed environments, as they normally consist of sequential code.Attempting to parallelise sequential code is not trivial A Condensed Graphscompiler is used to parallelise sequential applications, Figure 2 This com-piler converts existing code into an XML representation of the associated Con-densed Graph The compiler takes existing code and performs a data depen-dency analysis, using well known compiler techniques such as those described
in [2] This is illustrated in Figure 3
This analysis identifies parallelisable data blocks within the source code
This translation process can be described as either fully automatic or semi tomatic translation.
Trang 3Figure 3 Example translation of sequential C code to Condensed Graphs(CG) XML sentation The C program is analyzed to produce an Abstract Syntax Tree(AST) representation Applying CG rules to the AST results in the XML representation.
repre-Figure 4 Manually optimising data block dependences obtained from the compiler.
Trang 4For fully automatic translation, using appropriate interpretation rules, theidentified data blocks are converted into a Condensed Graphs application, suit-able for execution on WebCom-G.
For semi automatic translation, the uncovered data blocks are presented tothe programmer, via the WebCom-G Integrated Development Environment.This facilitates the further optimisation of data blocks by the programmer, ifpossible This is outlined in Figure 4
Annotation
Annotation is a mechanism for allowing programmers to identify parallelblocks within their source code This provides the programmer the opportu-nity to optimise their source code and hence the Condensed Graph obtainedvia the fully automatic extraction mechanism outlined previously This mech-anism may be used for high level languages as well as proprietary specificationlanguages
The Extraction(CG) Compiler being developed, will expose available lelism using a combination of the Extration and Annotation mechanisms out-lined in Section 2
paral-The compilation process depicted in Figure 2 comprises four stages: sourcecode analysis, source code restructure, data dependency analysis using theCondensed Graph rules, and the generation of parallel code in CondensedGraph representation
The CG compiler will attempt to fully automate the procedure of ing traditional source code into a Condensed Graphs representation, capable ofbeing executed in parallel This parallelising compiler inserts all the necessary
transform-CG information into its intermediate representation
The compiler, depicted in the Figure 2, accepts as input a source code Thisinput will be parsed to produce an Abstract Syntax Tree (AST)[21] The ASTrepresents the syntactic structure of the source code in a tree format This treestructure will then be converted into block representation and subsequentlyflow graph representation Block representation helps to identify the blocks ofprogram structure within the source code
The Parser module consists of Lexical Analyser and Parser The LexicalAnalyser scans the source code to identify tokens The Parser takes the tokensproduced by the analyser and produces a syntax tree corresponding to prede-fined Grammar rules ANTLR[1] is used to generate the AST from higher levellanguages such as C, C++ and Java
The flow graph representation outlines the program flow and identifies thedata blocks that may be parallelised within the Condensed Graph representa-
Trang 5Solutions must be developed to free application programmers from the lowlevel complexity of parallel programming in Grid environments In this paper,the WebCom-G Programming Environment is presented This environmentsupports the execution of existing applications on WebCom-G The program-mer is freed from the complexities of creating or employing complicated dis-tributed computing systems in order to develop solutions to problems.
Different mechanisms for application execution were presented, rangingfrom the extraction of parallelisable tasks from scripting language files, to an-notating preexisting native codes Using compilation techniques, data blocksobtained by using data dependencies are converted into Condensed Graphs for-mat and executed by WebCom-G
The goal of WebCom-G is to hide the Grid, by providing a vertically tegrated solution from application to hardware while maintaining interoper-ability with existing Grid technologies In addition to maintaining a verticallyintegrated solution, the available services will be exploited to increase func-tionality and effect interoperability The provision of such a Grid OperatingSystem will remove much of the complexity from the task of the applicationdeveloper
in-153tion Control flow, Data flow and Data dependency analysis is performed onthe flow graph to generate the Condensed graph representation, that will sub-sequently execute on WebCom-G
Cluster-Bor Yuh Evan Chang Iktara in ConCert: Realizing a Certifirf Grid Computing work from a Programmers Perspective School of Computer Science, Carnegie Mellon
Frame-University, Pittsbourghm June 2002, Technical Report: CMU-CS-02-150.
Karl Czajkowski, Steven Fitzgerald, Ian Foster, and Carl Kesselman Grid Information Services for Distributed Resource Sharing Proceedings of the 10th IEEE International
Symposium on High Performance Distributed Computing.
Trang 6D H J Epema, Miron Livny, R van Dantzig ans X Evers, and Jim Pruyne A
World-wide Flock of Condors: Load Sharing among Workstation Clusters Journal on Future
Generations of Computer Systems, Volume 12, 1996.
Steven Fitzgerald, Ian Foster, Carl Kesselman, Gregor von Laszewski, Warren Smith, and
Steven Tuecke A Directory Service for Configuring High-Performance Distributed
Com-putations Proceedings of the 6th IEEE International Symposium on High Performance
Distributed Computing.
Message Passing Interface Forum MPI: A Message-Passing Interface Standard The
International Journal of Supercomputer Applications and High-Performance Computing, Volume 8, 1994.
I Foster and C Kesselman Globus: A Metacomputing Infrastructure Toolkit
Interna-tional Journal of Supercomputer Applications, 11(2):115-128, 1997.
I Foster and C Kesselman The Grid: Blueprint for a New Computing Infrastructure.
Published by Morgan Kaufmann Publishers inc ISBN: 1-55860-475-8.
Al Geist, Adam Beguelin, Jack Dongerra, Weicheng Jiang, Robert Manchek, and Vaidy
Sunderam PVM: Parallel Virtual Machine A Users’ Guide and Tutorial for Networked
Parallel Computing MIT Press, 1994.
Globus Globus RSL http://www.globus.org/gram/rsl_spec1.html.
John P Morrison Condensed Graphs: Unifying Availability-Driven, Coercion-Driven
and Control-Driven Computing PhD Thesis, Eindhoven, 1996.
John P Morrison, Brian Clayton, David A Power, and Adarsh Patil WebCom-G: Grid
Enabled Metacomputing The Journal of Neural, Parallel and Scientific Computation.
Special issue on Grid Computing Guest Editors: H.R Arabnia, G A Gravvanis and M.P Bekakos April 2004.
John P Morrison, James J Kennedy, and David A Power WebCom: A Volunteer-Based
Metacomputer The Journal of Supercomputing, Volume 18(1): 47-61, January 2001.
John P Morrison, David A Power, and James J Kennedy An Evolution of the WebCom
Metacomputer The Journal of Mathematical Modelling and Algorithms: Special issue on
Computational Science and Applications, 2003(2), pp 263-276, Editor: G A Gravvanis Platform Computing: Load Sharing Facility http://www.platform.com.
Portable Batch System http://www.openpbs.org.
David A Power, Adarsh Patil, Sunil John, and John P Morrison WebCom-G
Proceed-ings of the international conference on parallel and distributed processing techniques and applications (PDPTA 2003), Las Vegas, Nevada, June 23-26, 2003.
Rob V van Nieuwpoort, Jason Maassen, Rutger Hofman, Thilo Kielmann, and Henri E.
Bal Ibis: an efficient Java-based grid programming environment Proceedings of the
2002 joint ACM-ISCOPE conference on Java Grande, p 18-27, November 03-05, 2002, Seattle, Washington, USA.
D A Watt Programming Language Processors Prentice Hall International, Hemel
Trang 7GRID SOLUTION FOR E-MARKETPLACES INTEGRATED WITH LOGISTICS
L Kacsukné Bruckner1and T Kiss2
The evolution of business models and technological solutionsadvance together like intertwining spirals motivating and supportingeach other Business requirements drive information technology (IT) tofind new tools and techniques that make businesses develop new needsagain Early electronic commerce – Electronic Data Interchange (EDI) -was started because telecommunication between computers facilitated anew relationship between businesses The rise of new business needsresulted in new communication standards and protocols Real e-commerce was born out of the opportunity offered by the World WideWeb and triggered new IT researches again After the unrealistic hype ofthe 90’s and the crises around 2000 e-commerce by now has entered thereality phase where efficiency drives the businesses, Internet usage adds
Abstract:
Key words: e-business, logistics, e-marketplace, Grid service, legacy code
Trang 8value and increases the profitability of the companies (Plankett Research2004)
A main target area of seeking business efficiency is supply chainmanagement (SCM) Today a substantial part of supply chains aremanaged across the Internet still they contain a surprisingly high amount
of inefficiencies (Oliver at al 2002) Both business and technology sidesshould be revised to find ways of improvement New e-commercemodels might be considered and the latest information technology toolssearched for to support them
The Grid concept has been created for solving computation intensivescientific problems, but the possibility of business applications was soondiscovered The convergence between Web services and Gridcomputing, that was triggered by the specification of OGSA (Open GridServices Architecture) (Foster et al., 2002), resulted in even moreintensive interest from large industry players towards Grid-basedsolutions OGSA defines a Grid architecture that is based on Web servicestandards and protocols As Web services are becoming more and morecommon in business applications, a Grid architecture based on SOAP(Simple Object Access Protocol) communication and WSDL (WebServices Description Language) service descriptions is the natural model
to adopt in a business environment
This paper would like to provide a step forward both in the fields ofbusiness and technology As an answer to SCM problems a new, three-sided e-commerce model is suggested that integrates buyers, sellers andlogistics service providers in the same negotiation process Thismarketplace helps trading partners to minimise their costs and increasetheir profit The online optimisation requires large amounts ofcomputation without any delay, which has focused attention on Gridtechnology Following the direction set by Kacsukné (2004) and usingthe Grid-based e-marketplace architecture introduced by Kiss at al.(2004) this article gives a complex picture of the new e-marketplacemodel and its planned implementation
B2B e-marketplace models can be classified as buyer-oriented, oriented and intermediary marketplaces The first type is facilitated by acompany or consortium with big buying potential aiming at optimisingprocurement costs and is contributed by a large number of sellers Aseller-oriented model is just the opposite while intermediarymarketplaces bring together a high number of buyers and sellers.According to industries we distinguish between vertical marketplacesserving one industry and horizontal models that serve several industries.Among seller-oriented vertical marketplace providers we can find many
E-MARKETPLACES
Trang 9Grid Solution for e-Marketplaces Integrated with Logistics 157third party logistics companies (3PL) that undertake packaging,transportation, warehousing etc However, none of the currently usedmodels enable optimisation that includes logistical costs as well E-marketplaces selling goods either do not offer logistical solutions at all
or offer a single solution or 2-3 possibilities in different time-cost ranges.Electronic distributors undertake the task of mediating between buyersand seller providing logistical services as well but this excludescompetition from the 3PL side
A new approach given by Kacsukné & Cselényi (2004) suggests amodel that really integrates logistics services providers to goods’ e-marketplaces In this model the logistics providers are placing theiroffers step by step during the negotiation of buyers and sellers then eachtime a combined proposal showing the total costs is created by the
marketplace The general framework of the integrated marketplace can
be seen on Figure 1 In the centre there is the marketplace engine towhich the entities from all the three sides – buyer, seller and 3PL – arejoining with the help of front-end processors The ratio of the number ofbuyers and sellers determines if the model is buyer-oriented, seller-oriented or intermediary type
It is assumed that all participants have advanced enterprise resourceplanning (ERP) systems and scheduling programs that can provide thefront-end processors with the relevant data within a short time
To illustrate the computational tasks of marketplaces integrated withlogistics we outline a three-sided auction algorithm involving multipleproducts, multiple buyers and multiple 3PLs in a single transaction:
1 The buyer issues a request for proposal (RFP) identifying therequirements
2 Sellers bid offering products
Figure 1.: Framework for an e-marketplace integrated with logistics
Trang 10We outline the optimisation algorithm of Step 4 Let us suppose thatthe buyer would like to purchase N different product items in amarketplace where M sellers and L logistics providers participate Theoptimal proposal can be chosen by minimising the total costs i.e the sum
of the purchase price, the transportation costs and the warehousing costsfor the period from the time of the actual delivery to the latest delivery asformulated in (1) We suppose that all the required amounts can bepurchased, which is expressed by (2)
unit price of the i product asked by the k seller
unit cost of getting the i product from the k seller via the 1.3PL
quantity of the i product purchased from the k seller
required quantity of the i product
the technical factor of storing of the i product for a timeperiod
ce capital tying up cost factor for a time period
the latest possible delivery time of the i product
the actual delivery time of the i product from k seller by 1.3PL
This is an integer programming problem with and binaryvariables that should be solved in each round of the auction process Thismodel is a grossly simplified illustration of the computational tasksbecause here we disregarded of the possible discounts for buying andtransporting in bulk that makes the problem much more complex
Trang 11Besides the substantial advantages, like reduced intermediation costs,integrated processes in supply chain, shortened purchase cycle, greatertransparency and lower administrative costs, e-marketplaces are stillfacing significant technical difficulties Integrating legacy back-officeapplications and ERP systems with marketplaces is a complex,expensive, but necessary task to utilise fully the opportunities ofexchanges Also, marketplaces offer only limited functionality todaybecause of the difficulties in integrating existing value-added services.These services may also require large computational power like theoptimisation algorithm described in section 2.
The following e-marketplace model based on Grid and Web servicesconcepts offers solutions for these problems If both back-office andmarketplace applications are implemented as Grid services they are able
to communicate with each other by exchanging standard SOAPmessages Interoperability is provided by Web services standards despiteany differences in hardware platforms, operating systems orprogramming languages applied A Grid service based model alsoprovides the possibility to extend the functionality of exchanges.Existing legacy applications run by participants or third party applicationservice providers are offered as Grid services and can easily beintegrated with the marketplace solution In addition, the substantialamount of calculation that may be required can be distributed byintroducing a special Grid implementation called Marketplace SupportGrid (MSG) that uses the computation power offered by participants whochoose this form of contribution instead of paying the registration fee.The general architecture of MSG is illustrated on figure 2 Themarketplace engine coordinates the business transactions communicating
with the front-end processors of the participants The optimisationproblems are passed to the Computation Centre This unit consists of the
Grid Solution for e-Marketplaces Integrated with Logistics 159
E-MARKETPLACES
Figure 2 Marketplace Support Grid Architecture