LuMoreau 1
,SimonMiles 1
,Carole Goble 2
, MarkGreenwood 2
,VijayDialani 1
,Matthew
Addis 1
, Nedim Alpdemir 2
, Rih Cawley 2
, David De Roure 1
, Justin Ferris 1
, Rob
Gaizauskas 6
,Kevin Glover 3
,Chris Greenhalgh 3
,Peter Li 5
, Xiaojian Liu 1
,Phillip Lord 2
,
Mihael Luk 1
,Darren Marvin 1
, Tom Oinn 4
,Norman Paton 2
, Stephen Pettifer 2
, Milena
V Radenkovi 3
, Angus Roberts 2
, Alan Robinson 4
, Tom Rodden 3
, Martin Senger 4
, Nik
Sharman 2
,Robert Stevens 2
, BrianWarboys 2
,PaulWatson 5
,ChrisWroe 2
(1) UniversityofSouthampton,(2) UniversityofManhester, (3) Universityof
Nottingham, (4) EMBLOutstation - European Bioinformatis Institute,(5)University
ofNewastle, (6) Universityof SheÆeld
Projetsite: www.mygrid.org.uk
Contat Author: LuMoreau
Email: L.Moreaues.soton.a.uk
Abstrat
MyGrid isan e-Siene Grid projet that aimsto help biologists and
bioinfor-matiians to perform workow-based in silio experiments, and helpto automate
themanagement of suh workows throughpersonalisation, notiationofhange
and publiationofexperiments. In thispaper, we desribethearhitetureof
my-Gridandhowitwillbeusedbythesientist. WethenshowhowmyGridanbenet
fromagentstehnologies. Wehaveidentiedthreekeyusesofagenttehnologiesin
myGrid: useragents,abletoustomizeandpersonalisedata,agentommuniation
languages oering ageneriandportableommuniationmedium,andnegotiation
allowingmultipledistributedentitiesto reah servielevel agreements.
1 Introdution
MyGrid is aGrid middleware projet in a bioinformatis setting. In biologialsienes,
itisnotprinipallythesizeofthedatathat mattersbuttheomplexityinvolvedinusing
it: the omplexity of the data itself, the number of repositories and tools that need to
be involved in the omputations required to answer the kind of questions posed by the
sientist, and the heterogeneity of the data and operation of tools. Rather than a few
international failities(e.g. CERNand Fermi Lab) produing vastamounts ofdata that
needstobeaessible, thepressingissuewith biologyisopingwithaverylargenumber
of sites (potentially thousands of individual laboratories) around the world, eah using
heap, ommoditytehnologytoontinuously generatesubstantialquantitiesof dierent
only raw data, but also additional annotations supplied by a small number of human
experts (urators)orautomatedsystems. Annotations aretypiallysemi-struturedtext
that make some use of keywords and ontrolled voabularies, and have to be parsed
omputationally or read by people. Therefore, as well as a large number of data types,
muh of the valuable knowledge is loked into semi-strutured text, under the premise
that the sientistwillread and interpretit.
In the past, this omplexity has been dealt with largely by the intelligene of the
pratising biologist. This has been possible beause biologists working on a spei
organism, ora spei aspet of it, have needed aess to only a smallnumberof these
resoures.
Interestingly, muh of the growth of moleular biology has been ontemporaneous
withthedevelopmentoftheWeb,whihprobablyexplainswhymanyresoureshavebeen
designed withthe intentionthat asientistwillinteratwithaWeb page,dealingwitha
singlequery atatime, andread the resultsdisplayed asreports inabrowser, navigating
between links in dierent databases by mouse-liking. (e.g. http://www.expasy.org
for SWISS-PROT). We all this approah \query by navigation". Where databases are
published,theyareusuallyreleasedasatles,eveninthoseases,suhasSWISS-PROT
and EMBL, where the prodution systems are relationaldatabases.
Although the volume of data isnot yet aomputationalproblem,the advent of high
throughputexperimenttehniquesmeansthathumananalysisisnowreahing its
limita-tions. Withsequene databasesreahinghundredsof MBytesand miroarrayexpression
data produing tens of GBytes, the limits of the non salable query by navigation are
rapidly being reahed, if not already passed.
The partiular fous of myGrid,therefore, is on inreasinglydata-intensive
bioinfor-matis and the provisionof a distributedenvironmentthat supports the in silio
exper-imental proess. The vision is of a \lab book" environment where the e-Sientist an
onstrut in silio experiments, and nd and adapt others, store partial results in loal
data repositoriesand havetheir own viewonpublirepositories,and bebetterinformed
as to the provenane and the urreny of the tools and data diretly relevant to their
experimental spae. For a less skilled user, myGrid should help in nding appropriate
resoures, oering alternativesto busy resoures and guiding the user through the
om-position of resoures intoomplex workows. In order toprovide suh an environment,
myGrid unequivoally needs to address the \Grid problem", i.e. the exible, seure,
oordinated resoure sharing, among dynami olletionof individuals and institutions,
| Virtual Organisations [8℄ | in this ontext, the Grid beomes egoentrially based
around the Sientist: myGrid.
The ontributionsof thispaperare threefold. First,we presentaservie-based
arhi-teture to support the vision of the \lab book" environment. Seond, we illustratehow
this arhiteture an be used during the enatment of workows. Third, we review how
Inthissetion,wedesribethedierentserviesthatareprovidedbymyGrid,andsketh
theirinterations. (Theyaredisplayed inFigure1.) Theexperimentalin silio proessis
expressed asaworkowsript bythe sientist. Serviesan beviewed asbeing provided
by agents and workow an be seen asan agent interation sript. Some initialwork in
this veinhas already been done [2,3℄.
2.1 Workow Enatment
At the heart of the myGrid runtime system, we nd the workow enatment engine
whih, given a workow sript, is able to exeute (or enat) the sript. Sientists and
their institutions may have preferenes that must be taken into aount when enating
a workow sript: e.g., some databases are preferred over others, or spei tools and
parameters are routinely hosen. It is the role of the workow resolution servie to
ustomise a sript's \free variables", possibly making use of a workow personalisation
servie able to obtain preferenes from a user (or a user agent ating on their behalf).
There exist several strategies to resolve a workow: eagerly before enatment, orlazily
if and when required by the enatment engine. (Both an be expressed at the level of
sripts through the use of anappropriate programtransformation.)
Theworkowenatmentansendrequeststoexistingrunningserviesoranativate
tools and interat with them: servies need to be disovered and proesses need to be
reated. For the former, a servie diretory is used as a repository of servie instanes
thatare urrentlyative,whereas thelattermakesuseofajobativationandsheduling
system. Generally, sripts may require spae to store temporary results, or may like to
ensure that omputational resoures are reserved at the same time as storage spae to
ensure theprompt exeutionof theworkow: alloationand reservation willbe handled
by the resoure management servie.
2.2 User Interation
The user, through an interfae, may interat with the workow enatment engine,
sus-pending and resuming workows, observing their progress, analysing their logs.
Sus-pended workows will be serialised and stored in a repository, potentially shared with
other users.
Some workows may take days, if not weeks, to omplete their exeution. Users
therefore need to be notied when workow exeution terminates. We prefer not to
as-sumetheexisteneofuseragentsabletohandleinomingnotiations. Indeed, usersare
not logged on permanently, and we feel that always runninguser agents would overload
the system unneessarily. Instead, we makeuse of a notiation servie able to forward
messages touser agents,when present, ortostore messages intheir absene. The use of
the notiation servie is of ourse not restrited tothe user agent,but may beused by
any servies in myGrid.
Sharinginformationbetweenusers,disoveringinformation,ndingoutusersor
Ontological Definitions
Ontological Reasoning
Workflow Provenance
Validation
Workflow Definition
Repository
Service Functionality
Metadata
Provenance
Repository
Serialised Workflow
Repository
Workflow
Resolution
Databases
Workflow
Personalisation
User
Agent
User Repository
Workflow
Enactment
Service
Directory
Notification
Job Scheduling
Resource Management
Distributed
Queries
Information
Extraction
Authentication
Authorisation
Groups, Roles Directory
User Directory
Figure 1: MyGrid Servies
diretories are used for that purpose: the user diretory holds information about users,
groups,rolesandinstitutions;theworkowrepositoryontainsinformationaboutsripts
and their funtionality.
2.3 Ontology Servie
All information about workows and users is what we all metadata and is strutured
aording to a set of ontologies | an ontology is generally dened as a shared
under-standing of a spei domain [10℄. Information about servies are also expressed using
suh ontologies, and are stored in the servie funtionality metadata servie; the latter
servie ontains metadata about lasses of servies, and must bedistinguished from the
servie diretory whih listsative servie instanes.
Not only are ontologies a shared understanding of some domains, but their logial
foundations alsoallowusers to perform reasoning over suh domains. Examples of
rea-soning inlude lassiation (i.e., the omputation of a onept hierarhy based on the
speialisation relation), or onsisteny heking (i.e., heking that a statement is not
inonsistent in a logi). An ontology-based reasoning faility is provided by myGrid to
relationships.
2.4 Data and Metadata
Most myGrid repositories will be implemented as databases. Additionally, biologial
informationisstoredinmultipleandheterogeneousdatabases. Distributedquerysystems
over suh databases are anessential omponent to failitate informationintegration. In
myGrid, databases will be aessed though a servie interfae [15℄, whereby strutured
datastoressupportonsistentinterfaesfordatabase aess,manipulationandmetadata
desription. As aomponent withinthe personalisationframeworkof myGrid, database
servies will be used to provide individual users with aess to (i) loally produed
data sets; (ii) the results of analyses run by the user over loal or remote data; and
(iii) distributed querying over loal and remote data resoures. The distributed query
proessor will benet from the onsistent servie interfaes and metadata desriptions
provided by loaland remote databases.
Above, we have disussed the existene of metadata that is strutured aording to
ontologies. In biologialsienes, it is alsoustomary to reate annotations in free text
form. Suh metadata ontains invaluable information assembled by database urators.
MyGridalsoprovidessupportfororrelatingsuhaninformationwithmedialliterature
through an informationextrationservie.
MyGrid providessupport forprovenane in two dierent ways. First,provenane
in-formation,inpartiularrelatedtoworkowenatement,anbeloggedinthe provenane
annotation servie; suh a servie is also used to store provenane information for
ser-vies having no built-in support for provenane. Additionally, the workow provenane
validation servie isable to re-enat workows toestablishhange overtime.
2.5 Seurity and Fault Tolerane
The myGridauthentiationservie extendsthe PKIinfrastruture toprovideX.509
er-tiates for users and objets (alled identities heneforth) needing veriation. It
sup-ports anotionoflogialdomain whihisdened by theset of identities itmanages. The
onfederation ofseveral logialdomains formsanenterprise infrastruture. Eahlogial
domain has assoiated domain administrators who are authorised to reate and revoke
identities withintheir logial domains.
In myGrid, a sub-omponent of the user agent ats as a redentials repository,
per-mitting simultaneous aess to multiple logial domains. This faility allows a user to
havesimultaneousaesstomultiplevirtualorganisations[8℄andobtaintheaessrights
tomultipleresoures arosssites.
MyGrid supports role-based aess ontrol [16℄ and dynami mappingbetween users
and roles. Within eah logial domain, there exists a hierarhy of user roles and aess
rights;rolesare statiallyassoiatedwithaessrights. Themodelisextensibleby
allow-ingthe denition ofnew roles and aessrights. In anenterpriseseurity infrastruture,
domainonto roles and aess rightsof another domain.
Mygridomputationsmaybelong-livedandinvolveaverylargenumberofomputing
resoures. Hene,they needtobedesignedwithfaulttoleraneinordertoberobust. To
thisend,myGridwillprovideasetofinterfaes,whihserviesarerequiredtoimplement,
and whih willprovide robustness to appliationsinvolving the use of multiple servies.
The omplete desription is beyond the sope of this paper, and we refer the reader to
a ompanion paper [4℄. The approah may be summarised as follows: implementors of
a servie have to implement an interfae (for hekpoint and rollbak); the arhiteture
dynamially extends the servie interfae by methods for fault tolerane; appliations
makinguseofdierentservieshavetodelaretheirinter-dependenies,whihareusedby
afault-managertoontrolhekpointsand rollbaks;anextensionoftheommuniation
layeris able to logand replay messages.
3 MyGrid Workow Enatment in Pratie
Wehave implementeda prototypeof this arhiteture, based ona subset of the servies
desribed inFigure1and exlusivelyrelyingonWebServiestehnology. Inthissetion,
we showhow the sientistis able toenat workows inmyGrid.
An in silio experimenttypiallyinvolvesusing several bioinformatis databases and
algorithmsavailable on the World Wide Web. Currently, these resoures are integrated
by a\query by navigation"proess, i.e. by uttingand pastingaross browser windows.
Alternatively, a sript (suh as perl sript or bat le) may be written to failitate the
frequentrepeatofin silioexperiments. Thereareanumberoflimitationsofthisurrent
state of pratie that workows inthe myGridenvironment address.
First, there is the problem of knowing what in silio experiment to perform. A
user typially has anunderstanding of what they are trying to ahieve inbioinformatis
terms and might know some spei Web resoures or sript, based on past experiene.
How they aquiredthis experiene, howthey keep their knowledge up-to-date, and how
they adaptprevious experienes to new tasksare essential elements of the experimental
proess, whihwe intend to make expliit.
Seond, there is the problem of inorporating new resoures. In most situations
the user is interested in a spei type of resoure, a SWISS-PROT database, rather
thana speiresoure instane suhasthe SWISS-PROT databasehosted ataspei
institution. If their rst (default) hoie is unavailable, then the user would like to use
analternativeof the same type. In the urrent state of pratie, sripts tend toinlude
hard-oded referenes tospei resoures.
Third,thereisthelimitedreordingofhowinsilioexperimentshavebeenperformed.
Without knowing what resoures have been used in the derivation of a result, there is
no way of knowing if it might be worthwhile re-running the in silio experiment in the
light of more reent knowledge (or if the result should be disregarded, as more reent
knowledge has rendered some of the experimental assumptionsinvalid.)
Fourth, there is diÆulty in propagating good in silio experimental pratie. This
indi-e-Siene ommunity, it isnot just the available data that isvaluable, but also knowing
the aeptable/proven ways of ombining that data togenerate new insights.
3.1 Prototype Experiment
In our prototype, a myGrid user has aess to a personal repository ontaining their
domaindata(andresults),aworkowrepositoryontainingtheavailableworkowsripts
and aservie diretoryofthe availableservie instanes. Eahdata iteminthe personal
repository has an assoiated onept type (a term in the ontology); suh onept types
are used to initiatethe enatment of in silio experiments, aswe now explain.
Potential workows are identied through a onversation with the ontology servie.
A spei user interfae is used to inrementally build up an abstrat desription of a
workow,startingwiththeseletedonepttype. Onetheabstratworkowdesription
is omplete, it an be lassiedto givea workow servie type identier (also aterm in
the ontology). This isused toretrievethe identiers of workowsripts that math this
required type, and from the identier, the workow sript itself. In this way, the user
interatswith theontologyservietodeterminethe onept thatmaththeirtask;then,
they get alistofallthe workowsripts ofthis typeand hoose the onetorun (perhaps
using some metadata tohelp inthe seletion).
3.2 Workow Details
Inspired by WSFL [11℄, the workow denition onsists of a set of servie providers,
ativities, data links and ontrol links between ativities. For many myGridworkows,
eah ativity has its own servie provider, whih inludes a loator element to identify
the Web Servie, to be used by the workow enatment engine. It is possible for the
loator to be stati and diretly referene the WSDL denition of the servie, but it is
moreusual forthe loator tobedynami. In this ase,itgivesthe servietypeidentier
that is used to lookup possible servies (using UDDI) from the servie diretory. Eah
ativity is desribed in terms of its servie provider and an operation, thus expressing
the spei provided operation that mathes the abstrat ativity in the workow. The
data links desribe how the outputs of an ativity are mapped to the inputs of other
ativities, whilethe ontrol linksare used todeide when the ativities shouldbered.
The enatmentofaworkowsriptstartsby sendingthe sriptandinputdatatothe
workow enatment servie. This responds by returning a workow instane identier
that the user interfae portal an use to query the workow status and identify the
workow result inthe personal repository.
Theuseofadynamiloatortoidentifyaservieproviderintheworkowsriptisthe
mainmehanismfor abstratingaworkowoverspei servieinstanes. Thedynami
loator gives the servie type identier; any servie instane that has registered under
this identier in the servie diretory is a potential math. The dynami loator also
givesthe poliytobeused forseletingbetween thepotentialservies. In theprototype,
hoie, where the listof servies is sent tothe user agent who makes hoie onbehalf of
the user, possibly interating with the user through the portal, if ongured todo so.
The workow enatment servie also reates a provenane log within the personal
repositoryfor eahworkowinstane. This traeinludes: the initialdata,the workow
sript, the intermediate results, the atual servie instanes seleted and the time taken
for the servie operations. These logsouldbeviewed through the portal tounderstand
the detailed derivation of a partiular result.
The denition of an in silio experiment as a workow means that it exists as an
expliitpieeofdatathatanbeshared,opiedandalteredbyaommunityofsientists.
Evenwithintheontextofthesimpleexamplesintheprototype,itwaslearthatwhata
usermightonsiderasinglein silioexperimentmightbesupportedbymanyworkows.
Thereare variantsof workows that havethe same type andthe hoiebetween themis
oftenthe personal hoie of the user. Some users willalways want tobe involved in the
dynamiseletionbetween alternativeservies, whileotherswillbeontenttoleavethat
totheenatmentengine,oranagentatingontheirbehalf. Anotherway thatworkows
of the same type might vary is in the ltering of sets of intermediate results. (In the
urrent state of pratie, this orresponds to a user who applies their knowledge to ut
and paste seleted data between resoures inan in silio experiment.)
While our projet is still at an early stage, we were able to enat workows that
expressed ratheromplex queriesinbioinformatis,suhas (i) Hasanyone elsestudied
the eet of neurotransmitters on the iradian rhythms of Drosophila? (ii) How do
the funtionsofthe lustersofproteinsfrommy experimentinterrelate? (iii)What are
the proteins with apartiular funtion? (iv) What is known about a given protein?
The enatement of workows has shown that there is a need for user preferenes to
guidetheseletionofserviestoinvoke. Thereissopeforuseragentsto(semi-)automate
the ustomisation of servie seletion, and also for negotiation when multiple servie
with omplementary harateristis are available to the user. This is preisely the role
of software agents, whihwe disuss inthe following setion.
4 Agents in Bioinformatis Grids
The bioinformatis domain is haraterised by rapid and substantial hange over time.
The volume of data poses problems, but the hange in the resoures available to the
biosientistisadistint problem;new resoures an appear, oldones andisappear, and
some an simply hange. Although there are several well-known and highly regarded
databases, limiting a system to only these ould impose undesirable onstraints. Thus,
any system intended for appliation to the bioinformatis domain should be able to
ope with this dynamism and openness, and nothing addresses these onerns in quite
the wayas theagentapproah. Agentsareexible, autonomousomponentsdesignedto
undertakeoverarhingstrategigoals,whileatthesametimebeingabletorespondtothe
unertaintyinherentintheenvironment. Ontheonehand,agentsprovideanappropriate
used for partiular purposes in ertain aspets of the system, inludingpersonalisation,
ommuniation,negotiation, whih we disuss below.
4.1 User Agent
The user agent of Figure 1 is an agent in the sense that it represents a user within the
myGridsystem(soouldalsobedesribedasapersonalagent [12℄). Itanautonomously
provide the personal preferenes and onditions of a user to other parts of the system.
This is useful, in partiular, when a workow is being enated and a hoie of servies
beomes available. The hoie should not be made arbitrarily, but on the priorities and
irumstanes of the partiular user. Forexample, a user may have greater trust in the
ability of one servie to produe aurate results than another, or the user's operating
system may only support some forms of interation between servies and the user. The
usershouldnothavetobequeriedeahtimeaserviemustbehosen,asthesepreferenes
and previous hoies an be reorded and ated upon by the user agent to selet from
eahset of optionspresented to it. We allthis funtionpersonalisation.
Another appliation of the user agent is as a ontat point between servies within
myGrid and the user. By having an intermediaryable toreeive, for example, requests
fromserviesfortheusertoenterdataornotiationsabouthangestoremotedatabases,
thesemessagesanbeprovidedtotheuseronlywhentheuserisableandwillingtoreeive
them. Conversely, theuseran delegatethedetailsofaproeduretothe useragent,suh
as authentiating itself with a servie before use, or for personalisation of workows as
desribed above.
4.2 Agent Communiation Language
AkeyrequirementofmyGridisthe designofafuture proof environmentinwhih
ollab-orative distributed bioinformatis appliations may be developed. Bioinformatis is not
a green eld, and multiple protools and standards are already supported by the
om-munity. Our methodology is to design a generi arhiteture able to support multiple
existing protools, languages and standards, and whih hopefully willbe able to
aom-modatefuturedevelopments. Inpartiular,wewanttodesignanabstratommuniation
arhiteture that we an map onto onrete ommuniation tehnologies.
At the same time, in the eBusiness ommunity, Web Servies have emerged as a
set of open standards, dened by the World Wide Web onsortium, and ubiquitously
supported by IT suppliers and users. They rely on the syntati framework XML, the
transportlayerSOAP[20℄,theXML-basedlanguageWSDL[19℄todesribeservies,and
the servie diretory UDDI [18℄. Web Servies therefore look likea strongontender for
GridComputing,asillustratedbythereentOpen GridServieArhiteture(OGSA)[7℄
whihextends Web Servies with supportfor the dynamilifeyle managementof Grid
Servies.
Theidea ofan\agentommuniationlanguage"datesbak fromtheDARPA
Knowl-edgeSharingEort,whihledtothedesignof KQML(KnowledgeQueryand
In agent systems, it is ommon pratie to separate intention from ontent in
om-muniative ats, abstrating and lassifying the former aording to Searle'sspeeh at
theory [17℄. An agent's ommuniations are thereby strutured and lassied aording
toa predened set of \message ategorisations",usually referred toas performatives.
In previous work, we have suessfully adapted a key onept of the Nexus
ommu-niation layer [9℄ to the world of agents, whih resulted in SoFAR, the Southampton
Framework for Agent Researh [14℄. Communiations between agents take plae over
a virtual ommuniation link, identied by a startpoint and an endpoint. An endpoint
identies anagent's abilitytoreeive messages using aspei ommuniation protool.
An endpoint extrats messages from the ommuniation link and passes them onto the
agent. Astartpoint istheother endof theommuniationlink,fromwhihmessages get
sent to an endpoint. Given a startpoint, one an ommuniate with a remote agent, by
ativating aperformativeon the startpoint,passing the messageontent.
In [13, 1℄, we have desribed how the idea of agent ommuniation languages, and
thestartpoint/endpointommuniationmodelouldbemappedontotheommuniation
stak of Web Servies. In [13℄, we onlyfousedonthe ommuniation layerby enoding
performativesandmessageontentsinSOAP.In[1℄,wemadeuseofthe WSDLlanguage
todesribe agentsand the performatives they support, sothat suhdenitions ould be
published in the UDDI registry,disovered and re-used like any other Web Servie.
This approah turns out to be promising, as it oers a delarative ommuniation
semantis, whihpromotes inter-operability,openness, and dynamidisovery and reuse
of agents. It alsoopens the agent world to the Web Servies ommunity, helping in the
design of more omplexinterations, asdisussed in the followingsetion.
4.3 Negotiation Broker
Anotherappliationofresearhfromtheagenteldisintheareaofnegotiation. Servies
andthe usersand servieproviders theyinteratwithwillhavedieringriteriaoverthe
preferable qualityand ontent of the servie they reeive.
An area in whih negotiation an be seen as partiularly useful in myGrid is
noti-ation support. The providers of various servies may want to send out into the wider
system notiations onerning improvements totools,hanges todatabases or updates
onerningthe stateof enatedworkows, et. Otherservies oragentswillwantto
reg-istertoreeivesomesubsetofthesenotiations. Forstability,wesupportasynhronous
messages, and manage their distribution usinga notiation servie.
4.3.1 Quality of Servie
The subjets (quantitative and qualitative) over whih negotiation takes plae ould
inludethe following formsof quality of servie.
The ost of reeiving the notiation,
daily, hourly,
the generality of the hangedesribed by the notiations,
the form inwhihthe informationinthe notiation message issupplied,
the auray of informationontainedwithin a notiation.
Qualityofservie refers tothese distintionsinboth whatapublisher produesandhow
it produesit.
A publisher of notiations will be able to produe notiations mathing (or
ex-eeding, where appropriate) one or more measures of quality of servie. For example,
a publisher may be able to publish notiations on a partiular topi every minute or
every hour. A onsumer of notiations may prefer, or demand, one measure of quality
of servie overanother. Whether, orhowwell, theirdemands an bemet by a publisher
depends onthe quality of servie that the publisher an provide.
If demands annot be met exatly, the onsumer may hoose to negotiate with the
publisher to nd the next best quality of servie that the publisher an provide. For
example,ifthe onsumerdesiresnotiationsweeklyand thepublisheran providedaily
or fortnightly notiations, the subsriber must nd this out from the publisher and
then deide between them, or deide not to subsribe at all, based on the subsribers
partiular priorities. Alternatively, the publisher may be able to exeed the quality of
servie inseveral ways whihthe subsribermay beunaware of,whihouldalsoleadto
negotiation.
4.3.2 Model
As the notiation servie must provide notiation supportfor a potentiallylarge and
varyingnumberofonsumers,itshouldnothangeitsontratbasedsolelyontheresults
of negotiation between a single onsumer and a publisher. Therefore, the notiation
servie should have some ontrol over the quality of servie agreed upon. There are
otherreasons thatthe notiationservie mayusefullylimitthe interationbetween the
publisher and onsumer, suh as limiting the knowledge of one by the other for reasons
of privay.
We propose using a quality of servie broker that is anagentoneptually ontained
within the notiation servie (available through the same ommuniation hannels).
Thequalityofservie brokerwillnegotiateonbehalfofeahonsumerwishingtoreeive
notiations ofa speied quality, then providea nal proposalto the onsumer. It an
negotiate with any of the publishers known to the notiation servie, and also limit
the agreed quality of servie to that aeptable to the notiation servie. We wish to
make thequalityof serviebroker abletonegotiate with publishers produed by various
providers, soweuse theonept of pluggablenegotiation algorithms,allowingthe quality
In this paper, we have presented the myGrid arhiteture and overviewed possible use
of agents. MyGrid aims to provide a personalised environment for the biosientists,
whih helps them to automate, repeat and therefore better ahieve their experiments.
AgentsarepartiularlyusefulintailoringthemyGridsystemtotheprioritiesofindividual
sientists,personalisingeahstepofaworkowandnegotiatingontheirbehalf. Itanbe
seen fromour disussion that, alongwith dynami workow enatment, standardisation
ofdatasemantisviaontologiesandthemanyotherfailitiesofmyGrid,agentsanmake
onduting in-silio experiments exible and more easily ontrolled by the individualor
ollaboratingsientists.
The examples of use of ageny wehave presented, while already oeringa apability
inexistent in urrent bioinformatis environment, still remain rather loalised to some
spei servies (useragentornegotiation overquality ofservie of notiationservie),
oromponents suh asa ommuniation layer.
For the long term, agent-based omputing also ounts in its armoury a range of
tehniques for enabling individual omponents toollaborate with others, as well as for
ompetingwithothersintheprovisionofserviesasmaybefound inbioinformatis. For
example,the formeraspets inludeissues intheonstrutionofthe virtualorganisation
mentionedearlier,whereby dierentservies ome together insome oherentwhole
sub-system for apartiular purpose; and issues in the regulationof open soietiesof servies
through the use of norms and eletroni institutions. The latter aspets, for example,
inludethepossibleuseofsophistiatedautionmehanisms,oreletroni marketplaes,
for obtainingthe best servies orresoures atthe least ost tothe user.
6 Aknowledgements
This researh is funded by EPSRC myGridprojet (referene GR/R67743/01).
Referenes
[1℄ Arturo Avila-Rosas, Lu Moreau, Vijay Dialani, Simon Miles, and Xiaojian Liu.
Agents forthe Grid: A Comparisonwith Web Servies(part II: ServieDisovery).
In Workshop on Challenges in Open Agent Systems,Bologna, Italy, July 2002.
[2℄ K. Bryson, M. Luk, M. Joy, and D. Jones. Agent interation for bioinformatis
data management. Applied Artiial Intelligene, 15(10):917{947, 2001.
[3℄ K.Deker,X.Zheng,andC.Shmidt.AMulti-AgentSystemforAutomatedGeneti
Annotation. In The fth ACM International Conferene on Autonomous Agents,
Montreal, Canada, May 2001.
[4℄ VijayDialani,SimonMiles,LuMoreau,DavidDeRoure,andMihaelLuk. T
born, Germany, August 2002. Springer-Verlag.
[5℄ T. Finin, Y. Labrou, and J. Mayeld. Software Agents, J. Bradshaw, Ed., hapter
KQML asan AgentCommuniation Language. MIT Press, 1997.
[6℄ FIPA: Foundation for Intelligent Physial Agents.
http://drogo.selt.stet.it/ fip a/.
[7℄ IanFoster, Carl Kesselman,Jerey M.Nik,and Steven Tueke. The Physiologyof
theGrid|AnOpenGridServiesArhitetureforDistributedSystemsIntegration.
Tehnial report, Argonne National Laboratory, 2002.
[8℄ IanFoster,Carl Kesselman,and Steve Tueke. The Anatomyof the Grid.Enabling
Salable Virtual Organizations. International Journal of Superomputer
Applia-tions, 2001.
[9℄ IanFoster,CarlKesselman,andStevenTueke. TheNexusApproahtoIntegrating
MultithreadingandCommuniation.JournalofParallelandDistributedComputing,
37:70{82, 1996.
[10℄ ThomasR.Gruber.Towardpriniplesforthedesignofontologiesusedforknowledge
sharing. Tehnial Report KSL-93-04, Knowledge Systems Laboratory, Stanford
University, August 1993.
[11℄ FrankLeyman. WebServies FlowLanguage (WSFL). Tehnial report,IBM,May
2001.
[12℄ PattieMaes. AgentsthatRedueWorkandInformationOverload. Communiations
of the ACM, 37(7):31{40,July 1994.
[13℄ Lu Moreau. Agents for the Grid: A Comparison for Web Servies (Part 1: the
transportlayer). In Henri E. Bal, Klaus-Peter Lohr, and Alexander Reinefeld,
ed-itors, Seond IEEE/ACM International Symposium on Cluster Computing and the
Grid (CCGRID2002),pages220{228,Berlin,Germany,May2002.IEEEComputer
Soiety.
[14℄ LuMoreau,NikGibbins,DavidDeRoure,SamhaaEl-Beltagy,WendyHall,Gareth
Hughes, Dan Joye, Sanghee Kim, Danius Mihaelides, Dave Millard, Sigi Reih,
Robert Tansley, and Mark Weal. SoFAR with DIM Agents: An Agent Framework
for Distributed Information Management. In The Fifth International Conferene
andExhibition onThe PratialAppliationof Intelligent Agentsand Multi-Agents,
pages 369{388,Manhester, UK,April 2000.
[15℄ N.W. Paton, M.P. Atkinson, V. Dialani, D. Pearson, T. Storey, and P. Watson.
Database aess and integration servies onthe grid. In Fourth Global Grid Forum
roles. In ACM Workshop on Role-Based Aess Control, pages 47{54, 1998.
[17℄ John Searle. Speeh Ats: An Essay in the Philosophy of Language. Cambridge
University Press, 1969.
[18℄ Universal Desription, Disovery and Integration of Business of the Web.
www.uddi.org, 2001.
[19℄ Web Servies DesriptionLanguage (WSDL). http://www.w3.org/TR/wsdl , 2001.
