Topic allocation method on edge servers for latency‐sensitive notification service 2021 Tomoya Tanaka.pdf

该【Topic allocation method on edge servers for latency‐sensitive notification service 2021 Tomoya Tanaka 】是由【可卿】上传分享，文档一共【17】页，该文档可以免费在线阅读，需要了解更多关于【Topic allocation method on edge servers for latency‐sensitive notification service 2021 Tomoya Tanaka 】的内容，可以使用淘豆网的站内搜索功能，选择自己适合的文档，以下文字是截取该文章内的部分文字，如需要获得完整电子版，请下载此文档到您的设备，方便您编辑和打印。:.
Received:10March2021Revised:25April2021Accepted:13May2021
DOI:
SPECIALISSUEPAPER
Topicallocationmethodonedgeserversforlatency-
sensitivenotificationservice
TomoyaTanaka1|TomioKamada2|ChikaraOhta3
1SchoolofInformationTechnologies,
TallinnUniversityofTechnology,Tallinn,Abstract
EstoniaTheimportanceofreal-timenotificationhasbeengrowingforsocialservices
2GraduateSchoolofSystemInformatics,andIntelligentTransportingSystem(ITS).AsanadvancedversionofPub/Sub
KobeUniversity,Kobe,Japansystems,publish-process-subscribesystems,wherepublishedmessagesare
3GraduateSchoolofScience,Technology
andInnovation,KobeUniversity,Kobe,spooledandprocessedonedgeservers,havebeenproposedtoachievedata-
,wepresentasystemthatallowsa
topictobemanagedonmultipleedgeserverssothatmessagesareprocessed
Correspondence
TomoyaTanaka,SchoolofInformationnearthepublishers,-
Technologies,TallinnUniversityofingmessagesongeographicallydistributedserverscouldenableimmediate
Technology,Ehitajatetee5,,theduplicatedmessagespool
Tallinn,Estonia.
Email:******@-
process-subscribesystemandformulatethetopicallocationasanoptimization

TomoyaTanaka,AkadeemiaTee7/2,
12611Tallinn,-hard,weproposeheuristicsleveragingthelocalityandthe
pub/subrelationshipsobservedbetweenclientstousetheedgeserver
Fundinginformation
JSPSKAKENHI,Grant/AwardNumbers:
JP18H03232,JP20K11841;Nationalreducesthedelaytodelivernotificationsandtheeffectivenessofthestrategy
InstituteofInformationandexploitingtherelationshipsbetweenclients.
CommunicationsTechnology(NICT)
1|INTRODUCTION
Inrecentyears,theimportanceofreal-timenotificationcombinedwiththeuseofdatahasbeengrowingforapplica-
tionssuchasIoTapplicationsandIntelligentTransportingSystem(ITS).1Forexample,real-timedecision-makingser-
vicesproposedinITSareexpectedtoreactimmediatelytochangesintrafficconditions,analyzethecurrentconditions,
-accessEdgeComputing(MEC)andPub/Submessagingmodelshave
beenexploitedtogeneratenotificationsimmediately,reflectingthechangingspatio-temporalconditions.
MEC,proposedbytheEuropeanTelecommunicationsStandardsInstitute(ETSI),enablesultra-lowlatencyand

beenappliedtoIoT,AR/VR,,Pub/Submodelcanbeadaptedtothe
designoflarge-,topic-basedandcontent-basedPub/Submodel
,6Amoresophisti-
catedversionofPub/Subcalledpublish-process-subscribeenablesdata-drivennotificationsbyspoolingandanalyzing

ThisisanopenaccessarticleunderthetermsoftheCreativeCommonsAttribution-NonCommercialLicense,whichpermitsuse,distributionandreproductioninany
medium,providedtheoriginalworkisproperlycitedandisnotusedforcommercialpurposes.
©&SonsLtd.
;
/:.
2of17TANAKAETAL.
-basedpublish-process-subscribe
,
publisherscansendmessagestothenearbyedgeserverwhichmanagesthemessage'-
,duplicatingmessageson
geographicallydistributedserverscouldenableimmediatenotificationsfrompublisherstoneighboringsubscribers.
However,duetothelimitedresourcesonedgeservers,duplicatingmessagesonnumerousedgeserverscouldcause
,which
performanalysisofthestreamingmessagesinthepublish-process-subscribeparadigm,-
trast,wemainlyfocusonutilizingstoragecapacityonedgeserversefficientlywhileenablingonetopictobemanaged
onmultipleedgeserverssothatthepublishedmessagesareprocessedatnearbyedgeserversandthegeneratednotifica-

capacityneededforduplicatedmessagespools,itdoesnotincreasethetotalcomputationalcostformessageprocessing
executedinadistributedmanner.
Wehaveproposedtheprototypedesignofourpublish-process-subscribesystemandconductedapreliminaryexper-
imenttoshowthetrade-,wemea-
suredthepathlengthfrompublisherstosubscribersandthestoragecapacityusedoneachedgeserverdependingon

servers.
Inthispaper,weproposeamodelthatlinksthepathfrompublisherstosubscribersandestimatestheconsumed
,weproposeamethodtoallo-
catetopicsonedgeserversinordertoachievereal-timenotificationsinapublish-process-subscribesystemunderlim-
:
•Weproposeadelaymodelofthenotificationdeliveryfromapublishertosubscribersandformulatethetopicalloca-
tionmethodasanoptimizationproblemofthedelayundertheresourceconstraintsonedgeservers.
•AstheformulatedoptimizationproblemisNP-hard,weproposeaheuristicnamedRELOC,whichallocatestopics

andtopic-derivedrelationshipsobservedamongpublishersandsubscriberstousestoragecapacityefficientlywhile
maintainingproximitybetweenclientsandamessageprocessor.

themotivationalcaseinmoredetails,enhancethemodelusedtorepresentcomputationalresourcesonedgeservers,
andimprovetheRELOCalgorithmanditsevaluation.
Theremainderofthisarticleisorganizedasfollows:Section2explainsrelatedworkstoclarifythepositionofour
,
Section5,weformulatethenotificationdeliverydelayforthepresentedpublish-process-,

topicmethodinSection7.
2|RELATEDWORK
Topic-basedPub/(called
publishers)toreceivers(subscribers).Instead,abrokerpreparedforeachtopicreceivesmessagespublishedtothistopic

,scalablePub/Subsystemsareusedtoconnectmulti-
-13ApacheKafka,11whichisdevelopedbyLinkedIn,canallocatemultiplebrokersfor
.
Thissystemfocusesonscalabilityandfaulttoleranceofthedistributedsystembutpayslittleattentiontothegeographi-
callocality.
Publish-process-subscribesystemsareanevolutionofPub/Submechanismswherepublishedmessagesare
-process-subscribemodel-basedreal-timecommunicationis
-time:.


balancesdatapublicationandprocessingloadforpublish-process-subscribesystemsoperatingattheedgesandensures
Quality-of-Service(QoS)onaper-
Nevertheless,manyresearchersfocusingonefficientedgeserverresourceuseassumethatonetopicismanagedby
,suchone-topic-to-one-edge-
example,itcouldtakelongertodelivermessagesfrompublisherstosubscribersinapplicationswherepublishersand
,computationalloadandtrafficloadcouldbeconcentratedwhen

hasthepotentialtoachieveimmediatemessagedeliveryfrompublishertosubscriberviatheirnearbyedgeserver,as
-topic-to-many-edge-serverarchitecturecoulddistributecomputationalloadandtraffic
,preparingmessageprocessorsonmultipleserverscouldcausestoragecapacityover-consumption.
Consideringthelimitedresourcesofedgeservers,manyresearchesfocusonproactivelyfetchingcontentonedge

cachedinclude(1)contentrequestprobability,15,16wherethemostpopularcontentiscachedonedgeservers;or(2)
clientmobility,17,,
inpublish-process-subscribesystems,themessagesordataobjectstobestoredonedgeserversaredeterminedbytopics
andpublishers/,weproposeanefficientuseofthelimitedstoragecapacity
,
topic-allocation-basedmessagesmanagementonedgeserversinpublish-process-subscribeparadigmhasnotbeenpres-
entedbefore.
,20isa
distributeddatastoreforFog/
andintroduceslocation-consciousreplicaplacementstrategyusingacontext(location)--
licasonneighboringhoststoenableefficientquorum-basedqueriesandalsoallocatesremotereplicastoprovidetoler-

publish/-localfederatedstoreoverunreliableMEC
,
federatedindexingmodelusingBloomfiltersanddifferentialreplicationschemeacrossedgeservers,butitdoesnot

,butit
doesnottakeintoaccountresourcecapacityordataprocessing.
3|SYSTEMOVERVIEW
Weconsiderapublish-process-subscribesystem,whereeachedgeservergeneratesandtransmitsdata-drivennotifica-
tionsnearpublisherssothatimmediatenotificationsaredeliveredtothenearbysubscribers.
Figure1presentsanoverviewofoursystem,whereeachpublisherisassignedtoasingleedgeserverwhichman-
,ClientCpublishesmessagestoTopicsPandQ,andEdge
B—theassignedserverofClientC—“home
server.”
andananalyzerofthespooledmessagesforatopic.
Onthistopicmanagementmechanism,thesystemgeneratesnotificationsanddeliversthemtosubscribersas
,itspoolsmessagesinthemessageproces-
,themessageprocessoranalyzesthespooledmessagesandgeneratesanotification.
Finally,