Seite - 185 - in Algorithms for Scheduling Problems
Bild der Seite - 185 -
Text der Seite - 185 -
Algorithms 2018,11, 76
AlthoughHEFT iswell-known, the studyofdifferentpossibilities for computing rankvalues
in a heterogeneous environment is limited. In some cases, the use of themean computation and
communicationas therankvalue in thegraphmaynotproduceagoodschedule [26].
In thispaper,weconsider twomethodsofcalculatingtherank: bestandworst. Thebestversion
assumesthat tasksareallocatedto thesameDSP.Hence,nodata transmission isneeded.Alternatively,
theworstversionassumesthattasksareallocatedtotheDSPfromdifferentnodes,sodatatransmission
ismaximal. Todetermine thecriticalpath,weneed toknowtheexecution timeofeach taskof the
graphandthedata transfer time,consideringeverycombinationofDSPs,where the twogiventasks
maybeexecutedtaking intoaccount thedata transfer ratebetweenthe twoconnectednodes.
Tasks labelingprioritizesworkflowtasks. Labelsarenotchangednorrecomputedoncompletion
ofpredecessor tasks. Thisalsodistinguishesourmodel frompreviousresearch(see, for instance [22]).
Tasklabelsareusedtoidentifypropertiesofagivenworkflow.Wedistinguishfourlabelingapproaches:
BestDownwardRank(BDR),WorstDownwardRank(WDR),BestUpwardRank(BUR),andWorst
UpwardRank(WUR).
BDRestimates the lengthof thepathfromconsideredtasktoarootpassingasetof immediate
predecessors in aworkflowwithout communication costs. WDRestimates the lengthof thepath
fromconsidered task to a rootpassing a set of immediatepredecessors in aworkflowwithworst
communications. The descending order of BDR and WDR supports scheduling tasks by the
depth-firstapproach.
BURestimates the lengthof thepath fromtheconsidered task toa terminal start taskpassing
asetof the immediatesuccessors inaworkflowwithoutcommunicationcosts.
Table3.Taskallocationstrategies.
Description
Rand Allocates taskTk toaDSPwith
thenumberrandomlygeneratedfromauniformdistribution
in therange[1,m]
BC(bestcore) Allocates taskTk to theDSPthatcanstart the
taskasearlyaspossibleconsideringcommunicationdelayofall
inputdata
PESS(pessimistic) Allocates taskTk fromtheordered list to
theDSPaccordingtoWorstDownwardRank(WDR).
OPTI (optimistic) Allocates taskTk fromtheordered list to
theDSPaccordingtoBestDownwardRank(BDR).
PHEFT(pessimisticHEFT) Allocates taskTk fromtheordered list to
theDSPaccordingtoWorstUpwardRank(WUR)
OHEFT(optimisticHEFT) Allocates taskTk fromtheordered list to
theDSPaccordingtoBestUpwardRank(BUR)
WURestimates the lengthof thepathfromtheconsideredtasktoaterminal taskpassingaset
of immediatesuccessors inaworkflowwiththeworstcommunicationcosts. Thedescendingorder
ofBURandWURsupports scheduling taskson thecriticalpathfirst. Theupwardrankrepresents
theexpecteddistanceofanytaskto theendof thecomputation. Thedownwardrankrepresents the
expecteddistanceofanytaskfromthestartof thecomputation.
5. ExperimentalSetup
Thissectionpresents theexperimental setup, includingworkloadandscenarios,anddescribes
themethodologyusedfor theanalysis.
185
zurück zum
Buch Algorithms for Scheduling Problems"
Algorithms for Scheduling Problems
- Titel
- Algorithms for Scheduling Problems
- Autoren
- Frank Werner
- Larysa Burtseva
- Yuri Sotskov
- Herausgeber
- MDPI
- Ort
- Basel
- Datum
- 2018
- Sprache
- englisch
- Lizenz
- CC BY 4.0
- ISBN
- 978-3-03897-120-7
- Abmessungen
- 17.0 x 24.4 cm
- Seiten
- 212
- Schlagwörter
- Scheduling Problems in Logistics, Transport, Timetabling, Sports, Healthcare, Engineering, Energy Management
- Kategorien
- Informatik
- Technik