archer user survey · 2016-07-05 · status final author(s): anne whiting reviewer(s) alan simpson,...
TRANSCRIPT
![Page 1: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/1.jpg)
ARCHERUserSurvey
2015
![Page 2: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/2.jpg)
2
1. DocumentInformationandVersionHistoryVersion: 1.0Status Final
Author(s): AnneWhiting
Reviewer(s) AlanSimpson,AndyTurner
Version Date Comments,Changes,Status Authors,contributors,
reviewers0.1 2016-03-07 Initialdraft AnneWhiting0.2 2016-03-22 Furtherupdates AnneWhiting0.3 2016-04-13 Updates AnneWhiting0.4 2016-04-18 Reviewed AndyTurner0.5 2016-04-19 Post-reviewupdates AnneWhiting0.6 2016-04-19 Reviewed AlanSimpson1.0 1016-04-19 VersionforEPSRC AlanSimpson,AnneWhiting
![Page 3: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/3.jpg)
3
2. DescriptionoftheSurveyTheARCHERUserSurveyclosedon17thFebruary2015.230responseswerereceivedfromARCHERusers.Thesurveyaskedforratings(onascaleof1to5)withthefollowingquestions:
1. PleaserateyouroverallexperienceoftheARCHERService(required)[VeryUnsatisfied(1)–VerySatisfied(5)]
2. HastheARCHERhardwareconfigurationmettherequirementsofyourresearch?(required)[Notmetanyrequirements(1)–Exceededrequirements(5)]
3. HasthesoftwareonARCHERmettherequirementsofyourresearch?(required)[Notmetanyrequirements(1)–Exceededrequirements(5)]
4. IfyouhaveusedtheARCHERhelpdesk,pleaserateyourexperience[VeryUnsatisfied(1)–VerySatisfied(5)]
5. IfyouhaveusedtheARCHERdocumentation,diditprovidetheinformationyourequired?[DidnotprovidetheinformationIrequired(1)–ProvidedalltheinformationIrequiredandmore(5)]
6. IfyouhaveusedtheARCHERwebsite,pleaseratethequalityofthecontentandeaseofnavigation[Verypoor(1)–Excellent(5)]
7. PleaserateyourexperienceofanyARCHERTrainingyouhaveused(eitheronlineorface-to-face)?[VeryUnsatisfied(1)–VerySatisfied(5)]
8. IfyouhaveattendedanyARCHERwebinarsorvirtualtutorials,didyoufindthesessionworthwhile?[Acompletewasteoftime(1)–Extremelyinterestinganduseful(5)]
9. IfyouhaveattendedanyARCHERonlinetrainingmaterial(e.g.OnlineDrivingTest,screencasts),howusefuldidyoufindthematerial?[Ofnouse(1)–Extremelyuseful(5)]
Onlythefirstthreequestionswerecompulsoryforallsurveyresponders,butover93%ofrespondersalsoprovidedfeedbackinatleastsomeoftheoptionalquestions.Userswerealsoprovidedwiththeopportunitytooffercommentsorsuggestionsunderalloftheaboveheadingsandprovidedwithspaceforanyothercommentsorsuggestionsattheendofthesurvey.ThesurveywasconstructedusingGoogleFormsandembeddeddirectlyintotheARCHERwebsite.
![Page 4: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/4.jpg)
4
3. SelectedQuotesThefollowingquotesreflectthetoneofthemajorityofresponderstothesurveywithregardtotheARCHERservice:
“IfindtheARCHERservicetoberesponsiveandhelpful,easytouse,andflexible.Excellentservice.”“TheonlinedrivingtestwashowIobtainedaccesstoARCHER,andIthinkitisabrilliantideaforwideningaccess.ThishasallowedmetoputtogetheraneCSEapplicationforthenextcall,andhopefullyacceleratemyresearchinfuture.”“Archerreallyisagreatservice.IhaveusedothersupercomputersbuttheassistanceandtheguidanceonArcherissignificantlybetter.”“TheUKNationalSupercomputerserviceisanessentialtoolinsupportofscientificresearchwithintheUK.”“Anexcellentservice!Thankyouforallyourhardwork!”
Quotesonthehelpdesk(whichalsoreflectonthecentralisedCSEteam)echothehighratingsforthisaspectinparticularareshownbelow:
“Justtore-iteratethehelpdesk/supportmakesarchersobeneficialandhasreallysavedmeanimmenseamountoftimewhentryingtocompilesoftware,orwhenaproblemhasariseninacalculation.”“Fantastic.Theyknowwhentobringexpertsinonparticularpackagespecificquestionsalso.”“Alwayshelpful,andIlikethefactthattheycheckwithyoubeforeclosingajobtomakesureeverythingisfixed.““Allhelphasbeenpromptanddirectedtowardsimprovingmyexperience.Somesuggestionstoimproveperformancehavebeenofferedwithoutmyrequestandthat'salsoprovedveryuseful.”“Thehelpdesktendstobequicktorespond,efficient,andhelpful,evenwithfairlycomplexrequests.Greatjob!It'salwaysapleasuretoworkwiththehelpdesk.”
Therewereonlytwocommentsonthefilesystemsproblems:
“DiskfailuresandprolongeduseofRaidCheckdiagnosticsgreatlyreducedtheperformanceofthesystemandthethroughputofmyjobsoverthepast12months.““Theissueswiththefilesystemshavehadanadverseeffectonourresearchwork,mostlyintermsofatimelost.“
AfulllistofthecommentscanbefoundinSection5.
![Page 5: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/5.jpg)
5
4. RatingsAllquestionsaskedresponderstoratetheirsatisfactionwitheachparticularaspectofthesurveyonascaleof1to5with1representing“VeryUnsatisfied”and5representing“VerySatisfied”.Table1summarisestheratingsforeachaspectandrevealstheallaspectsoftheARCHERServiceareratedhighlybyusers.Thenumberofresponseswasupfrom153in2014to230in2015.Table2showstheresponsestothe2014surveyforcomparisonpurposes.ServiceAspect
TotalResponses MeanScore(outof5) MedianScore(outof5)
OverallSatisfaction 230 4.3 4Hardware 230 4.1 4Software 230 4.0 4Helpdesk 198 4.5 5Documentation 215 4.1 4Website 221 4.2 4Training 147 4.1 4Webinars 102 3.9 4Onlinetraining 104 4.0 4Table1:SummaryofscoresfordifferentaspectsoftheARCHERService2015ServiceAspect
TotalResponses MeanScore(outof5) MedianScore(outof5)
OverallSatisfaction 153 4.4 4Hardware 153 4.1 4Software 153 4.0 4Helpdesk 198 4.5 5Documentation 142 4.1 4Website 144 4.1 4Training 81 4.1 4Webinars 41 3.6 4Onlinetraining - - -Table2:SummaryofscoresfordifferentaspectsoftheARCHERService2014Table33showsthatthemeanratingsin2015fordifferentaspectsoftheserviceareverysimilartotheequivalentratingsinthepreviousyear.AllaspectsoftheARCHERservicecontinuetoreceiveveryhighsatisfactionratingsfromtheusers.Inparticular,theHelpdeskcontinuestostandoutasthehighestratedaspectoftheserviceinbothsurveys,withanextremelyhighrating.Thisistestamenttothehardworkofallservicepartners(SP,CSEandCray)inensuringthatresponsestotheusersthroughthehelpdeskaretimely,accurate,usefulandpolite.ServiceAspect
2014MeanScore(outof5) 2015MeanScore(outof5)
OverallSatisfaction 4.4 4.3Hardware 4.1 4.1Software 4.0 4.0Helpdesk 4.5 4.5Documentation 4.1 4.1Website 4.1 4.2Training 4.1 4.1Webinars 3.6 3.9Onlinetraining - 4.0Table3:Comparisonofmeanscoresfrom2014and2015UserSurveysfordifferentaspectsoftheARCHERService
![Page 6: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/6.jpg)
6
AscanbeseenfromFigure1,theoverallsatisfactionwiththeARCHERserviceisextremelyhighwithonly8respondersratingtheservicebelow3ona1-5scalefrom“VeryUnsatisfied”to“VerySatisfied”,i.e.3percent.Themeanratingis4.3andthemedianratingis4.Theseratingsareverysimilartotheratingfrom2014UserSurveywherethemeanratingwas4.4andthemedianratingwas4.
Figure1:DistributionofscoresforoverallsatisfactionwiththeARCHERservice(230responsesintotal).
Similarly,forthehardwareandsoftware(Figure2andFigure3respectively),theoverallsatisfactionwiththeserviceishighwithonly5usersratingthehardwarebelow3and9usersratingthesoftwarebelow3.Thesingleratingof1(“VeryUnsatisfactory”)forthehardwareonARCHERwasaccompaniedbyacommentstatingthatthehardwareisfinebuttheturnaroundforNERCjobswasnotgoodrecently.Theratingof1forthesoftwareonARCHERisfromthesameuserwhogavethehardwareascoreof1andhasanegativecommentaboutrecentjobturnaround.Theresponsehasbeensubmittedanonymouslysowecannotfollowupthecommentswiththeresponder.Themeanratingforhardwareis4.1(medianis4)andthemeanratingforthesoftwareis4.0(medianis4).Theseratingsareexactlythesameasthosefromthe2014UserSurvey.
Figure2:DistributionofscoresforsatisfactionwiththeARCHERhardware(230responsesintotal).
0
20
40
60
80
100
120
1 2 3 4 5
Respon
ses
Racng
0
20
40
60
80
100
120
140
1 2 3 4 5
Respon
ses
Racng
![Page 7: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/7.jpg)
7
Figure3:DistributionofscoresforsatisfactionwiththeARCHERsoftware(230responsesintotal).
ThesatisfactionratingsfortheARCHERHelpdeskshowed4responseswithascoreunder3andameanratingof4.5(medianis5).Theseareidenticaltothemeanandmedianinthe2014UserSurvey.Ofthe198responses128,65%,gaveascoreof5(“Excellent”).Oneoftheuserswholeftascoreof1indicatedthattheydidnotwishtobecontactedregardingtheirresponsetothesurveyandtheotherhasbeencontactedforfeedbackastheydidnotleaveacomment.
Figure4:DistributionofscoresforsatisfactionwiththeARCHERhelpdesk(198responsesintotal).
ARCHERdocumentation(fig5,mean=4.1,median4)andwebsite(fig6,mean=4.2,median4)showthesamehighlevelofsatisfactionasthatshownfortheoverallserviceandhavehighrespondentrates.TheresultsforARCHERtraining(fig7,mean=4.1,median=4)arehighandconsistentwiththecoursesurveyresultspresentedintheCSEServicequarterlyreports.Therearenocommentsfromuserswithscoresunder3andanumberofrespondershavescoredfortrainingwithacommentthattheyhadnotattendedanytraining.Asfortheotheraspects,thesescoresareverysimilartothescoresfromthe2014UserSurvey.Thewebinarsandonlinetraininghaveamuchlowerresponderrate(possiblyduetothefactthatthetechnicalnatureofthewebinarsisofinteresttoasubsetofARCHERusers)butshowahighsatisfactionrating(figs8and9,mean=3.9and4.0respectively,median=4).Therewassome
0
20
40
60
80
100
120
140
160
1 2 3 4 5
Respon
ses
Racng
0
20
40
60
80
100
120
140
1 2 3 4 5
Respon
ses
Racng
![Page 8: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/8.jpg)
8
apparentresponderconfusionbetweentrainingandonlinetrainingandwebinars,andthusthequestionsinthesurveyfor2016willberefinedtoavoidthis.
Figure5:DistributionofscoresforsatisfactionwiththeARCHERdocumentation(215responsesintotal).
Figure6:DistributionofscoresforsatisfactionwiththeARCHERwebsite(221responsesintotal).
0
20
40
60
80
100
120
1 2 3 4 5
Respon
ses
Racng
0
20
40
60
80
100
120
1 2 3 4 5
Respon
ses
Racng
![Page 9: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/9.jpg)
9
Figure7:DistributionofscoresforsatisfactionwiththeARCHERtraining(147responsesintotal).
Figure8:DistributionofscoresforsatisfactionwiththeARCHERwebinars(102responsesintotal).
0
10
20
30
40
50
60
70
1 2 3 4 5
Respon
ses
Racng
0
5
10
15
20
25
30
35
40
45
1 2 3 4 5
Respon
ses
Racng
![Page 10: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/10.jpg)
10
Figure9:DistributionofscoresforsatisfactionwiththeARCHEROnlineTraining(104responsesintotal).
0
5
10
15
20
25
30
35
40
45
50
1 2 3 4 5
Respon
ses
Racng
![Page 11: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/11.jpg)
11
5. ListofcommentsThenumberinbracketsindicatestheIDoftherespondent.Hardware• Itisnotpossibletousepartofprocessorsonanode:thisconstrainsscalabilitytoamultipleof24,
i.e.thenumberofprocessorsonanode(14)• Itwouldbeextremelyusefultohaveafewnodeswithmorememory(17)• Lesscrashing(23)• ThefocusontraditionalCPUs,ratherthanaccelerators,isgreatforscientistsneedingtospend
timeonscience,notrewritingtheirsoftware(wheretheiralgorithmisamenable)forGPUs.Formyapplications,interconnectisnotthemostimportantthing,andthefastestcorespossibleisthepriority.(28)
• WouldbenicetohaveaccesstoafewGPUsfortrialpurposes.(32)• Bigmemorynodesareveryuseful-moreofthesewouldbegreat.(34)• Verypleasanttouse,thesystemfastandthequeuesaregenerallyquitequickundernormal
usageloads.(40)• Pleasestopdoingmaintenancemidweek...itcauseshugequeuebacklogsanddestroystheflow
oftheweekattimes(48)• Thehardwareandinterconnectisexcellent,andmuchbetterthantheregionalN8machine
(scalabilityis~200coresonN8vs6000coresonARCHER).(50)• Theissueswiththefilesystemshavehadanadverseeffectonourresearchwork,mostlyinterms
ofatimelost.(55)• Thereisinsufficientstoragespacetopost-processanydata.Datatransfersareslow.(58)• Samearchitectureandsoftwareacrosscomputeandserialqueues.Longeraccess(>12hrs)and
betterperformanceofserialqueuenodes.Infutureinmemorycomputingwouldbeusefultoreducedatasizeandoutputtodisk.(60)
• RAMonindividualnodesareonlowerside.Mostcodesscaleswellwith8GBpercoreRAM.Also,ifgeneralqueueallow72hourjobs,itwillbehelpful.(62)
• Anywaytobetterhandlemanysmallfiles?(66)• Longqueueandwaitingtimefortheinteractivesessions(67)• unknown,yet.(68)• IoftenusethelargememoryCPUsavailableonArcherandduethesmallernumberofthose
specificCPUsthewaitingtimesaremuchlongercomparedtothestandardCPUs.PossiblyIwouldliketoseeanincreaseofthenumberofsuchCPUs.(73)
• Niceandfast!(74)• Itwouldbeniceiftherewaslesslagwhenaccessingfile.(77)• Todate,IhaveonlyusedARCHERfortrainingcourses.(80)• CurrentlytryingoutDDTreverseconnection.Itworkedbrieflywhichwasgoodbutonlyatoy
codefortesting.Itsgreatthatwehavethisfacility(89)• Itwouldbeniceiftherewasadefault6monthextensionofCPUhoursbeyondtheendofa
grant.Manypapersaresubmittedattheendofagrantandextratimeforreviewercommentswouldbeuseful.(91)
• IhaveproducedasubstantialnumberofsimulationsonARCHERthisyear.MostlyIhavebeenveryimpressedwiththespeedofjobsgoingthroughthequeue.However,therehavebeeninstanceswherejobshavewaitedforseveraldaysbeforebeingprocessedandthishasheldupmyresearch.(102)
• ThejobsIrunonArchertypicallyperformbestwithnumbersofcoresthatarepowersof2.With24corespernode,itisnotpossibletoreachapowerof2anduseallavailablecoresatthesametime.Itwouldbemoreusefuliffuturesystemshadsomethinglike16or32corespernode.It
![Page 12: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/12.jpg)
12
wouldalsobeusefultoincreasethenumberofhighmemorynodestoatotalof512.TherearesomejobsIwouldliketorunthatwouldbenefitfromthisnumberofhighmemorynodes.(104)
• ARCHERcanprovidemoreabilityforthepost-processingnodessinceIfoundthatthespeedformypost-processingwithOpenFOAMutilitiesisalwaysveryslow.ThesejobsaresubmittedtoserialqueueinArcher.(108)
• Veryreliable,goodperformance(109)• Greatforrunninglargejobs.Howeverbecausethequeuesareoftenlongitisn'tthatpracticalto
debugsmallerjobsetc.(119)• NoneatpresentasIhaveonlyjustbeentrainedonhowtouseit(121)• Transferspeedsbetween/workdiskto/nercand/nerctoJASMINevenusinglightpatharequite
slow.Thathasbeenthemajorlimitationofhowmuchsimulationswecanperformwithingivendaysratherthanthecomputationspeedorlengthofqueue.That'squiteashameisn'tit?(126)
• Thecomputetimesandqueuetimesaremanageable.However,thefilesystemisoftenextremelysluggishduringweekdayswhenpeopleuseARCHER.Jobsthatnormallytake~2hrscansometimeshitthe5hrwalltimeduetoslowread/writespeeds.(127)
• ExcellenthardwareformyGroup'sneeds!(128)• ForbigdatatasksmoreuseoftheMOMnodesforamotherprogram(i.e.forextremelyparallel
quantummontecarlo)wouldbeniceinthefuture.Otherwisegreat.(131)• It'sbehelpfultohavealarger/workspace(132)• TheonlyissueIhaveisthatreadingandwritingfilesseemsquiteslow.ParticularlywhenIwant
totarupasetofresultssoIcancopythemelsewhere.It'safairlyminorpoint.(133)• Stable,fast,goodamountofmemorypercore.Onlyslightcaveatisthelackofdiskspace.(138)• Stable,fast,goodamountofmemorypercore.Onlyslightcaveatisthelackofdiskspace.(143)• Availabilityof>2GB/corememoryaswellastheexistenceoflargememorynodehasbeen
fundamentaltosustainourresearch.(146)• Thehardwareisfine,whenjobsrun.However,overthepastfewmonthstheturnaroundforour
NERCjobshasbecometerrible.WithdecentturnaroundArcherwouldbeexcellentforourwork.(151)
• Fastercoresandfasternetworking(probablylowerlatencyratherthanbandwidth)wouldgivebetterperformanceforourmassiveatomisticsimulations.Don’tgetmewrong,ARCHERisfantasticbutour40millionatomsimulationsscalelinearlyto3072coresbutonly80%efficientat6144.SoIcantgive"5Exceededrequirements"althoughIlovethemachine.(152)
• Youmaywanttoexperimentaddingacceleratorsonsomenodes.Icanseethatyoumaygainexperienceonthisfield,whichmaybecomemoreimportantinthefuture.Mysuggestionhowever,istokeeptheARCHERastheverylargecomputerclusterthatitisrightnow,andupgradeitwithfastermemoryandmorepowerfulprocessors.(153)
• Goodperformancepernode(155)• DiskfailuresandprolongeduseofRaidCheckdiagnosticsgreatlyreducedtheperformanceofthe
systemandthethroughputofmyjobsoverthepast12months.Additionally,thescheduler'sstrongpreferenceforlarge(>200node)jobsmeansthatuserswho,likeme,requiresmallernumberofcoresforlongerperiodsoftimeareatadisadvantage.Manyofmysimulationsrequireweekstorun,butuseonly16-64nodes.Throughputforthesejobsispoor.(160)
• Itmaybeasoftwareissueratherthanhardware.Thefilesystemslowsdownfrequentlyandonceseveralfilesareinafolder(about1000orso)thefilesystembecomesdifficulttoworkwith(161)
• Morestoragespacewouldbeevenbetter.(171)• IamveryhappywithARCHERandhavenothadaproblemyet.(178)• Thehardwareisverygood,mydown-votingismoreintermsofavailability.Toworkefficiently
I'dneedveryquickaccesstoadevelopmentqueueavailablefordebugging,can'twaitfor2hinthedevelopmentqueueto"sortouttheedges"...Alsoforproductionrunsqueuetimeshavebeenridiculouslylongthisyear.(182)
• Iwaswondering,whetherArcherisabletohavealargememoryloginnodeforsomesimplepost-processinguseifpossible.(183)
![Page 13: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/13.jpg)
13
• InthelastyearstheBarcelonaSupercomputingCenterhasmaintainedsimultaneouslyaCPUbasedmachine(Marenostrum)andaGPUbasedmachine(Minotauro).WiththegrowingpopularityofGPUbasedsimulationcodes,thismaybeasuitableoptiontoexploreinthefuture.(186)
• Asalways,wecoulduseabiggerandfastercomputer...(187)• Prioritisediskspacebeforenumberofcores.ItisveryeasytogenerateTBsofdatawiththe
currentsizeofthemachine,butthereisn'tthediskspaceavailabletoholdit.Thesharednodesareutterlyuselessforcompilingcode,andmakemenotwanttousetheserviceatall.Asthesharednodesareexactlythat,andareoftenoverloaded,whatshouldtake~10minutestocompiletimesoutatanhourormore.(188)
• Matlabonthefewpost-processingnodeswouldbequithelpful.(195)• Archerhardwareexcellent,especiallylargememcores.(198)• justwewaitsometimesforlongtimeinthequeue,ifthereisanysolutionforthisproblemwe
willbegrateful.(208)• Verypowerful.(209)• JustasIneedittobe.(216)• IusetheRDFalotatthemoment.It'sveryusefulformodeldatastorage.IwillusetheCPUs
laterthisyearagain,forUKESM.(218)• Goodarchitecture,butcoulddowithbeingabitlarger.I/Ocanbeextraordinarilyslow,and
there'salotofnoiseinruntimesstill(+/-20%istypical,butsometimesmuchlarger)anditwouldbegoodtosolvethis.(225)
• Pleaseopensomequeuewithlongerwalltime.Forlong(10days)simulations,the24hwalltimemeanswastingourtimeinrestart(229)
• Hardwareisgood.Thenumberofprocessorspernode(24)canmakedomaindecompositionawkward,sinceapowerof2isusuallypreferred,butthisisasmallmatter.(230)
![Page 14: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/14.jpg)
14
Software• InstallationofoftenrequiredMatlab(1)• Sometimespackageupgrades(whichoftenbreakcode)arenotaswelldocumentedasremovals.
(13)• ThelinearequationsolverthatIusedseemedslowincomparisontoothersystems,butIonly
foundthisoutaftertheeventasitwasn'tamajorpartofmycode.Somedocumentationpointingtoperformanceofdifferentlinearalgebratasksondifferentcompilerswouldbeuseful.(18)
• YoucouldaddBerkeleyGWasapreinstalledmodule.(19)• Apermanenttestqueueisneededwithshortwall-times.Withthecurrentsetup,testingisvery
difficultasevensmalljobsmayqueueforinexcessof24hours.(22)• TheCraycompilershaveoftencausedtrouble,andIhaveinsteadhadtorelyontheGNU
programmingenvironment.Librariesandpackageshoweverhavenevercausedmeproblems.(28)
• I'dbehappyifMidnightCommander(unixfilemanager)operatesinafullregime,i.e.withthesubshellwhichactuallyworks.(29)
• Cantakeaverylongtimeforsmallerjobstostartrunning-sometimesupto5days.Wouldbegoodtotweaktheschedulingsystemtomakethiswaitabitshorter.(34)
• Paraviewrunsslowwithlargedatasets,butthismaybeinevitable.(36)• TheonlycommentthatIhaveisthatIwouldlikeautomatichighlightingontheheadnodeforfile
typese.g.tar.gzandfordirectoriestomakeitalittlesimplertodistinguishbetweendifferentitems.(40)
• ARCHERrdfcouldbenefitfromhavingafewmoresoftware.Forexample,NCLcouldbeinstalledaswellasasoftware(otherthangs)forviewingpdfdocuments.(44)
• IwouldneedtousesoftwaretoperformQMcalculationslikeGuassian,QChem,Orca.IunderstandtheseprogramsdonotparallelizeveryefficientlybutitiswhatIneedfordoingmyresearch(48)
• Generallythecompilerisgood,butthecurrentdefaultversionoftheCrayCcompilerisbuggyanddoesn'tcompilemycode(itcompilesonallotherplatforms/compilerswithoutissue).ThenextversioncompilesthecodeOK.(50)
• SomesoftwarewasoutofdatebutonceIemailedthehelpdesk,theyupdatedittothelatestversionquicklyenough.(53)
• Thepythonchangeshavebeenabitofapain-hopefullyit'llbestablenow.(56)• ThecodeI'musingisLAMMPS(nottheprovidedversionbutonewhichincludesmodifications).It
isapitythatLAMMPSdoesnotcompileproperlywiththedefaultcompiler,anditisrecommendedtoswitchinsteadtotheGNUcompilerasthissurelyaffectsperformance.Idon'tknowwhetherthisisacompilerproblem,aLAMMPSproblemorsomethingelse.(57)
• Samesoftware/compilersforcomputeandpostprocessingnodes.Graphical/visualparalleldebugging+performanceanalysiswouldbeusefuloncomputenodes-couldbedoneoninteractivenodesifsamearchitecture+softwareenvironmentisused.Itcanbedifficulttodebug/testusingsmallnon-productiontyperuns.(61)
• Isitpossibletohaveanestimationthewaitingtimeaftersubmittingajob?(67)• Compilersandenvironmentmodulessystemisexcellent.ARCHERisbyfartheeasiest
environmentIhaveusedtocompilescientificsoftware,andagoodselectionisavailable(74)• Todate,IhaveonlyusedARCHERfortrainingcourses.(80)• Maybeitisalreadypossible,andifsopleaseacceptmyapologiesforthiscomment.Ithinkthatit
wouldbeusefulforsomepeople(IknowactuallyquiteafewofthemIworkwith,particularlyinthebio-physics,photo-physics,moleculardynamicssectors)togetaccesstovery-longqueues.Somethinglikelastingaweek.Theydonotneedlargesystems,orlargenodes(materialsscientistslikemedo)justafewCPUsworkinguninterruptedlyforaweekortwo.Theyarenotwillingoritisnotinherentlypossibletoimplementmiddlecheckpointsintheirsoftware.
![Page 15: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/15.jpg)
15
Furthermore,youwouldgetmoreusersfromEngineeringbackgroundbyaddingsomemoremulti-physicssoftwareorfiniteelementssoftware.(86)
• IuseonlymysoftwarebutIhaveaccesstoallthelibsIneedtocompileit(91)• supportabitslow,pythonmodulesstrugglingtokeepupwithe.g.releasesonUbuntuLTS?(92)• Morecompletepythoncomputesetofmodulesforscientificcomputingwouldbeusefuland
savesometimeinstallingthemlocally.(101)• Itwouldbeveryusefulifuserswereallowedtorunscreenontheloginnodes.Also,Iwasasked
recentlytostopusing"ls--color"asitslowsdownthefilesystem.ThisisabitsadconsideringhowusefulitisandthatthisisallowedoneveryothersupercomputerIhaveeverused.(104)
• I'mgladthatARCHERhasPython!(105)• Verygood.(108)• Mostofthesoftwareweuseisourownbutalllibrariesweneedwereavailable.(109)• tmux!(115)• tmux!(116)• Wheneveranythinghasn'tbeenavailableithasbeenaddedquickly.(119)• itisgoodtohavethemakefilesforsomeoftheinstalledsoftwarepackagesavailableattheweb
page;thisservesasastaringpointforowncompilationswherechangesinthesourcecodehavetobeaccountedfor(123)
• Ispent3monthstryingtodiscoverwhatfeatureofmycodeleadtosomenodesrunningmoreslowlythanothers,onlytohavethaterraticperformancebehaviourvanishasaresultofa`systemupgrade'.(124)
• LackofIDL,ncdump,ncview,ncohavebeenthelimitationofARCHER.ThishasbeenareasonwhyIneedtotransferdatatoJASMINorlocalcomputerwhichtakestime.(126)
• Afewdaysago,IhadanincidentwhenIcouldnotcompilemycodeduetoproblemswiththepathtoalibrary.Onthefollowingdaytheproblemseemedtohavebeenresolvedasthesamecodecouldcompilewithoutanyproblems(althoughIdidnotnotifyanyonefromtheARCHERSupportTeam).(128)
• Onesmallthingwouldbeifyourpersonalqstatlistedhowmanyjobsoutofyourtotalyouhavesubmittedsofar-i.e.14/16jobsinqueue.(131)
• Theerrormessagesthatyougetwhenajobfailsareratherimpenetrable-particularlywhenitisjobcontrolfault.(133)
• WenormallyuseRandrgdalpackagetoprocessgeographicaldatabutitrequiresexternallibrariessuchasGDALandPROJ4.WoulditbepossibletohavetheselibrariesinstalledontheArcher?(134)
• Itendtoself-installallnecessarysoftware.(138)• nocomments.Allavailableasneeded.(139)• IdeallyIwouldliketobeabletousethepost-processingsoftwareParaviewremotelyinparallel
forvisualizingresultsfromOpenFOAMCFDsimulations.(141)• WouldbeusefultobeabletorunParaviewonArcherforpost-processing.Thiswouldsavehaving
toregularlydownloadlargequantitiesofdata.(142)• CurrentlyArcherhaslimitedsupportforIntelcompilertools.ItwouldbegreatifthelatestIntel
CompilerSuiteandperformanceanalysistoolsarealsosupported.(147)• Fortrancoarraycornercasesneedmoreresearch.Someunexplainedbehaviourencountered.
Craysupporthasbeenexcellent,butthisseemsanareawhereimprovementscanbemade.(148)• Againthemachineisgreatwhenjobsrun.However,theturnroundhasbecometerrible.Even
quitesmalljobs(64cores)queueformanydays(includingweekends).Notsureifthisasoftwareissueorhardwareorpriority....(151)
• Thereisagreatup-to-dateamountofsoftwareinstalled.Thankyou!(153)• ProvidingcustommodulestootherARCHERusersisstillrathercumbersome.(154)
![Page 16: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/16.jpg)
16
• Thesoftwareisupdatedtoofrequently,particularlythecompilersuitesandthesystemlibraries(e.g.,MPICH),requiringrecompilationofcode.Noreasonsareprovidedtousersfortheupdatesexcepttoincreaseversionnumbers.Iwouldpreferamorestablesoftwareenvironment.(160)
• Themessingwiththepythonstackhasbeenabitfrustrating,butotherwiseit'sgreat.(173)• Iusegromacsanditisworkingverywell(178)• Frequentlybrokenmodulesafterupdatecycles,issueswithpythonlibraries,guilibraries...(182)• great!(183)• IamgenerallyhappywiththesoftwareonARCHER,andanythatisn'tavailablehasbeeninstalled
reasonablywell.Iwouldsaythatsometimesonlythegnuversionisinstalled,whenitwouldbegoodtohaveaversioncompiledunderthecceinstead.(188)
• ModuleconflictsbetweenscientificlibrariesandMPIlibrariesareunfortunate(butdifficulttoavoid).ThereisnotoolthatcanproduceacommunicationtimelinealaVampir.(201)
• SomesortofmoreautomatedarchivingfromworktotheRDFwouldbeuseful.(203)• ReallysatisfiedwithcoherentpackagesofMPI,FFTWandotherlibrariesfordifferent
programmingenvironment.Ifonlythiswasavailableonsmalleruniversityclusters.Unfortunately,theparallelprofilersturnedouttobebuggyandtheydidn'tworkwithmycodeineCSE05-14.Wehavespentalotoftimewiththem.(205)
• Morejobscriptsforthosescientificsoftwaresunderindividualwebpages.(209)• I'mnotanadvanceduser,don'thaveintensivedemands.Soyes,hasmetmyexpectations.(211)• Itwouldbenicetobeabletorunsmallshortscriptstomovefilesaround.(215)• Inmyexperience,ARCHERisstrippedbacktojustwhatisneeded,andIamhappytoanalyse
elsewhere(220)• Havingagraphicsessionandsometooltovisualiseatomistructureswouldhelpourjo(229)
![Page 17: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/17.jpg)
17
Helpdesk• Thehelpdeskisexceptional,allmyinteractionswiththemhavebeenpromptandveryhelpful.
(13)• Fantastic.Theyknowwhentobringexpertsinonparticularpackagespecificquestionsalso.(22)• Alwayshelpful,andIlikethefactthattheycheckwithyoubeforeclosingajobtomakesure
everythingisfixed.(34)• supportwasveryhelpfulinmakingnecessaryscriptstorunjobsonArcherandintestingthe
scriptstounderstandoptimalscalability(35)• Ionlyeverhadtousethehelpdeskonceasthedocumentationonthewebsitecoveredalmost
everythingthatIhaveneededsofar.ThetimethatIdidcontactthehelpdesktheresponsewasquickanditsolvedtheissuethatIwashaving.(40)
• Veryfastresponseandveryhelpfulatsolvingtheproblem.(45)• Veryhelpfulstaffandquickreply.(49)• Didwhattheysaidtheywouldwithinthetimeframetheysaidtheywould.(53)• Greatresponsesandresponsetime.(56)• Fastandhelpful.(58)• Onlinevideos/webinarsuseful,pleasekeepthesegoing.(60)• Makeitpossibletoreplythroughthewebpageforhelpdeskqueries.(66)• veryefficient(68)• Theresponsesarealwaysveryfastandhelpful.(73)• Actuallynothingtosay...youaregreat!Thequerysystemworksperfectlyandefficientlyasnever.
SinceIamaskedandsinceIdiditrecently,Iwouldstandardizetheprocedureofbuyingcomputationaltimeforexternalusersandcenters,tobeabletoaccessArcherbypaying(Imean,otherthanapplyingforgrants).(86)
• Waswaitinganadditional(beyondduedate)twomonthstohavequeryresponse.(92)• Thesupporthasbeenandcontinuestobeexcellent.WithoutthesupportgiventomeIwould
definitelyhavestruggledtoconductmyresearch.(97)• Allhelphasbeenpromptanddirectedtowardsimprovingmyexperience.Somesuggestionsto
improveperformancehavebeenofferedwithoutmyrequestandthat'salsoprovedveryuseful.(102)
• Ihaveusedthehelpdeskalot,withanythingfromquick1linequeries,tomorecomplicatedcompilinganddebuggingquestions.Theteamhavebeeninvaluabletomyresearch(103)
• Thehelpdesktendstobequicktorespond,efficient,andhelpful,evenwithfairlycomplexrequests.Greatjob!It'salwaysapleasuretoworkwiththehelpdesk.(105)
• Thehelpdeskcanalwayssolvemyproblems.Veryhelpfulandprofessional.(108)• Someofmystudentshavecomplainedthatresponsetimehasbeenslowinsomeinstancesand
norealsolutiontoaproblemwasprovided.(109)• Helpdeskalwaysreplyveryquicklyandareveryfriendlyandhelpful.(111)• Veryhelpful.(119)• quickandkindhelp!Thanks!(125)• Helpdeskhasbeenextremelyhelpful.Ireallyappreciateit.(126)• Oftenquickandhelpfulreplies.(127)• Thisworkswell.(133)• AsarecentARCHERuser,I'veaskedanumberofquestionsandrequiredconsiderablesupport
fromtheARCHERhelpdeskandtheresponseshavealwaysbeentimelyandincrediblyhelpful,thankyou.(141)
• Havebeenveryhelpfulwhenansweringallofourqueries.(142)• Helpdeskqueriesareansweredquicklyandinformatively.TheARCHERhelpdeskisoneofthe
mostusefulsupportservicesIhaveused.(160)• Theyhaverepliedpromptlyandhelpedmewhenrequired(178)• Usuallygood,inonecasedidn'treceivefurtherfeedbackandafteragoodwihle(aweekorso)
foundtheissuemyself,fedbackthesolutionbutneverheardback...(182)• great!(183)• Itseemstoworkfine.(188)
![Page 18: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/18.jpg)
18
• Thehelpdeskhasalwaysbeenquicktorespondtomyquieries(203)• Veryefficient.(209)• Notusedsufficientlytoprovideusefulcommentary(220)• Ihaven'treallyuseditthatmuch.(225)• Responsewasveryquickandhelpful.(230)
![Page 19: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/19.jpg)
19
Documentation• Moredocumentationonoptimisingcodeperformancewouldbeuseful.Particularlyfor
estimatingtheamountofAUstobeusedpriortorequestingtime.(9)• Needtoimproveinsomeaspectssuchasthereisnodetaileddocumentationforsubmittingjobs
withdependencyandrestartjobs.(10)• Ithinkthattheweblinkssometimesledmeincircles.(18)• Iverymuchenjoyedtheintroductorytutorialvideos.(19)• Possiblyabitmoreinformationonthebigmemorynodes.(34)• Ihaveonlyhadtousethehelpdeskoncetofindtheanswertoaproblemasthedocumentation
onthewebsiteisverygood.(40)• Thestructuringofthedocumentationisabitconfusingattimes:someinformationisintheuser
guide,someinthebestpracticeguideetcanditisnotclearwherethebestplacetolookis(41)• TheYoutubevideosarehelpfulandthereisalotofwritteninformationbutitcanbehardtofind
whatyou'relookingforinthewrittendocumentation.Atipsandtrickssectioncouldbehelpful.(43)
• Ihavefoundithardtofindinformationonthequeueingsystemandwhatqueueacceptscertainjobs.Particularlyrecentlywhenthelongqueuechangedtogreaterthan25hours-Ifoundthisbytrialanderror.(45)
• Thedocumentationisexcellent.(50)• Mostlyveryclear.(60)• AddhowtheusedAUsarecomputed(mycalculations,usingwhatIthinkhowitisdone,are
factor2off).havenotdiscussedwithhelpdesksofar.(68)• ThedocumentationofGlobusandGrid-FTPintheDataManagementGuidewouldbenefitfrom
anupdate.Moredetailedinstructionsoninstallationoftherequiredtoolsandcertificatesontheclientsystemwouldespeciallybehelpful.(71)
• Irecentlyfoundthetaskfarmjobsinstructionsextremelyhelpful.(73)• Itcouldbealittleeasiertofindthings,buttheinformationiswell-writtenandofferedatabout
therightdepth(74)• Itisverydifficulttofindoutwhichlibrariesareavailable.Itwouldalsobeusefultohavesome
moreinformationonhowthingsarelaidout(e.g.locationoflibrariesandincludepaths).(84)• Justperfect.Whatyouneedinaclearandconcisestyle.(86)• Iknowitisdifficulttokeepituptodatebutreferencingaversionofsoftwarethatisnolonger
availablecouldbebetter.Ifstatingversionnumbersandmodulesoneshouldreallyprovidethedatethatitwasvalid.i.e.thedefaultversionasof....(89)
• Thishasbeenforthebpsscriptsmainly(91)• Thereisnotaspecificdocumentationforlibrariesandmodulesinstalledonarcher.Sometimes
onewanttoknowspecificenvironmentalvariablesandpathsforspecificmodules,butthereisnodocumentationforthat.(94)
• itwouldbenicetohaveitorganizedinamoreuser-friendlyway,witheasyaccesstotheimportantinformation,butmaintaininghigh-levelhelp,possiblyupdatedwiththehelpofusers.Forinstance,aspecialCASTEPorVASPsectionwithbenchmarks,suggestionsandutilitiesfromthecommunitywillbegreat.Ithinkthesectionwithhelpforwindowsusertoconnectandbackupfilescouldbeupgradedwithsuggestionsforspeedingupconnectionanddatatransferaswellaslinkstousefultoolssuchasxming.(96)
• I'velookedatthedocumentationforsubmittingmultiplejobsusingarraysandfounditbaffling.Therewereveryfewcluesastohowtostartwithsubmittinginthisfashion.(102)
• Thereareafewtoomanyguidesavailable,suchastheQuickStartGuide,theUserGuide,andtheBestPracticesGuide.Thedivisionofinformationbetweenthesethreeisnotatallclear,especiallybetweentheUserandBestPracticesGuide.Thesewouldbebetteroffcombined.(104)
• Thedocumentsareveryhelpfulandcansolvemostofmyquestions.(108)• Thedocumentationisgood-clear,rightlevelofdetail,wideranging.However,it'snotalways
easytofindwhatyouneed-it'snotclearwhatisintheuserguideandwhat'sinthebestpracticeguide,soit'seasytomissthingsthatareinthebestpracticeguide.(114)
![Page 20: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/20.jpg)
20
• Sometimesabithardtofindwhatyou'relookingfor.e.g.runningtheserialnodes,accessingthepostprocessingbitofarcher.(119)
• Noteasytogettheinformation...asearchsystemorabetterorganisationasawikicouldbeverygood.(122)
• Thedocumentationonthewebcanbebetter.TherehavebeenafewoccasionswhenIfollowedthedocumentation,didn'twork,consultedthehelpdesk,helpdeskofferedhelpandupdatedthedocumentation.Thatmeansthedocumentationwasnotinitiallycorrectornotupdated.(126)
• IcanfindmostthingsOK.TheonlythingthatIfindhardtofindisthedetailsofthecurrentlyavailablequeuelimits,andtherecommendedsizesofjobstomaximiseutilisation.(133)
• Idon'trelyonthedocumentationoftenanymorethesedays,butit'stothepointandusefulwhenIdo.(138)
• Availabilityofwikicontaininguptodateinformationoncompilationofsupportedmodellingpackagesisawell-definedmeritoftheARCHERdocumentation(146)
• Simplifydocumentationsothatit'seasytopulloutspecifictasks(e.g.compilationagainstspecificlibrariesetc)(149)
• Thedocumentationisscatteredthroughoutseveral"guides",butitisnotclearwhy.Theinformationcouldbemoreusefullyorganisedintoasingledocument.RecentlyIencounteredanissuewithrunningmixed-modeMPI/OpenMPjobswhencompilingsoftwarewiththeIntelcompiler,inwhichthejobwoulduseonlyoneOpenMPthreadnomatterhowmanywerespecifiedthroughthe'aprun'command.Thesolutiontothisproblemwasinthedocumentation,butlistedonlyinthesectionon"pureOpenMP"jobs,whichIhadskippedoverbecausemyjobwasamixed-modeone.Includingthisinformationinthecorrectsectionwouldhavesavedmetwoweeksoftestinganddebuggingmycode.(160)
• Again,usuallygoodbutnotalwayscatchingupwiththeupdatecycles(182)• Thedocumentationisgood,itwouldbebettertoprovidesomekeywordsortingfunction.(183)• WheneverI'veneededtouseit,I'vefoundwhatIneed.(188)• Theuseofthequeuesisstillnotentirelycleartome.Istheallocationofnodesdependenton
totalresourcerequest(i.e.nodehours)orjustnumberofnodesusedinajob.i.e.forlongrunsthatneedtoberuninsequence,isitbettertoruntheseinthelongqueue,orinthestandard?Thisismainlyrelatedtoaproblemforwhenyouneedoneexperimenttoruninconsecutivedependentchunks(i.e.climatemodelexperiments),ratherthanhavingmanyindependentexperimentsthatcansubmittedtothestandardqueue.Sometimes,iffeelslikeotherusersareatanadvantageastheycanhavemanyexperimentsrunningatonce,whileasinglerunagesinthequeues.(203)
• Ifoundalotofusefulthingsithedocumentation.StillIhadtoasktheusersupportforsomedetails,butmostusersprobablydon'tneedthose.(205)
• Thedocumentationisveryusefulforme,especiallynewtoARCHER.(209)• Generallyhelpfulsofar,onlybeenauserforaboutamonth.(211)• Itdidn'tmentiontheworkdirectory!(215)• Notusedsufficientlytoprovideusefulcommentary(220)• TheguidanceonusingDDTwasalittleunclear,nothelpedbytherebeingboth"ddt"and
"allinea"modules.(225)
![Page 21: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/21.jpg)
21
Website• TheonlywayIcanfindthingsisthroughsearch,whichIsupposeisokay.(16)• TheservicestatusinformationisnotasdetailedasitwasonHECToR,whichprovidedacountof
whichqueueswerebusiest.Thiswasusefulwhenconsideringwhethertosplitlargejobsorrunwhole.(28)
• IwastryingtofindinformationonhowmanyAUsusedcertainsoftwarepackagemodules(e.g.CASTEP)onARCHER,togetaroughestimateonhowmuchARCHERtimegoestowardsCASTEP,LAMMPS,etc(beingawarethatpeoplemayforgotheARCHERmodulesinfavouroftheirowncompiles.Anyway,Icouldnotfindsuchastatisticonmoduleuse.(37)
• Safecouldbeimproved.Therightbuttoncanbehardtofindtoperformanoperation.Fornewuserstogainprojectaccessitisnotobviouswhattheuserortheprojectmanagerwillsee/do.(60)
• Navigationcouldbeeasier,butagoodrangeoffunctions.(74)• It'susuallyeasyenoughtofindthematerialIneedviagoogle/searchonwebpage.(75)• Findinginformationiscomplicated.ArcherSAFEusedtobeparticularlynon-trivialtolocate.(77)• Itismoredifficultthannecessarytoaccessthecoursedescriptionsfromthecourse/event
schedulingpage.Description/registration/schedulinginformationshouldbeincludedinthesameplace.(80)
• itwoulddefinitelybenefitofanupdateinstyle(abittoomuchold-fashioned)andcontent.(96)• Sometimesthecompilation/jobscribsareoutofdatebuthelpwasalwaysathandwhen
contactingthehelpdesk.(97)• Thetextisalittleconfrontational,butnavigationisgood.(102)• Ifindtheoverallstatus/livereportsveryhelpful(103)• Itwouldbeveryusefultoshowinformationoncompletedorexpiredallocationsontheuser-
specificpage.Onceallocationsarefullyusedorexpired,theydisappearcompletelyasiftheyneverexisted.Suchinformationisusefulwhenwritingproposalsandisalsousefulwhenanallocationstopsbeingactivebeforeauserexpects.(104)
• Sometimesabithardtofindwhatyou'relookingfor.e.g.runningtheserialnodes,accessingthepostprocessingbitofarcher.(119)
• Atextbookhasanindex.Websitesrequireyoutoknowwheretheinfoisbeforeyoustartlookingforit.Archer'swebsiteisnodifferent.(124)
• Servicestatusdoesnotalwaysshowstheproblem.BylookingatthegraphsdownthepageIcanseeit'squitecrowded,butitdoesnotprovideanyclueabouthowmuchlongermyjobneedstobeonthequeue.(126)
• Thiscanbetrickytonavigate-especiallytheSAFEtool...notentirelyclearhowcertainprocesses(e.g.getanewpassword)aremeanttohappen.(137)
• Thewebsiteisveryuseful.However,theremainingbudgetsectiononlyseemstoupdateoccasionallywhichhasledtoover-usingtheavailablebudgetonoccasion.(142)
• Navigatingbetweenthemainsiteandarcher-safecouldbeeasier.(152)• Someusefullinksandinformation,sometimeshardtofindcertainpagesforexample,ifthere
wasalinkinanemailandIwantedtolookthepageupagainlaterwithoutthelinkfromtheemail,usingthesearchtoolonthearcherwebsiteitcanbehardtofindthatpagesometimes.(181)
• great!(183)• Somethingshardtofindbecausenotsurewhatwordsyouuse.Onceaccustomedtothe
languageallwasfine.(187)• Nicelayout.(209)• Good(211)• Couldbemadeclearer,especiallythedocumentation,currentformatlooksmessy(214)• I'vealwaysfoundthewebsitetohavewhatIneed.Inpart,thisprobablystemsfromthelimited
wayinwhichIneedtointeractwithARCHER(e.g.runninglargeoceanjobsonit).(220)• Uptodate.cool!=D(229)
![Page 22: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/22.jpg)
22
Training• Iappreciatethehighfrequencyofwebinars.(19)• notused,butnowIcannotunclicktheanswer.(37)• Ihavenotcompletedanyofthese,ofmyownfault,theyarewelladvertisedthroughthemailing
lists.Onlygivena3becauseIhavenevertakenadvantageofthisfacility.(40)• Audioqualitycanbeanissuewhenfollowingvideoguides.(43)• Therangeoftrainingavailableisgreat.(50)• Onlinevideos/webinarsuseful,pleasekeepthesegoing.(60)• IattendedtheScientificPythoncourseparticularlyforthesectiononintegratingPythonwith
otherlanguages,e.g.C,Fortran.However,asthecoursehadrunovertimethissectionwasalmostcompletelyskipped.Iwouldbeinterestedinmoremedium-levelcourse,orhands-onworkshopsforpeoplewithsomeHPC/codingexperiencebutnotadvancedusers,onthingslikedatamanagement/analysisusingbatchprocessing,particularlyhowtousedifferentsoftwarepackagestocompletedifferentpartsoftheanalysis.(Iamawareofhowtodothiswithbash+gnuplot,buthavenoideahowtointegratesay,datageneratedinfortranwithanalysisdoneinMatLab/Mathematica/Origin.)(80)
• UnfortunatelyIdidnotusethem,butIwillgiveitagointhefuture.Ibelieve,lookingatthetrendsaround,thatdeliveringlecturesonbasicandadvancedFORTRAN(orCandC++)andontheuseofGPGPUandCUDAwillbeenoughfornormalusers.Furthermore(andIamnotsurehereagainalreadyyouofferthispossibility)youcanincreasetheinterestofprogrammingcourses(sometimesboring)ifyouaddspecificonesonapplications(Iamthinkingaboutfluid-dynamics,astrophysics,high-performancemoleculardynamicsetc...)(86)
• HavenotattendedsomeastheyarebasedinEdinburgh.(89)• Iwouldhavelovedtoparticipatetothecomputationalsoftwaretraining,butitisdifficultforPIs
toattendafullday(ormore)oftraining.EventsinLondonoreventsonlineworkbetterforme.Iwouldincrediblyappreciateifthematerial,examples,tutorialsandpresentationsvideosandslideswouldbemadeavailabletotheARCHERcommunity,especiallytousersthatcouldnotattendthesessions.(96)
• IattendedthePythonandFortranTrainingsessionandtheywereveryuseful.Mayberunningdifferentlevelsoftrainingmightbeworthwhileinthefuture.(97)
• TheylookveryinterestingalthoughIhaven'tactuallyattendedany.Moreonpython.(119)• Veryinformative(121)• Iattendedone.itwasverygood.Ijustfeltnotenoughdaysandpracticalworks.(122)• IhaveonlywatchedoneTutorialfromtheARCHERwebsite.TherestinformationIneededI
foundinwrittenform(ontheARCHERwebsite).(128)• Iwishsomeofthesewereeasiertogetto...(133)• MPIinPythonmightbeausefuladditiontotheMPIcourse(whichcurrentlyincludesCand
Fortran)(137)• Ihavenotreceivedrecenttraining,soIgiveaneutralscore.(138)• IhaveattendedseveralTrainingsessionsandtheyhavebeenreallyusefultome.(147)• Ihavenotattendedtoanyofthewebinars,asIhavelongexperienceinHPCand
supercomputing,buthavelookedatthewebsiteandthematerialslookgreat.However,youcouldtakeastepforwardanddosomeworkshopsonnumericalapproaches,specificlibrariessuchasPETScorTrilinos...YoucanfindinspirationintheworkshopsandtrainingsessionsthatareofferedintheSuperComputingconferenceinthestates.(153)
• IfoundtheAdvancedMPIcourseabittoofastpaced+toomuchnewinformationcovered.Thiswasfedbackonthefeedbackformforthecoursealongwithsomesuggestionsforimprovementetc.(156)
• Itisveryusefulindeed!thanks!(183)• -Foryourinfo,Ihaven'tattendedanyARCHERtrainingsofar,butitisusefulforARCHERto
providethese.(184)• Didn'tdothismyself,butheardgoodstoriesfromothers.(187)• Couldyouimproveyourannouncementsaboutyourtrainingcoursestobemoreinternational.I
meantherearemanyinternationalstudentsneedyourcoursesbuttheydon'tknowaboutyourservices.(208)
![Page 23: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/23.jpg)
23
• Itcoversmanyusefultrainingcoursesforme.(209)• Ihaven'tbeentoany,butI'veheardgoodreports.(225)
![Page 24: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/24.jpg)
24
Webinars• Thiswasveryuseful.(18)• Someoftheintroductionstoagiventopicwereabittoobasicforme,butIsupposethecluewas
inthetitle...(53)• Thewebinarsoftwareisabitclunky,asimplewebinterfacewouldbenice.(74)• TheHPCoverviewcourse(recommendedasthestartingpointfornewusersplanningtotakethe
HPCdrivingtest)washardtofocusonremotely.Thereweretechnicaldifficultiesatthebeginning(videosonlyshowedpresenterwithslidesinthebackground,ratherthanavideoofthepresentationnexttotheslidesthemselves).Thepresentersdidagoodjobofinformingparticipantsoftheissuesandupdatingtheslidesviathecoursemailinglist,butIfounditdifficulttosetasidethetimeduringtheworkdaytowatchthe3hourpresentationaftertheineffectivefirstsession.(80)
• Oftencannotattendastheyclashwithotherwork..HaveusedYouTubechannel.Workedwellforme.Althoughononeoccasionthepresentationwasdifficulttofollowduetofontsize.(89)
• GoodforaskingquestionsbutgenerallyItendtoswitchoffhalfwaythrough.(97)• Somewereuseful,someabittoobasic.Thearchiveofmaterialsfrompasttutorialsisveryuseful.
(114)• It'sgreatthatthewebinarsareavailableonyoutubeaftertheeventdate.(158)• Usefulandinteresting!(183)• Didn'tdothismyself,butheardgoodstoriesfromothers.(187)• Lateafternoontimeslotsformanyofthemissomewhatinconvenient.Otherwise,excellent(198)• Veryuseful(209)
![Page 25: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/25.jpg)
25
OnlineTrainingMaterial• Honestly,IonlydidthedrivingtesttogettheextrakAUs,levelwasverybasic.(20)• Possiblyhaveextensionstothedrivingtestformorecomplextasks.(34)• OneofmystudentsrecentlyworkedthroughtheinitialtrainingforArcher'snewuserandhewas
verysatisfiedbytheinstructions.(73)• Needtobemorecarefulwhenreferencingspecificversions(89)• OnlinedrivingtestisagreatwaytogetpeopleinvolvedinArcherandHPC!(115)• OnlinedrivingtestisagreatwaytogetpeopleinvolvedinArcherandHPC!(116)• ItooktheOnlineDrivingTest,andthinkthatitisagreatideatoasitintroducesthefutureuser
nicelywhilemotivatinghim/hertolearnhowthemachineisconfiguredbeforeusingit.(153)• Itisextremelyusefultomycodingandparallelismimplementation.(183)• IdidtakethedrivingtestIthink,butIcan'trememberitbeingparticularlyusefulI'mafraid.(188)• extendawardlimits(twice-triplekAuallocations)ifresourceswouldallowthat.(195)• Needmoreexamplesforexplanation.(208)• Itisveryusefultoprovidestep-by-stepinformation.(209)• Drivingtestwasgood.IrepeatedittillIgot100%,couldn'tfigureoutwhetherthiswasrequired
ornot,don'tthinkIreaditanywhere(211)
![Page 26: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/26.jpg)
26
Othercomments• OverallmyexperiencewiththeARCHERservicehasbeenverypleasant.(13)• I'vehadafewMAJORproblemswithARCHERoverthepastyear.First,thequeuesystemis
absolutelydisgraceful.Havingtoqueuemorethan48hoursforasmall,3-minuterunisridiculous.Ifoundthatsubmittingtheseshortrunstothelongqueuesolvedmyproblem,whichIrealizeisnotwhatyouhadinmindforthelongqueue,butwhenyouareundertimeconstraints,sometimesthisisnecessary.ThefactthatARCHERwasdownforthebetterpartoftwomonthsisanationaldisgrace.YouaresupposedtobetheUK'sbestandfastestsupercomputer.ThatwasthekeytimeformetorunsimulationsformyPhD.HavingmymodelspostponedfortwomonthsnearlycostmemyPhDasIhadnooptionofextendingmyfunding.Ihopeyouhavelearnedfromthoseproblemsandhaveasysteminplacetopreventthisfromhappeninginthefuture.(16)
• Itwouldbegreattobeabletousetheshortqueuealldaylongandnotonlybetween09:00-16:00.(19)
• Archerreallyisagreatservice.IhaveusedothersupercomputersbuttheassistanceandtheguidanceonArcherissignificantlybetter.(27)
• Thestandardqueuesareoftenpainfullylong.Myjobstake6hourstoruniftheyaresuccessful,butinthelastfewmonthshavetypicallyqueuedfor12-24hoursbeforestartingtorun.24hoursisalongtimetowaitforsomethingtocrashassoonasitstartstorun,whichisacommonoccurrencewhentestinganewsetup.Isthereanywaythatthequeuingsystemcouldbechangedtoreducewaitingtimesforuserswithrepeatedlyfailingjobs?(38)
• Itisaterrificresource(42)• Theonlyfrustratingthingiswhenthequeuesalwaysbecomeblockedjustbeforearenewalof
budget.IknowthereisnotmuchthatcanbedonetopreventthisbutitstheonlyfrustratingthingaboutusingArcher.(45)
• TheonlinedrivingtestwashowIobtainedaccesstoARCHER,andIthinkitsabrilliantideaforwideningaccess.ThishasallowedmetoputtogetheraneCSEapplicationforthenextcall,andhopefullyacceleratemyresearchinfuture.(50)
• Archershouldmovetowardshigh-throughputcomputation,muchmoreusefulforthevastmajorityofresearch,thanthecurrentsinglempiprocesspreference.(76)
• Pleasegivemetheprizesof2000kAUonArcher;-)!!!(86)• Justtore-iteratethehelpdesk/supportmakesarchersobeneficialandhasreallysavedmean
immenseamountoftimewhentryingtocompilesoftware,orwhenaproblemhasariseninacalculation.(97)
• Ihaveafewcriticalcomments.1)Whyisthe'short'testingqueueonlyopenfrom9-5UKtime?Iamnotalwaysabletodomytestingworkduring"normalbusinesshours".IhavetobeflexibleastowhenandwhereIwork.SometimesIneedtoruntestswhileworkingfromsomeothertimezone-whyshouldIbepenalisedforworkingfromabroad?Havingatestingqueuethatisonlyopenfrom9-5hampersmyprogressandisveryfrustrating.Itisalsoveryunusual-IhaveworkedonalotofHPCsystems,andARCHERistheonlymachineIhaveencounteredwithsuchastrangetime-limitedqueue.Itreallyisbizarre.2)Theattitudethat"ARCHERisnotatestingplatform"doesnotrespecttheneedsofocean/atmosphere/cryospheremodellers.Ourmodelsare*always*indevelopment.Wetypicallyhavetocompileandrun*each*case(e.g.differentcompile-timephysicalpackages,adjointsetups)andtestit.Whyshouldn'tIbeabletotestmysetupwithashortrun?Mytypical"testrun",justtomakesurethatmyinputsareconfiguredcorrectly,islessthan1000coresforabout10minutes.Ishouldn'thavetowaitonthequeueforhoursjusttoperformthissmalltest.Whydoesn'tARCHERrespectourneedsforthisveryminimalkindof"testing"?Inshort,Ireallyneedalarger"short"testingqueuethatisopen24hoursaday.3)Whatcanbedoneaboutthe"rushtoburnhours"inFebruaryandMarch?ARCHERisnearlyunusableforthosemonthsbecausebothNERCandEPSRCusersaretryingtouseupalltheir
![Page 27: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/27.jpg)
27
hoursfortheyear.Whycan'tthoseallocationyearsstartandendatdifferenttimes?4)It'stoocumbersometoshareinfobetweenprojects(e.g.n01,n02).Ihavecollaboratorsinbothn01andn02,andsometimessharingfilesonARCHERbetweenprojectsisarealheadache.Therearetoomanywalls.5)PleaseclearlypublishanychangesinyourSSH/SCPpolicies.Sometimeinthepastyear,adecisionwasmadetodisallowoutgoingSSHconnections.Thismadeitimpossibleformetodomywork,sinceIuseTAF(aFastOpt.comproduct),andTAFusesoutgoingSSHconnectionsforserver-to-servercommunication.IwastedalotoftimewonderingwhyTAFwasn'tworkingforme.TheARCHERhelpdeskdidrespondtomyrequestfairlyquickly,andtheissuewasfixedoverthenextcoupleofweeks.Ifpossible,pleasebemoreexplicitaboutthesethingsinthefuture.Forexample,perhapsinsteadofjustkillingtheSSHsessions,youcouldhaveanerrormessageintheterminalexplainingthatthesessionisbeingkilled(andwhy)?(105)
• OntheSAFEwebpage,itisveryusefultohavethetablewhichsummarisestheprojectsyouareinvolvedin.Remainingbudgetisreportedonthistable,anditwouldbereallyusefultohavereportedhowmuchofyourtotalallocatedtimethiswas,andwhenitrunsout.(111)
• Itwouldbeveryhelpfultogetawarningmessageifyoutrytosubmitajobtotheshortqueueoutofthe9-5hours;alsoifyousubmitajobbefore5pmwhichislikelytobestillrunningat5pm.It'seasytoforgetaboutthelimitedhoursforthisqueue.(114)
• Itwouldbeinvaluabletohaveaseparateresourcewithbetterqueuesforsmallerruns.(119)• Ifthereisawaytoincreasebandwidthforthefilesystem,pleasedoso.I'vehadseveralcrashes
duetoextremelyslowread/writespeedsformyjobs,andpostprocessingdataismuchsloweronARCHERthanonothermachines.(127)
• IfindtheARCHERservicetoberesponsiveandhelpful,easytouse,andflexible.Excellentservice.(132)
• Ihavefoundthatmanagingthebudgetofprojectshasbeenalittledifficultduetoadelay(~24hours)inupdatingtheremainingbudget.Ihavefoundthatthequeuingperiodscansometimesbequitelong.IunderstandthatreducingthewalltimeofsimulationsgivesprioritybutinmyworkItendtorunrelativelyfewlongjobssoendupwithlongqueuetimes.Theuploadanddownloadspeedoffilescanbeincrediblyslowascandeletingfiles.ForsimulationsgeneratingTBsofdatathiscantakeareallylongtime.(141)
• Downloading/uploadingdatafromArchercanseemverysloworevenstallattimes.Whenthishappensitisoftenquickertostopthedownloadandrestartitagain.Furthermore,deleting/copyingdataonArcherseemstobeslow.(142)
• Itwouldbegreatifthereisfurthersupportforshorterdurationdebugqueues(147)• welldone!(148)• Theserviceisworsenowthaneverinthepastduetothejobturnaround.Ireallyhopethatcan
besortedsoon.ItisironicasArcherhasbeenupgraded,buttheserviceisworse.Therestofthefacility(machineandstaff)isexcellent.(151)
• Keepuptheexcellentwork&Thanks(152)• Pleasecommunicatesystemissuestousersmorefrequentlyandinamoretimelymanner.I
understandthatthereisabalancetobestruckbetween"bombarding"userswithemailsandkeepingusersinformed,butARCHERalreadyhasdifferentlevelsofmailingliststhatcouldbeusedforthispurpose.Overthepastyear,therewereseveralinstancesofovernightsystemfailures/machinerebootsthatwerenotcommunicatedtousers.Thoseofuswhohadjobsrunningatthetimereceivedcrypticerrormessages,whichinitiallyledustoblamesoftwareerrorsratherthanahardwarecrash.OnlyafteremailingtheARCHERhelpdesktoinquiredidwediscoverthattheentiremachinehadgonedownovernight!Hardwareerrors(particularlydisks)andsubsequentmaintenancehavemarredperformanceofthesystemoverthepastyear.(160)
• 16joblimitonqueueisfrustrating--Iunderstandtheneedforthrottling/controllingthenumberofjobs/coresactiveperuser,butthequeueshouldhandlethat,notrejectjobsbasedonahard(andquitesmall)limit.(167)
• Wonderfulservice,myresearchwouldn'tbethesamewithoutit!(169)• Anexcellentservice!Thankyouforallyourhadwork!(172)
![Page 28: ARCHER User Survey · 2016-07-05 · Status Final Author(s): Anne Whiting Reviewer(s) Alan Simpson, Andy Turner Version Date Comments, Changes, Status Authors, contributors, reviewers](https://reader033.vdocuments.mx/reader033/viewer/2022050502/5f9499860f92bc47481d36ca/html5/thumbnails/28.jpg)
28
• Thetimetakentoqueuejobstorunisquitefrustrating.A1.5hrjoboftenneedstobesetupthedaybeforetorun.(175)
• someofthemoreusefulworkshopsfillupreallyquickly,e.g.pythonrelatedones-woulditbepossibletohavetheseclassesmorethanonceinayear?(181)
• Great-Thanks.Keepgoing.(183)• -WerunOpenFOAMsimulationsthatproduceafewthousandsoffiles.Unfortunatelythisisa
problemofOpenFOAM.HoweverwhenafewresearchersonaprojectareusingARCHERintensively,wefindthatthelimitofallowablefiles(3million)onARCHERissometimesexceededandbringsahalttoalloursimulations.Thiscouldbeimproved,bysayallowingthesespecialcasesanincreaseinthislimit.(184)
• Myresearchfieldisdataassimilation,inwhichhugegeosciencemodelsofe.g.theatmosphere,withupto10^9variables,arecoupledtohugedatasets,upto10^8every6hours.SincetheseproblemsarenonlinearMonte-Carlomethodsareneededforthis.Present-dayARCHERisnotfitforpurpose,andonlycentresliketheMetOfficecandothisatthemoment.However,theyneedtheacademiccommunitytopushthequalityoftheirmethodsforward,bothintermsofaccuracyandefficiency.IntheUSAtheacademiccommunitydoeshaveaccesstomachinesthatcandothis,weintheUKarebehind,whileatthesametimethenewestandmostexcitingnewideasarecomingfromtheUK,throughclosecollaborationofmathematicians,statisticians,andgeoscientists.Thisisnotonlyforweatherpredictionandbetterunderstandingoftheweather,butalsoforoceanography,biology,ecology,land-surfacescience,climatemodelling,etc.Thereareveryexcitingandinnovativethingswecoulddowithabiggersystem.(187)
• Changethewaythesharednodeswork,oratleastmakesomenon-sharedserialnodesthatcanbeusedforcompiling.Itisverydifficulttocompilewhenyoudon'tknowhowlongthingswilltaketocomplete.ThisisparticularlyproblematicwhenrunningtrainingcoursesonARCHER,whichIdoeveryyear(UKCA).Whilewecanreserveparallelnodes,Ifirstgetthestudentstocompile.Whilethisshouldtake10minutes,theyallthentime-outafteranhourasthereare20+peoplehittingthesharednodes.Itdoesn'tjustwastetheirtimeandmine,butalsowe'repayingforparallelnodesthatwecan'tuse.Thisismybiggestproblemwiththeserviceasitiscurrentlyconfigured.IfIcouldcompileintheparallelqueue,orreserveserialnodes,thiswouldimprovethingsgreatly.WhenIcalledthehelpdeskaboutthistherewasnothingtheycoulddo.(188)
• TheUKNationalSupercomputerserviceisanessentialtoolinsupportofscientificresearchwithintheUK.(201)
• Pleaseprovidedataanalysistutorials(foratmosphericmodeloutputse.gUKCA)(207)• Goodexperience.(209)• Pleaseincreaseuser'squotaandgiveusmorespace.Muchtimeiswastedinbackups(229)• Standardqueuewaitingtimescanbeexcessive:Ihaverecentlybeensubmittingjobsofabout11
nodesfor12hours,andthesehavequiteregularlyspent2-3daysinthequeue.Irealiselargerjobsareprioritised,butafasterturnaroundonthesesmallerrunswouldbeuseful.(230)