LUMI AI Fac o y Se ice Cen e
Empowe ing Eu ope’s AI Ecosys em
D5.1 Tooling o objec s o age
2
Deli e able 5.1
Tooling o objec s o age
D5.1 Tooling o objec s o age
3
P ojec Ti le
LUMI AI Fac o y Se ice Cen e
P ojec Ac onym
LUMI-AIF
P ojec Numbe
101234208
Type o Ac ion
HORIZON-JU-RIA
Topic
HORIZON-JU-EUROHPC-2025-AI-01-IBA-01
S a ing Da e o P ojec
01.03.2025
Ending Da e o P ojec
29.02.2028
Du a ion o he P ojec
36 mon hs
Websi e
lumi-ai- ac o y.eu
Wo k Package
WP5
Task
T5.3, T5.1
Lead Au ho s
Ma in Golasowski (IT4I), Jan Ma ino ič (IT4I)
Con ibu o s
Abdul ahman Azab (Sigma2), Ja no Lai inen (CSC), Lo and
Janos Szen annai (Sigma2)
Pee Re iewe s
Pauliina Some koski (CSC)
Ve sion
1.0
Due Da e
31.8.2025
Submission Da e
29.8.2025
Dissemina ion le el
X
PU: Public
SEN: Sensi i e – limi ed unde he condi ions o he G an Ag eemen
EU-RES. Classi ied In o ma ion: RESTREINT UE (Commission Decision 2005/444/EC)
EU-CON. Classi ied In o ma ion: CONFIDENTIEL UE (Commission Decision 2005/444/EC)
EU-SEC. Classi ied In o ma ion: SECRET UE (Commission Decision 2005/444/EC)
D5.1 Tooling o objec s o age
4
Ve sion His o y
Re ision
Da e
Edi o s
Commen s
0.1
9.7.2025
Ma in Golasowski
Ini ial s uc u e o he documen
0.2
7.8.2025
Ma in Golasowski
In oduc ion comple e
0.3
22.8.2025
Abudl ahman
Azab, Ma in
Golasowski, Ja no
Lai inen
Consolida ed desc ip ion o na ional
esou ces, added sec ion abou da a
ans e s
0.4
26.8.2025
Ma in
Golasowski,
Lo and Janos
Szen annai
Finished da a spaces and summa y sec ion,
added No wegian esou ces
0.5
29.8.2025
Ma in Golasowski
Finaliza ion, LUMI AIF abb e ia ion ix, AI
decla a ion added
1.0
29.8.2025
Anna Luoma
Final quali y checked pe o med by he
PMO, sen o o icial e iew.
Decla a ion on he Use o AI Assis ance
This epo has been p epa ed wi h he suppo o Cha GPT and GPT-5-enabled Mic oso Copilo ,
which we e used o assis in d a ing ex , checking language and con en consis ency, and ga he ing
backg ound in o ma ion. All con en has been ho oughly e iewed, ac -checked, and edi ed by he
au ho s o ensu e accu acy and alignmen wi h he objec i es o he epo .
Glossa y o Te ms
I em
Desc ip ion
ACL
Access Con ol Lis
API
Applica ion P og amming In e ace
AWS
Amazon Web Se ices
CLI
Command Line In e ace – e minal clien
EOSC
Eu opean Open Science Cloud
FTP
File T ans e P o ocol
GUI
G aphical Use In e ace
GPU
G aphics P ocessing Uni
HPC
High Pe o mance Compu ing
HTTP
Hype Tex T ans e P o ocol
IETF
In e ne Enginee ing Task Fo ce
D5.1 Tooling o objec s o age
5
iRODS
In eg a ed Rule-O ien ed Da a Sys em
ISO
In e na ional O ganiza ion o S anda diza ion
NFS
Ne wo k File Sys em
POSIX
Po able Ope a ing Sys em In e ace
SSH
Secu e Shell – p o ocol o emo e command line in e aces and da a
ans e
URL
Uni o m Resou ce Loca o
Execu i e Summa y
Desc ip ion o objec s o age concep s, implemen a ion and a ailable ooling ha should be p o ided
along wi h suppo o he LUMI AIF use s. In eg a ion wi h compu ing esou ces ocusing on e icien
da a s aging and caching is discussed as well as connec ion o ex e nal da a eposi o ies and esou ces
such as Eu opean Da a Spaces. A ailable s o age esou ces pe each LUMI AIF pa ne coun y a e
summa ized along wi h o e iew o common echnologies a ailable in Eu opean HPC cen e s.
D5.1 Tooling o objec s o age
6
Table o Con en s
1. In oduc ion ............................................................................................................... 7
S a e o he a in he objec s o ages 7
LUMI objec s o age sys em 7
S3 implemen a ions 7
2. Cu en ooling o objec s o age ............................................................................... 8
Command line ools 9
So wa e De elopmen Ki s (SDKs) 9
G aphical Use In e aces (GUIs) 9
AI o ien ed s o age managemen solu ions 10
O he objec -o ien ed solu ions 10
Connec ion o ex e nal da a esou ces 11
3. In eg a ion wi h compu ing esou ces....................................................................... 12
S aging on compu ing esou ces 13
3.1.1 SSH based access 13
3.1.2 In-clus e s aging 13
3.2 Local da a caching – inges ion om ex e nal esou ces 14
4. Ex e nal da a eposi o ies and esou ces .................................................................. 15
Na ional esou ces ac oss he conso ium pa ne s 15
4.1.1 Czechia 15
4.1.2 Denma k 16
4.1.3 Es onia 16
4.1.4 Finland 17
4.1.5 No way 17
4.1.6 Poland 18
Objec s o ages ac oss Eu opean HPC cen e s 18
5. Summa y .................................................................................................................. 18
Re e ences ...................................................................................................................... 19
D5.1 Tooling o objec s o age
7
1. In oduc ion
S a e o he a in he objec s o ages
The concep o objec s o age eme ged in he ea ly 2000’s as o ganiza ions aced inc easing challenges
s o ing and managing la ge olumes o uns uc u ed da a. I s de elopmen aimed o o e come he
limi a ions o adi ional s o age sys ems in e ms o scalabili y, me ada a handling, and access me hods.
One o he i s majo miles ones was he launch o Amazon S3 (Simple S o age Se ice) in 2006, which
showcased how objec s o age could powe la ge-scale cloud in as uc u e. Since hen, i has become a
ounda ional echnology in cloud compu ing, used widely by public cloud p o ide s, en e p ises, and
so wa e-de ined s o age solu ions o i s du abili y, cos -e ec i eness, and ease o in eg a ion wi h
APIs.
Howe e , despi e widesp ead usage, he e is no s anda d o he S3 p o ocol and he e can be suppo
issues be ween di e en clien s and e sions. The e a e di e en implemen a ions o he p o ocol, which
implemen he S3 API. Con a y o he HTTP as an example, he e a e no IETF o ISO s anda ds o S3
API.
I is wo h men ioning ha objec s o age is p ima ily o e ed as a se ice wi h a well-de ined API, which
is supposed o be ei he by expe use s o de elope s. The e a e many use-cases o objec s o ages like
backup o a chi e s o age and collec ion o da a om IoT de ices. Howe e , all hese use cases o en
build hei own in e aces and se ices on op o he objec s o age API. AI ela ed use-cases a e no
di e en , app op ia e ooling and se ices mus be buil on op o hese in e aces making he echnical
complexi y anspa en o he use .
LUMI objec s o age sys em
LUMI-O is an objec s o age se ice accessible h ough a web in e ace and S3 API, and LUMI AIF
con inues o ope a e simila se ice. LUMI-O is based on comple ely di e en echnology han he Lus e
pa allel ilesys em (LUMI-P and LUMI-F). Also, unlike he Lus e pa allel ile sys em, LUMI-O is a sepa a e
se ice om LUMI compu e pa i ions. We discuss his opic in mo e de ail in he chap e "In eg a ion
wi h Compu ing Resou ces" below.
LUMI-O o e s a o al o 30 PB s o age space o s o ing, sha ing, and s aging o da a. By de aul , 150 TB
o s o age space is alloca ed pe LUMI p ojec , bu p ojec s can eques o mo e LUMI-O s o age space
by con ac ing he LUMI helpdesk.
S3 implemen a ions
This sec ion p o ides a lis o common implemen a ions o he S3 API, including he AWS i sel . I is wo h
no ing ha many o he solu ions o e good scalabili y wi hin a single da acen e , and scaling ac oss
mul iple loca ions equi es building a complex ne wo k in as uc u e, including load balance s and
D5.1 Tooling o objec s o age
8
poin s-o -p esence o hide he physical limi s o a single da acen e o he use s (i.e. hype scaling). Gi en
he dis ibu ed na u e o compu e and s o age esou ces ac oss Eu ope, good da a ans e solu ion is
needed o mo e da a e icien ly be ween loca ions, which is a ea u e ha is usually no o e ed by he
solu ions lis ed below.
Amazon S3 (Simple S o age Se ice) is a highly scalable, du able, and secu e objec s o age se ice
p o ided by Amazon Web Se ices (AWS). Launched in 2006, S3 was one o he i s se ices o e ed by
AWS and emains a co ne s one o i s cloud in as uc u e. S3 allows use s o s o e and e ie e any
amoun o da a om anywhe e on he web using a simple web in e ace o REST ul API. I o ganizes da a
in o bucke s and assigns each objec a unique key, enabling ine-g ained access con ol, e sioning, and
me ada a agging. Wi h buil -in ea u es like li ecycle policies, c oss- egion eplica ion, and in eg a ion
wi h AWS analy ics and machine lea ning se ices, S3 is widely used o backup, disas e eco e y, da a
lakes, con en deli e y, and applica ion hos ing.
Se e al open-sou ce and comme cial implemen a ions o e S3-compa ible objec s o age o
o ganiza ions seeking al e na i es o AWS o o on-p emises deploymen s.
MinIO is a high-pe o mance, ligh weigh , and Kube ne es-na i e objec s o age sys em designed o
cloud-na i e applica ions. Fully compa ible wi h he S3 API, MinIO suppo s ad anced ea u es like
e asu e coding, enc yp ion, e sioning, and mul i- enan isola ion. I s simplici y, speed, and scalabili y
make i a popula choice o p i a e cloud, edge, and hyb id cloud en i onmen s.
Ceph RGW (RADOS Ga eway) is ano he p ominen S3-compa ible solu ion, pa o he b oade Ceph
dis ibu ed s o age sys em. Ceph RGW p o ides a ga eway in e ace ha allows use s o access Ceph's
objec s o age using S3 o Swi APIs. Ceph is known o i s high a ailabili y, aul ole ance, and uni ied
s o age capabili ies, o e ing block, objec , and ile s o age in a single pla o m. While mo e complex o
deploy han MinIO, Ceph RGW is a o ed in la ge-scale en e p ise o esea ch en i onmen s equi ing
deep cus omiza ion and in eg a ion wi h o he s o age modali ies.
OpenS ack Swi is an open-sou ce, dis ibu ed objec s o age sys em designed o s o e and e ie e
uns uc u ed da a a scale. I is a co e componen o he OpenS ack cloud pla o m and p o ides a highly
a ailable, aul - ole an s o age backend minimizing single poin o ailu e. Swi is ideal o s o ing la ge
olumes o da a such as backups, media iles, and a chi es. I suppo s ea u es like eplica ion, e en ual
consis ency, and me ada a, and can be accessed ia a REST ul API o compa ible clien s. Swi is
commonly used in p i a e clouds o se ice p o ide en i onmen s whe e open-sou ce con ol and
scalabili y a e key equi emen s.
OpenS ack Swi o e s S3 API suppo h ough middlewa e such as Swi 3 o adosgw (when used wi h
Ceph), enabling compa ibili y wi h S3-compa ible ools and applica ions. This allows use s o in e ac
wi h Swi using amilia S3 clien s while bene i ing om Swi 's scalable, open-sou ce a chi ec u e.
2. Cu en ooling o objec s o age
The objec s o age ooling consis s mainly o So wa e De elopmen Ki s (SDKs) o a ious
p og amming en i onmen s and applica ions p o iding CLIs and desk op GUIs. These ools in e ac wi h
he S3 di ec ly, he e o e i is a use esponsibili y o se up he common equi ed pa ame e s such as
URL o he endpoin , access and sec e keys, zone and egion name, e c.
D5.1 Tooling o objec s o age
9
Command line ools
s3cmd
Websi e: h ps://s3 ools.o g/s3cmd
A Py hon-based CLI ool o managing S3 bucke s and iles. I suppo s ope a ions like uploading,
downloading, syncing, and modi ying me ada a o Access Con ol Lis s (ACLs) o en o ce da a access
policies. I 's well-sui ed o sc ip ing and e minal-based wo k lows ha don' equi e p og amming.
s5cmd
Websi e: h ps://gi hub.com/peak/s5cmd
A high-pe o mance CLI ool w i en in Go, designed o as , pa allel S3 ope a ions such as ba ch
uploads, downloads, and sync. I is ideal o use s handling la ge-scale o ime-sensi i e S3 wo kloads,
o e ing signi ican speed ad an ages o e adi ional ools. I o e s supe io pe o mance o he na i e
s3cmd/bo o based implemen a ions.
So wa e De elopmen Ki s (SDKs)
Lib a ies wi h bindings o many p og amming languages exis s, he e we lis he mos popula ones
including Bo o3, which is pa o he o icial AWS SDK.
AWS SDK o Py hon (Bo o3)
Websi e: h ps://bo o3.amazonaws.com/ 1/documen a ion/api/la es /index.h ml
The o icial Py hon SDK o AWS se ices, including ull suppo o S3 ope a ions like uploads,
mul ipa ans e s, and bucke managemen .
AWS SDK o Go ( 2)
Websi e: h ps://aws.gi hub.io/aws-sdk-go- 2/
Mode n, modula SDK o Go ha suppo s S3 ope a ions and in eg a es easily wi h Go-based cloud
se ices o au oma ion ools.
MinIO Ja aSc ip Clien SDK
Websi e: h ps://min.io/docs/minio/linux/de elope s/ja asc ip /minio-js.h ml
Ligh weigh Node.js SDK o wo king wi h S3-compa ible s o age, ideal o use wi h MinIO bu wo ks
wi h any S3 endpoin .
G aphical Use In e aces (GUIs)
GUIs a e he mos use - iendly way o access he objec s o ages. The e is a numbe o desk op
applica ions allowing s eamlined in e ac ion wi h objec s and bucke s as well as web based in e aces,
such as he na i e MinIO web applica ion. He e a e wo examples o popula GUI applica ion wi h S3
suppo :
D5.1 Tooling o objec s o age
16
compliance wi h FAIR p inciples by o e ing access h ough MyAccessID and implemen ing he Da aCi e
me ada a model.
On a na ional le el, a coo dina ed na ional laye o esea ch-da a eposi o ies is being de eloped wi hin
he Eu opean Open Science Cloud amewo k unde EOSC CZ Ini ia i e
10
. The co e componen is he
Na ional Reposi o y Pla o m o Resea ch Da a
11
, which p o ides eposi o y in as uc u e, including a
gene al-pu pose na ional eposi o y, while also enabling in eg a ion wi h ins i u ional and domain-
speci ic eposi o ies. Me ada a om hese eposi o ies a e ha es ed in o he Na ional Me ada a
Di ec o y
12
, which se es as he cen al disco e y se ice. To ensu e in e ope abili y and alignmen wi h
EOSC, he me ada a a e exposed in s anda d o ma s such as DCAT-AP, making da ase s indable bo h
na ionally and in e na ionally. This pla o m will be connec ed o he LUMI AI Fac o y, he eby
acili a ing a ailabili y o la ge da ase s in he AI Fac o y ecosys em.
The access o la ge da ase s will be aligned wi h he goals o he EOSC CZ Ini ia i e, which ac i ely
implemen s he Open Science app oach and FAIR p inciples in he esea ch da a managemen domain
in he Czech Republic. The Czech na ional node o he Eu opean Open Science Cloud is cu en ly being
es ablished, aiming o building a Na ional Da a In as uc u e as a common pla o m o sha ing,
managing and accessing da a and compu ing esou ces o esea ch pu poses. EOSC CZ ini ia i e also
includes ac i i ies o suppo o he in eg a ion o a dis ibu ed la ge da a ans e se ice in o he Czech
EOSC ecosys em, which will allow he s aging o he da a close o he compu ing esou ces (such as GPU
enabled HPC clus e s).
4.1.2 Denma k
In DK he e a e se e al eposi o ies wi h accessibili y policies ha a e a ailable o being used by he LUMI
AI ac o y. The “Da a ej ise ” ac s as he main en y poin o publicly a ailable da ase s. The
speci ica ions a e de i ed om he Eu opean Commission's speci ica ion DCAT-AP 2.0, which aims o
s anda dize and es ablish cohe ence and in e ope abili y be ween da a po als in membe coun ies. The
Eu opean Commission's speci ica ion likewise builds upon he W3C speci ica ion DCAT (Da a Ca alog
Vocabula y).
These eposi o ies a e eachable ia he CKAN-API (h ps://docs.ckan.o g/en/2.9/api/)
(h ps://gi hub.com/ckan/ckanex -dca # d -dca -endpoin s) The CKAN’s Ac ion API is a, RPC-s yle API
ha exposes all o CKAN’s co e ea u es o API clien s. All o a CKAN websi e’s co e unc ionali y can be
used by ex e nal code.
4.1.3 Es onia
UT HPC has been ope a ing an objec s o age sys em h ps://s3.hpc.u .ee since 2024. I is based on open
sou ce objec s o age MinIO. Objec s o age became ele an o UT HPC as use s asked o a s o age
solu ion ha is S3 compa ible, om he e MinIO was chosen. S3 is cu en ly being used o HPC jobs, web
apps and as a s o age solu ion o a my iad o o e ed se ices. The u u e cu en ly holds a need o
10
EOSC CZ Ini ia i e: h ps://www.eosc.cz/en
11
Czech Na ional Reposi o y Pla o m o Resea ch Da a: h ps://da a.na odni- epozi a .cz
12
Czech Na ional Me ada a Di ec o y: h ps://nma.eosc.cz/
D5.1 Tooling o objec s o age
17
mig a e o e o a CEPH o SeaweedFS based solu ion, bu as long as e alua ions a e unde way no hing
is se .
4.1.4 Finland
CSC has been ope a ing Allas objec s o age sys em (h ps://docs.csc. i/da a/Allas/ ) based on CEPH
backend since 2019. The e o e CSC has expe ience on ope a ing such sys em be o e i was aken in o use
also in LUMI. The main pu pose o Allas is o s o e da a o p ojec li e ime, which does no need o be
ac i ely p ocessed on he HPC sys ems. CSC has sc ip s, which will dele e da a om Lus e sc a ch space
o he p ojec based on he ile imes amp and quo a o he sc a ch. CSC has been implemen ing w appe
sc ip s called "a-commands" o he clone wi h suppo on packaging and comp essing he iles. This is
help ul in cases whe e he e is a la ge amoun o iles. CSC has been suppo ing bo h Swi and S3
p o ocols, bu is now mig a ing o S3, which is he only p o ocol LUMI is using. CSC has p o ided guidance
how o use Allas om LUMI. A sc ip called allas-con suppo swi and S3 con igu ing he access. CSC
ad ices also g aphical use in e aces such as Cybe Duck and has made a simple web UI o he Allas
backend. O he al e na i e b owse based access is he OpenS ack Ho izon ( e y limi ed) and CSC
speci ic sensi i e da a ans e ool including enc yp ion. Fo back-up he e is a sc ip o he Res ic ool.
4.1.5 No way
Sigma2, No way’s na ional esea ch in as uc u e p o ide , has been ope a ing he Na ional e-
In as uc u e o Resea ch Da a (NIRD) since 2017. NIRD, now in i s second i e a ion and ope a ional
since he beginning o 2023, was designed wi h use - ocus, scalabili y, and adap abili y in mind. I se es
as a gene ic s o age in as uc u e comp ising wo independen bu in e connec ed esou ces: NIRD
Da a Peak and NIRD Da a Lake. NIRD enan s and use s ha e he op ion o apply o esou ces om
ei he NIRD Da a Peak o NIRD Da a Lake, o bo h, depending on hei speci ic unc ional equi emen s
and usage pa e ns. These esou ces p o ide cos -e icien da a and s o age se ices o use s.
NIRD is buil on IBM ha dwa e and so wa e, u ilizing IBM S o age Scale (GPFS) as i s co e. I o e s
unc ionali ies such as enan sepa a ion, da a li ecycle managemen , anspa en ie ing, au oma ic
eplica ion, snapsho s, s o age policies, enc yp ion, me ada a ca aloging and mo e.
NIRD Da a Lake p o ides uni ied ile- and objec -s o age access h ough edundan , high-a ailabili y
p o ocol nodes, and he objec s o age endpoin is h ps://s3.ni d.sigma2.no. The objec s o age se ice
u ilizes IBM’s High-Pe o mance Objec (HPO) s o age wi h NooBaa as i s co e. I suppo s he S3
p o ocol and S3 APIs and is in eg al o Sigma2’s da a-cen ic app oach, whe e da a is a he co e o
esea ch and se ices - such as cloud se ices (NIRD Se ice Pla o m) - a e buil a ound i and igh ly
in eg a ed o p o ide a seamless use expe ience. Cu en ly, Sigma2 is wo king on u he enhancing he
wo k low o No wegian HPC use s, and enabling da a s aging and caching be ween emo e HPC
sys ems and NIRD based on he S3 p o ocol.
In addi ion o NIRD’s gene ic s o age in as uc u e, Sigma2 coo dina es No way’s na ional se ices o
sensi i e esea ch da a. These se ices p o ide secu e, policy-complian en i onmen s o s o ing,
p ocessing, and sha ing da a classi ied unde GDPR and na ional heal h and p i acy egula ions. The
solu ions a e designed o suppo p ojec s ac oss heal h, social sciences, and o he domains whe e
sensi i e da a handling is manda o y. While physically and logically sepa a ed om he gene ic NIRD
D5.1 Tooling o objec s o age
18
se ices, he sensi i e da a pla o ms o m pa o No way’s b oade e-in as uc u e ecosys em and a e
being aligned wi h Eu opean ini ia i es o T us ed Resea ch En i onmen s (TREs).
4.1.6 Poland
AI s o age is p o ided as S3 se ice ia he PLG id po al. Two zones o S3 S o age a e a ailable,
geog aphically dis ibu ed o wo da acen e s, loca ed 10 km om each o he . Access o se ice is
a ailable by g an s accep ed ia PLG id po al. S3 is one o s o age a ailable o he use s, S3 is mainly
used o long e m s o age ha is no needed o compu a ional pu poses in nea u u e. The e a e
modules a ailable on supe compu e s o using s3 on compu ing clus e s ( o example clone o minio
clien ). We ha e se ice documen a ion a ailable o use s
(h ps://guide.s3p.cloud.cy one .pl/quicks a _en.h ml). Addi ionally, a WEB in e ace based on
iles ash is a ailable o use s o ge access o his ype o se ice. Use s can check ia a dedica ed po al
hei s a us o g an s, quo as and check numbe o objec s o da a amoun . This objec s o age is planned
o be in eg a ed wi h LUMI/LUMI AI o he AI se ice p o ision.
Objec s o ages ac oss Eu opean HPC cen e s
The Eu opean HPC cen e s also ope a e objec s o ages. Below we p o ide a non-exhaus i e lis o
se e al HPC cen e s, illus a ing ha S3 access is becoming qui e common, al hough i s local
implemen a ion ype a ies a lo .
• Jülich Suppo s S3 p o ocol. Access oken o he MinIO clus e is acqui ed ia web po al. MinIO
clien (mc), py hon bo o3 and s3cmd ools a e ad ised
13
• SURF objec s o age
14
is implemen ed wi h Ceph RADOS Ga eway and he au hen ica ion is ia
OpenS ack Keys one. Bo h S3 and Swi a e suppo ed. Use guide lis s s3cmd, awscli, S5cmd,
clone, py hon swi clien , cu l and Cybe duck (GUI).Also ead-only moun ing ia S3QL, S3FS,
clone and Goo ys is men ioned as back-up solu ion.
• MeluXina uses DDN S3DS sys em
15
, which suppo s S3 p o ocol. C eden ial a e ecei ed ia
helpdesk. Thei use guide has a able compa ing he ea u es o s3cmd, awscli, Cybe Duck and
py hon Bo o3.
5. Summa y
In his documen we p o ide a b ie desc ip ion o objec s o age as a concep o s o ing da a compa ed
o adi ional iles and olde s. We p o ide a comp ehensi e lis o exis ing clien applica ions and a ious
usage models. Impo an opic o his documen is use expe ience wi h objec s o age and e icien da a
mo emen close o he compu e esou ces when execu ing complex AI wo kloads. We desc ibe se e al
concep s which will be le e aged o p o ide ooling o he LUMI AIF use s, which a e expec ed o use
many e y la ge da ase s epea edly, he e o e app op ia e da a locali y mus be ensu ed h ough
anspa en caching a easonable amoun o abs ac ion.
13
Jülich objec s o age: h ps://apps. z-juelich.de/jsc/hps/jus /objec -s o age.h ml
14
SURF: h ps://se icedesk.su .nl/wiki/spaces/WIKI/pages/29557216/Objec +S o e
15
MeluXina: h ps://docs.lxp.lu/web_se ices/s3/
D5.1 Tooling o objec s o age
19
Signi ican da a mo emen is expec ed in LUMI AIF, be ween he AI ac o y i sel , local s o ages as well
as emo e da a esou ces and eposi o ies. The documen also con ains an o e iew o na ional s o age
esou ces ele an o AI use-cases ac oss he LUMI AIF conso ium pa ne s as well as s o ages ope a ed
by Eu opean HPC cen e s.
We p o ide desc ip ion o ooling sui able o common AI use-cases. As he e a e many echnologies and
p o ocols in ol ed in da a mo emen , he LUMI AIF should p o ide ailo ed suppo o i s use s by
p o iding app op ia e ooling and au oma ion o he da a ans e s by wo king closely wi h he use s and
ensu ing ha op imal da a access pa e ns a e used.
Re e ences
[1] NAKAMURA, Ryo; KUGA, Yohei. Mul i- h eaded scp: Easy and Fas File T ans e o e SSH.
In: P ac ice and Expe ience in Ad anced Resea ch Compu ing 2023: Compu ing o he Common Good.
2023. p. 320-323.
