Sa ing ene gy in G id Compu ing.
A. Fe nández-Mon es1, J. I. Sánchez-Venzalá1,J. A. O ega1, L. González-Ab il2
1Depa men o Compu e Science, Uni e si y o Se illa, Se illa, Spain
{a dez,jisanchez,jo ega}@us.es
2Depa men o Applied Economics, Uni e si y o Se illa, Se illa, Spain
[email p o ec ed]
Abs ac
This a icle is ocused on simula ing he p og ess o
G id’5000 in o de o choose bes policies in e ms
o a anging jobs and managing esou ces. A Ja a-
based simula o has been de eloped in o de o e-
play he condi ions o he G id’5000 om ecen
yea s his o ical da a . This way, an s udy o di -
e en policies has been ca ied ou looking o en-
e gy e iciency. The policies s udied a e based in
ma hema ical models which y o p edic he mos
e icien beha io o he G id’5000.
1 In oduc ion
Sa ing ene gy is a key ac o in compu e science. Ene gy e -
iciency is looked o in all kind o sys ems om li le de ices
o la ge scale compu ing.
The huge amoun o ene gy consumed by g id compu ing
is a good eason o s udy sa ing ene gy me hodologies ei he
om an economical o ecological poin o iew. G id ope -
a ional policies mus be ma hema ically analyzed in o de o
be op imized.
The analysis ha his pape p esen s has been accom-
plished o e ench G id’5000, desc ibed nex .
2 G id’5000 O ganiza ion
G id’5000 is a scien i ic ins umen designed o suppo
expe imen -d i en esea ch in all a eas o compu e science
ela ed o pa allel, la ge-scale o dis ibu ed compu ing and
ne wo king. I aims o supply a highly econ igu able, con-
olable and moni o able expe imen al pla o m o i s use s.
The G id’5000 p o ides a es bed which allows expe imen s
in all he so wa e laye s be ween he ne wo k p o ocols up o
he applica ions.
G id’5000 has been buil upon a ne wo k o dedica ed clus-
e s. I is no an ad hoc g id. The in as uc u e o G id’5000
is geog aphically dis ibu ed on di e en si es, ini ially 9 in
F ance: Bou deaux, G enoble, Lille, Lyon, Nancy, O say,
Rennes, Sophia-An ipolis and Toulouse. Po o Aleg e, in
B azil, and Luxembu g, a e now o icially becoming he 10 h
and 11 h si es espec i ely.
The p ojec began in 2004 as an ini ia i e o ench min-
is y o Educa ion and Resea ch, INRIA, CNRS, he Uni e -
si ies o all si es and some egional councils.
Figu e 1: G id’5000 ench si es.
The ini ial aim was o each 5000 p ocesso s in he pla -
o m. I has been e amed a 5000 co es, and was eached
du ing win e 2008-2009. On Ma ch 16 h 2010, 1569 nodes
(5808 co es) we e in p oduc ion in G id’5000.
Nowadays, si es see each o he s inside he same VLAN a
10Gbps hanks o he da k ibe in as uc u e which connec s
hem, in a no comple e g aph scheme.
G id’5000 allows expe imen s a g id o a clus e le el,
which gua an ees a mo e homogeneous ha dwa e and band-
wid h, al hough g id le el expe imen s a e a o ed in plan-
ning.
Each si e o G id’5000 hos s se e al clus e s, because ha d-
wa e has been acqui ed by inc emen al s eps on each si e,
o ming clus e s a each pu chase.
Each clus e is o med by wo kind o nodes:
• Compu e node, which con o ms he base elemen o a
clus e , on which compu a ions a e un.
• Se ice node, which a e dedica ed o hos he g id in-
as uc u e se ices, as con ol o deploy.
Each node can supply se e al co es, which a e he ines
g ain o esou ce in G id’5000.
2.1 Tasks
The pla o m can be used in wo di e en modes: submis-
sions and ese a ions.
• Submission: an expe imen is submi ed and he sched-
ule decides when o un i .
• Rese a ion: when a ese a ion o he pla o m o a
ce ain ime is made (al hough he expe imen has o be
launched in e ac i ely).
The so wa e used o ask schedule is OAR. I is a esou ce
manage (o ba ch schedule ) o la ge clus e s which allows
clus e use s o submi o ese e nodes ei he in an in e ac i e
o in a ba ch mode.
3 G id’5000 Simula o
The G id’5000 simula o ies o simula e he p og ess o he
eal G id ega ding jobs and esou ces ope a ion. The ob-
jec i e is o be able o compu e he ene gy consumed by
G id’5000 om his o ical da a om pas yea s, which is
s o ed in a da abase, applying di e en policies o a ang-
ing jobs and managing esou ces. A anging jobs policies a e
called A anging Policies while managing esou ce policies
a e called Ene gy Policies.
The simula o ope a ion is based on an agenda whe e jobs
a e egis e ed and a lis o esou ces ep esen ing he eal e-
sou ces om he si es.
The simula o s a s o co e he agenda om he beginning
o he end, modi ying esou ces s a es as would be needed
o execu e hem in he eal wo ld, aking in o accoun he
policies es ablished o manage esou ces and jobs. The con-
sumed ene gy compu a ion is made s ep by s ep by means o
he in o ma ion abou ene gy consump ion o each esou ce
and esou ce s a es poin ed ou in he esou ce lis .
The esul o simula ion execu ion is a log whe e he be-
ha io o g id, esou ces, and asks acco ding o he policies
employed a e shown oge he wi h he ene gy consumed com-
pu a ion.
I has been implemen ed in Ja a, which makes possible an
easy in eg a ion o new componen s, as he g aphical in e -
ace, o he de elopmen o new ex ensions by o he s.
4 Ene gy Policies
Ene gy policies es ablish he managing o he g id esou ces.
They desc ibe wha o do wi h a esou ce when a ask inishes
i s execu ion. The e a e se e al op ions:
• Always On: i lea es esou ces always on, ne e swi ch
hem o .
• Always Swi ch O : i always swi ch esou ces o a e
a jobs execu ion.
• Swi ch O in Ts: a e a jobs execu ion, i wai s o a
de e mined ime (Ts) o swi ch-o he esou ce.
O he ene gy policies a e being s udied and simula ed in
o de o op imize ene gy sa ing in he g id.
5 A anging Policies
A anging policies es ablish he a anging o he jobs o i s
execu ion. They can mo e a job om one esou ce o ano he ,
o e en can mo e a planned job execu ion in ime in o de o
aking ad an ages o esou ces ha a e al eady swi ched on.
• Do No hing: does no mo e jobs nei he in ime o om
a esou ce o ano he , hey a e execu ed as hey we e
de ined in he agenda.
• Simple Agg ega ion o Tasks: which ies o execu e he
jobs in he same esou ces, i possible, al hough i does
no change planned jobs s a ime.
O he a anging policies a e being conside ed, in o de o
op imize he execu ion o asks.
90