ARGONNE ! CHICAGO foste r@m cs.a nl.gov Grid Com puting Ia n Foste - - PowerPoint PPT Presentation

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

G

Grid Co m pu tin g : Re s o u rce S h a rin g a n d Co o rdin a te d Pro ble m S o lv in g in S ca la ble Dis tribu te d Co m m u n itie s

G

Grid com puting te chnologie s e na ble controlle d re source sha ring in distribute d com m unitie s a nd the coordina te d use of those sha re d re source s a s com m unity m e m be rs ta ckle com m on goa ls. The se te chnologie s include ne w protocols, se rvice s, a nd APIs for se cure re source a cce ss, re source m a na ge m e nt, fa ult de te ction, com m unica tion, a nd so forth, tha t in te rm e na ble ne w a pplica tion conce pts such a s virtua l da ta , sm a rt instrum e nts, colla bora tive de sign spa ce s, a nd m e ta com puta tions. In this ta lk, we re vie w a pplica tions tha t a re m otiva ting wide spre a d inte re st in Grid conce pts within the scie ntific a nd e ngine e ring com m unitie s. The n, we de scribe the Globus Grid a rchite cture tha t ha s be e n a dopte d by m a ny Grid proje cts, focusing in pa rticula r on our se curity, re source m a na ge m e nt, a nd da ta m a na ge m e nt te chnologie s.

Grid Com puting

Ia n Foste r

Ma the m a tics a nd Com pute r Scie nce Division Argonne Na tiona l La bora tory De pa rtm e nt of Com pute r Scie nce The Unive rsity of Chica go

Ca rl Ke sse lm a n

Inform a tion Scie nce s Institute Unive rsity of Southe rn Ca lifornia

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

Ove rvie w

G

Grid com puting conce pt

G

Grid te chnica l la ndsca pe

G

Da ta grids & da ta grid proje cts

The Grid Conce pt

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

The Grid: The We b on Ste roids

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

The Grid: The We b on Ste roids

http: / / http: / /

We b: Uniform a cce ss to HTML docum e nts

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

The Grid: The We b on Ste roids

http: / / http: / /

We b: Uniform a cce ss to HTML docum e nts Grid: Fle xible , high-pe rf a cce ss to a ll significa nt re source s

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

The Grid: The We b on Ste roids

http: / / http: / /

We b: Uniform a cce ss to HTML docum e nts Grid: Fle xible , high-pe rf a cce ss to a ll significa nt re source s

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

The Grid: The We b on Ste roids

http: / / http: / /

We b: Uniform a cce ss to HTML docum e nts Grid: Fle xible , high-pe rf a cce ss to a ll significa nt re source s

Se nsor ne ts Da ta a rchive s Com pute rs Softwa re ca ta logs Colle a gue s

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

The Grid: The We b on Ste roids

http: / / http: / /

We b: Uniform a cce ss to HTML docum e nts Grid: Fle xible , high-pe rf a cce ss to a ll significa nt re source s

Se nsor ne ts Da ta a rchive s Com pute rs Softwa re ca ta logs Colle a gue s

On-de m and cre ation of pow e rful virtual com puting syste m s

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

Grid Com puting: Ta ke 2

G Ena ble com m unitie s (“virtua l orga niza tions”)

to sha re ge ogra phica lly distribute d re source s a s the y pursue com m on goa ls—in the a bse nce of ce ntra l control, om niscie nce , trust re la tionships

G Via inve stiga tions of

– Ne w a pplica tions tha t be com e possible whe n re source s ca n be sha re d in a coordina te d wa y – Protocols, a lgorithm s, pe rsiste nt infra structure to fa cilita te sha ring

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

Grid Com m unitie s a nd Applica tions: NSF Na tiona l Te chnology Grid

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

Grid Com m unitie s a nd Applica tions: Ne twork for Ea rthqua ke Eng. Sim ula tion

G NEESgrid: na tiona l

infra structure to couple e a rthqua ke e ngine e rs with e xpe rim e nta l fa cilitie s, da ta ba se s, com pute rs, & e a ch othe r

G On-de m a nd a cce ss to

e xpe rim e nts, da ta stre a m s, com puting, a rchive s, colla bora tion

NEESgrid: Argonne , Michiga n, NCSA, UIUC, USC

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

Grid Com m unitie s a nd Applica tions: Ma the m a ticia ns Solve NUG30

G Com m unity= a n inform a l

colla bora tion of m a the m a ticia ns a nd com pute r scie ntists

G Condor-G de live rs 3.46E8

CPU se conds in 7 da ys (pe a k 1009 proce ssors) in U.S. a nd Ita ly (8 site s)

G Solve s NUG30 qua dra tic

a ssignm e nt proble m

14,5,28,24,1,3,16,15, 10,9,21,2,4,29,25,22, 13,26,17,30,6,20,19, 8,18,7,27,12,11,23

Me ta NEOS: Argonne , Iowa , Northwe ste rn, Wisconsin

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

G Com m unity =

– 1000s of hom e com pute r use rs – Phila nthropic com puting ve ndor (Entropia ) – Re se a rch group (Scripps)

G Com m on goa l=

a dva nce AIDS re se a rch

Grid Com m unitie s a nd Applica tions: Hom e Com pute rs Eva lua te AIDS Drugs

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

A Little History

G Ea rly 90s

– Giga bit te stbe ds, m e ta com puting

G Mid to la te 90s

– Ea rly e xpe rim e nts (e .g., I-WAY), a ca de m ic softwa re proje cts (e .g., Globus), a pplica tion e xpe rim e nts

G 2000

– Ma jor a pplica tion com m unitie s e m e rging – Ma jor infra structure de ploym e nts – Rich te chnology ba se – Grid Forum : > 300 pe ople , > 90 orgs, 11 countrie s

Te chnica l La ndsca pe

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

Grid Te chnica l Cha lle nge s: Sha ring, Coordina te d Use

G Ne w proble m solving m e thods a nd tools

– Da ta , com puting, colla bora tion, se nsors, …

G Num e rous cross-cutting te chnica l proble m s

– Authe ntica tion, a uthoriza tion, policy, a udit; re source discove ry, a cce ss, a lloca tion, control; fa ilure de te ction & re cove ry; broke ring; … ; …

G No ce ntra l control, om niscie nce , trust; he nce

– Ne e d to pre se rve loca l site inde pe nde nce – Policy discove ry a nd ne gotia tion im porta nt – Myria d inte re sting fa ilure m ode s

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

How Are Cross-Cutting Te chnica l Issue s Addre sse d?

G De ve lopm e nt of Grid protocols & se rvice s

– Protocol-m e dia te d a cce ss to re m ote re source s – Ne w se rvice s: e .g., re source broke ring – “On the Grid” = spe a k Inte rgrid protocols – Mostly (e xte nsions to) e xisting protocols

G De ve lopm e nt of Grid APIs & SDKs

– Fa cilita te a pplica tion de ve lopm e nt by supplying highe r-le ve l a bstra ctions

G The (huge ly succe ssful) m ode l is the Inte rne t G The Grid is not a distribute d OS!

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

La ye re d Grid Archite cture (By Ana logy to Inte rne t Archite cture )

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

La ye re d Grid Archite cture (By Ana logy to Inte rne t Archite cture )

Fa bric

“Controlling things loca lly”: Acce ss to, & control of, re source s

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

La ye re d Grid Archite cture (By Ana logy to Inte rne t Archite cture )

Fa bric

“Controlling things loca lly”: Acce ss to, & control of, re source s

Conne ctivity

“Ta lking to things”: com m unica tion (Inte rne t protocols) & se curity

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

La ye re d Grid Archite cture (By Ana logy to Inte rne t Archite cture )

Fa bric

“Controlling things loca lly”: Acce ss to, & control of, re source s

Conne ctivity

“Ta lking to things”: com m unica tion (Inte rne t protocols) & se curity

Re source

“Sha ring single re source s”: ne gotia ting a cce ss, controlling use

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

La ye re d Grid Archite cture (By Ana logy to Inte rne t Archite cture )

Fa bric

“Controlling things loca lly”: Acce ss to, & control of, re source s

Conne ctivity

“Ta lking to things”: com m unica tion (Inte rne t protocols) & se curity

Re source

“Sha ring single re source s”: ne gotia ting a cce ss, controlling use

Colle ctive

“Ma na ging m ultiple re source s”: ubiquitous infra structure se rvice s

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

La ye re d Grid Archite cture (By Ana logy to Inte rne t Archite cture )

Fa bric

“Controlling things loca lly”: Acce ss to, & control of, re source s

Conne ctivity

“Ta lking to things”: com m unica tion (Inte rne t protocols) & se curity

Re source

“Sha ring single re source s”: ne gotia ting a cce ss, controlling use

Colle ctive

“Ma na ging m ultiple re source s”: ubiquitous infra structure se rvice s

Use r

“Spe cia lize d se rvice s”: use r- or a ppln-spe cific distribute d se rvice s

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

La ye re d Grid Archite cture (By Ana logy to Inte rne t Archite cture )

Fa bric

“Controlling things loca lly”: Acce ss to, & control of, re source s

Conne ctivity

“Ta lking to things”: com m unica tion (Inte rne t protocols) & se curity

Re source

“Sha ring single re source s”: ne gotia ting a cce ss, controlling use

Colle ctive

“Ma na ging m ultiple re source s”: ubiquitous infra structure se rvice s

Use r

“Spe cia lize d se rvice s”: use r- or a ppln-spe cific distribute d se rvice s

Applica tion

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

La ye re d Grid Archite cture (By Ana logy to Inte rne t Archite cture )

Fa bric

“Controlling things loca lly”: Acce ss to, & control of, re source s

Conne ctivity

“Ta lking to things”: com m unica tion (Inte rne t protocols) & se curity

Re source

“Sha ring single re source s”: ne gotia ting a cce ss, controlling use

Colle ctive

“Ma na ging m ultiple re source s”: ubiquitous infra structure se rvice s

Use r

“Spe cia lize d se rvice s”: use r- or a ppln-spe cific distribute d se rvice s

Applica tion

Inte rne t Tra nsport Applica tion Link Inte rne t Protocol Archite cture

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

Protocols, Se rvice s, a nd Inte rfa ce s Occur a t Ea ch Le ve l

La ngua ge s/ Fra m e works Fa bric La ye r Applica tions

Loca l Acce ss APIs a nd Protocols

Colle ctive Se rvice APIs a nd SDKs Colle ctive Se rvice s

Colle ctive Se rvice Protocols

Re source APIs a nd SDKs Re source Se rvice s

Re source Se rvice Protocols Use r Se rvice Protocols

Use r Se rvice APIs a nd SDKs Use r Se rvice s Conne ctivity APIs

Conne ctivity Protocols

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

Grid Se rvice s Archite cture : Fa bric La ye r Protocols & Se rvice s

G Just wha t you would e xpe ct: the dive rse

m ix of re source s tha t m a y be sha re d

– Individua l com pute rs, Condor pools, file syste m s, a rchive s, m e ta da ta ca ta logs, ne tworks, se nsors, e tc., e tc.

G Fe w constra ints on low-le ve l te chnology:

conne ctivity a nd re source le ve l protocols form the “ne ck in the hourgla ss”

G De fine d by inte rfa ce s not physica l

cha ra cte ristics

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

Grid Se rvice s Archite cture : Conne ctivity La ye r Protocols & Se rvice s

G Com m unica tion

– Inte rne t protocols: IP, DNS, routing, e tc.

G Se curity: Grid Se curity Infra structure (GSI)

– Uniform a uthe ntica tion & a uthoriza tion m e cha nism s in m ulti-institutiona l se tting – Single sign-on, de le ga tion, ide ntity m a pping – Public ke y te chnology, SSL, X.509, GSS-API – Supporting infra structure : Ce rtifica te Authoritie s, ke y m a na ge m e nt, e tc.

GSI: www.globus.org

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

Grid Se rvice s Archite cture : Re source La ye r Protocols & Se rvice s

G Grid Re source Alloca tion Mgm t (GRAM)

– Re m ote a lloca tion, re se rva tion, m onitoring, control of com pute re source s

G GridFTP protocol (FTP e xte nsions)

– High-pe rform a nce da ta a cce ss & tra nsport

G Grid Re source Inform a tion Se rvice (GRIS)

– Acce ss to structure & sta te inform a tion

G Ne twork re se rva tion, m onitoring, control G All inte gra te d with GSI: a uthe ntica tion,

a uthoriza tion, policy, de le ga tion

GRAM, GridFTP, GRIS: www.globus.org

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

Grid Se rvice s Archite cture : Colle ctive La ye r Protocols & Se rvice s

G Inde x se rve rs a ka m e ta dire ctory se rvice s

– Custom vie ws on dyna m ic re source colle ctions a sse m ble d by a com m unity

G Re source broke rs (e .g., Condor Ma tchm a ke r)

– Re source discove ry a nd a lloca tion

G Re plica ca ta logs G Co-re se rva tion a nd co-a lloca tion se rvice s G Etc., e tc.

Me ta dire ctory: www.globus.org; Condor: www.cs.wisc.e du/ condor

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

S o u rce Co de Re po s ito ry S DK API Lo o ku p Pro to co l

Exa m ple : Use r Porta l

We b Porta l Source code discove ry, a pplica tion configura tion Broke ring, co-a lloca tion, ce rtifica te a uthoritie s Acce ss to da ta , a cce ss to com pute rs, a cce ss to ne twork pe rform a nce da ta , … Com m unica tion, se rvice discove ry (DNS), a uthe ntica tion, a uthoriza tion, de le ga tion Stora ge syste m s, cluste rs, ne tworks, … Us e r Appln Co lle ctive Re s o u rce Co n n e ct Fa bric

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

S o u rce Co de Re po s ito ry S DK API Lo o ku p Pro to co l

Exa m ple : Use r Porta l

We b Porta l Source code discove ry, a pplica tion configura tion Broke ring, co-a lloca tion, ce rtifica te a uthoritie s Acce ss to da ta , a cce ss to com pute rs, a cce ss to ne twork pe rform a nce da ta , … Com m unica tion, se rvice discove ry (DNS), a uthe ntica tion, a uthoriza tion, de le ga tion Stora ge syste m s, cluste rs, ne tworks, … Us e r Appln Co lle ctive Re s o u rce Co n n e ct Fa bric

Co m pu te Re s o u rce S DK API Acce s s Pro to co l

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

Ch e ckpo in t Re po s ito ry S DK API C-po in t Pro to co l

Exa m ple : High-Throughput Com puting Syste m

High Throughput Com puting Syste m Dyna m ic che ckpoint, job m a na ge m e nt, fa ilove r, sta ging Broke ring, ce rtifica te a uthoritie s Acce ss to da ta , a cce ss to com pute rs, a cce ss to ne twork pe rform a nce da ta Com m unica tion, se rvice discove ry (DNS), a uthe ntica tion, a uthoriza tion, de le ga tion Stora ge syste m s, sche dule rs Us e r Appln Co lle ctive Re s o u rce Co n n e ct Fa bric

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

Ch e ckpo in t Re po s ito ry S DK API C-po in t Pro to co l

Exa m ple : High-Throughput Com puting Syste m

High Throughput Com puting Syste m Dyna m ic che ckpoint, job m a na ge m e nt, fa ilove r, sta ging Broke ring, ce rtifica te a uthoritie s Acce ss to da ta , a cce ss to com pute rs, a cce ss to ne twork pe rform a nce da ta Com m unica tion, se rvice discove ry (DNS), a uthe ntica tion, a uthoriza tion, de le ga tion Stora ge syste m s, sche dule rs Us e r Appln Co lle ctive Re s o u rce Co n n e ct Fa bric

Co m pu te Re s o u rce S DK API Acce s s Pro to co l

Da ta Grids

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

Da ta Grid Proble m

G “Ena ble a ge ogra phica lly distribute d

com m unity [of thousa nds] to pool the ir re source s in orde r to pe rform sophistica te d, com puta tiona lly inte nsive a na lyse s on Pe ta byte s of da ta ”

G Note tha t this proble m

– Is not unique to physics – Ove rla ps strongly with othe r Grid proble m s

G Da ta Grids do introduce ne w re quire m e nts

a nd R&D cha lle nge s

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

Da ta Grid Hie ra rchy Conce pt

Tier2 Centre ~1 TIPS Online System Offline Processor Farm ~20 TIPS CERN Computer Centre FermiLab ~4 TIPS France Regional Centre Italy Regional Centre Germany Regional Centre Institute Institute Institute Institute ~0.25TIPS Physicist workstations ~100 MBytes/sec ~100 MBytes/sec ~622 Mbits/sec ~1 MBytes/sec

There is a “ bunch crossing” every 25 nsecs. There are 100 “ t riggers” per second Each t riggered event is ~1 MByt e in size Physicist s work on analysis “ channels” . Each inst it ut e will have ~10 physicist s working on one or more channels; dat a f or t hese channels should be cached by t he inst it ut e server

Physics data cache

~PBytes/sec

~622 Mbits/sec

• r Air Freight (deprecated)

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

Ma jor Da ta Grid Proje cts

G Ea rth Syste m Grid (DOE Office of Scie nce )

– DG te chnologie s, clim a te a pplica tions

G Europe a n Da ta Grid (EU)

– DG te chnologie s & de ploym e nt in EU

G GriPhyN (NSF ITR)

– Inve stiga tion of “Virtua l Da ta ” conce pt

G Pa rticle Physics Da ta Grid (DOE Scie nce )

– DG a pplica tions for HENP e xpe rim e nts

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

Da ta Grid Archite cture

Discipline -Spe cific Da ta Grid Applica tion Cohe re ncy control, re plica se le ction, ta sk m a na ge m e nt, virtua l da ta ca ta log, virtua l da ta code ca ta log, … Re plica ca ta log, re plica m a na ge m e nt, co-a lloca tion, ce rtifica te a uthoritie s, m e ta da ta ca ta logs, Acce ss to da ta , a cce ss to com pute rs, a cce ss to ne twork pe rform a nce da ta , … Com m unica tion, se rvice discove ry (DNS), a uthe ntica tion, a uthoriza tion, de le ga tion Stora ge syste m s, cluste rs, ne tworks, ne twork ca che s, … Us e r Appln Co lle ctive Re s o u rce Co n n e ct Fa bric

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

Globus Da ta -Inte nsive Se rvice s Archite cture

Libra ry Progra m

Le ge nd

globus-url-copy Re plica Progra m s Custom Se rve rs globus_ga ss_copy globus_ftp_clie nt globus_ftp_control globus_com m on GSI (se curity) globus_io Ope nLDAP clie nt globus_re plica _ca ta log globus_re plica _m a na ge r Custom Clie nts globus_ga ss_tra nsfe r globus_ga ss

Alre a dy e xist

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

High-Le ve l Vie w of Ea rth Syste m Grid: A Mode l Archite cture for Da ta Grids

Me ta da ta Ca ta log Re plica Ca ta log

Ta pe Libra ry Dis k Ca ch e

Attribu te S pe cifica tio n Lo g ica l Co lle ctio n a n d Lo g ica l File Na m e

Dis k Arra y Dis k Ca ch e

Applica tion

Re plica Se le ction

Mu ltiple Lo ca tio n s

NWS

S e le cte d Re plica GridFTP co m m a n ds Pe rfo rm a n ce I n fo rm a tio n & Pre dictio n s Re plica Lo ca tio n 1 Re plica Lo ca tio n 2 Re plica Lo ca tio n 3

MDS

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

Grid Physics Ne twork (GriPhyN)

Ena bling R&D for a dva nce d da ta grid syste m s, focusing in pa rticula r on Virtua l Da ta conce pt

Virtual Data Tools Request Planning and Scheduling Tools Request Execution Management Tools Transforms Distributed resources (code, storage, computers, and network) Resource Management Services Resource Management Services Security and Policy Services Security and Policy Services Other Grid Services Other Grid Services Interactive User Tools Production Team Individual Investigator Other Users Raw data source

ATLAS CMS LIGO SDSS

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

The Virtua l Da ta Conce pt

“[a virtua l da ta grid e na ble s] the de finition a nd de live ry of a pote ntia lly unlim ite d virtua l spa ce of da ta products de rive d from

• the r da ta . In this virtua l spa ce , re que sts

ca n be sa tisfie d via dire ct re trie va l of m a te ria lize d products a nd/ or com puta tion, with loca l a nd globa l re source m a na ge m e nt, policy, a nd se curity constra ints de te rm ining the stra te gy use d.”

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

Virtua l Da ta in Action

?

Major Archive Facilities Network caches & regional centers Local sites G Da ta re que st m a y

G Acce ss loca l da ta G Com pute loca lly G Com pute re m ote ly G Acce ss re m ote

da ta

G Sche duling subje ct

to loca l & globa l policie s

G Loca l a utonom y

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

Abbre via te d Acknowle dgm e nts

G Othe r Globus principa ls include : Ste ve

Tue cke , Ann Che rve na k, Bill Allcock, Gre gor von La sze wski, Le e Lim ing, Ste ve Fitzge ra ld

G Da ta Grid colla bora tors: Arie Shosha ni,

Re a ga n Moore , Miron Livny

G GriPhyN Co-PI: Pa ul Ave ry G PPDG PIs: Richa rd Mount, Ha rve y Ne wm a n

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

Sum m a ry

G Grid com puting conce pts a re re a l: the y

unde rlie a va rie ty of m a jor proje cts in physics, othe r scie nce s, a nd industry

G Prom ise is tha t by a llowing coordina ting

sha ring, e na ble qua ntita tive & qua lita tive cha nge s in proble m solving te chnique s

G Cle a r picture of Grid a rchite cture e m e rging G Da ta Grid conce pts a nd te chnologie s a re

le ss m a ture but a re a ppe a ring ra pidly

foste r@m cs.a nl.gov ARGONNE ! CHICAGO

For More Inform a tion

G Book (Morga n Ka ufm a n)

– www.m kp.com / grids

G Globus

– www.globus.org

G Grid Forum

– www.gridforum .org

G PPDG

– www.ppdg.ne t

G GriPhyN

– www.griphyn.org

The End