A Grid Scheduling Algorithm Considering Dynamic Interconnecting Network

(1)

Centro Comunicaciones CSIC RedIRIS.

Spanish NREN

Distributed Systems Architecture and Security

Group Dpto. de Arquitectura de Computadores y Automática Universidad Complutense de Madrid !#" $ "

%& '& &

( )*+ ,-+ . + /0 1 2 ( ) -) 3 ) 4

A Grid Scheduling Algorithm

Considering

Dynamic Interconnecting

Network

The 3

rd

_{Cracow Grid WorkShop. October 2003}

Antonio Fuentes, Eduardo Huedo, Ruben S. Moreno, Ignacio Martín Llorente

Antonio Fuentes

(2)

OutLine

1. Motivation.

2. Applications and Framework.

3. Environment of the experiment.

4. Resources Selection Algorithm.

5. Example of Resource Selection.

6. Conclutions.

(3)

Motivation

1. Need use dynamic bandwidth. Why?

The network is not only for Grid

Computing.

2. Need Selection resources strategy

considering interconnecting Network.

1. Evaluation the transfer costs

;=< > ? @BA CED F G HJI C I < GLK I < A M K NI O H ? P A I < D G < QI F PSR T A G A U G D >I O ;=< G HJI < H I P M T I P N A F< D P I K D F V G H I WYX Z []\ ^_ ` O ;=< G HJI < N HJI P M T I P NA F< D P IK D F V G HJI a X _cb Z de O ;=< G HJI < HJI P T P A F< D P K D V _ ` b W b f _ \ X e O

(4)

Applications and FrameWork

GridWay FrameWork

gBh i j k lnm o pq rst u rv pq lnw i h m r x x uzy u r{ | m} m~ k i q i q m m hi m q m i o pq l l q w p k ~ h k l o

•

p o k s s | s ¡¢ r w i l p h l mo k q

•

p o k t y s ¢ s ¡¢ r l m~ m q h p k¤£ m l ph ~ k m~ h m

•

p o k r ¢¥ ¦ s ¡¢ r o m h¨§ o pq l p h l o m h j k ~ m o

•

p o k s ¢ u y s ¡¢ r © k l m h pq m i ª p« « k ~ p k i oq

(5)

Aplications and Framework

Network Weather Service

1. Distributed Resource Performance

Forecasting

Service for

Metacomputing.

2. Provide accurate forecasts of

dynamically

changing

performance characteristics.

Server

NS

PS

Server

¬® ¯E° ± ° ¯ ¬® ¯E° ± ° ¯

S

Client

F

Iperf

(6)

Environment of the experiment. IRISGrid

RedIRIS Local Network

Dacya Group

Host.Globus 2.4

²´³ µ¶ ·¸ ¹º »¼ º ²´³ µ¶ ·¸ ¹º »¼ º ½¿¾ ÀÁ Á½ Â ½¿¾ À Á ½Á Â Ã#Ä ÅÇÆ ÈÉÊ ÅË Ì ÍÇÎ ÏÐ Ñ#Ò Ë ÈÄÊ È Î Ó Ô#Õ Ö É Å Ö Ã#Ä ÅÇÆ ÈÉÊ ÅË Ì ÍÇÎ ÏÐ Ñ#Ò Ë ÈÄÊ È Î Ó Ô Õ Ö É Å Ö

GIIS

IRISGrid

GRIIS

1Gbit/s

2Gbit/s

The Spanish Thematic Network

on Grids in the framework of

e-Science Initiatives

S

NS M

NS

S

M

NWS NameServer

NWS Memory

NWS Sensor

(7)

Resources Selection Algorithm

×ÙØ ÚÛ Ü ÝÞ Ø ßáà â àã ä åÇæ ç ßáà â àã ä åÇæ ç èéê ë èé ê ë èéê ê ë èéê ê ë èéê ê ë èéê ê ë ìíîï ðñ òó ô òõ é÷ö ø ù é÷ö ø ù úû ü úý ü û þ ÿ þ û ö ø ö ø é÷ö ø ù !#" $å&% ' ( ä ã ) * ( ç (+ à, × ØÚÛ Ü ÝÞ Ø $å&% ã æ- à, . $å&% ã æ- à, . /ô01 òó ò 2 ì íîï õ ò3 ó4 5 67#8 7 9 Þ :<; = Û Ý> 8 7 9 Û ; ? 6A@ B @C 8 ÝÞ D97 Ø Þ 7 Ü ÝØ E

•

F % à, G ' , æ-åH à H I àJ K å , à Làç ä % F % à, G ' , æ-åH à H I à J K å , à Làç ä %

•

M Ü 7 D Û Ý 9ON 8 7 9 Û ; 7 ØÚ 7 POQ ;8 > 9 Þ :<; = Û Ý> 8 7 9 Û ; ?RS TU Û 8 ; V C W E

•

I ( ç X àY ' , à % % åÇæ ç I ( ç X àY ' , à % % å æç

•

T Ú Ø ÝÛ Z Ý [ 9 V Ø V Ø\ Ø Þ Ü 7#8 ] U Ø

•

R D 8 8 Ý Þ 7 Ø Ý 9 N Ø V9 Ú Û Þ [Ø ÝØ V D Û Ú 7 Ú @ ^ Ø 7 8 Û Ý V Ø ÝØ VU 9 Ú 7 Û = ÝØÚÛ Ü ÝÞ ØÚ ]OQ Ý 8 ; _ è ` a ø b øcö ø è ` a ø b øcö ø d e f bg ë b f e a d ë d e f bg ë b f e a d ë èéê ê ë èé ê ê ë

Network Measurements

Rank

Calculation

(8)

Experimental TestBed. RedIRIS-UCM

hji k l m i npo q r&s o o n t r mjo uiv w x i uy z|{

aquila

Pentim III

700Mhz Linux 2.4

128M

dacya.ucm.es

cygnus

Pentium IV

2.5Ghz Linux 2.4

512M

dacya.ucm.es

cepheus

Pentium III

600Mhz Linux 2.4

256M

dacya.ucm.es

hydrus

Pentium IV

2.5Ghz Linux 2.4

512M

dacya.ucm.es

aristoteles Pentium III

1.4Ghz Linux 2.4

1G

rediris.es

platon

Pentium III

1.4Ghz Linux 2.4

1G

rediris.es

heraclito

Celeron

700Mhz Linux 2.4

256M

rediris.es

}~

CPU intensive artificial WorkLoad.

Initially, the job submision from cygnus

O ¡ ¢ £ O¤ ¥ ¦ § ¥ ¨ ¡ © ª¨ © ¥ « ¡ ¢ « ¦ £¬ ¡ ¢ £ O¤ ¥ ¦ § ¥ ¨ ¡ © ª¨ © ¥ « ¡ ¢ « ¦ £¬ ¯® °#± ²³µ´ ¶ ·¹¸ º» ¼½ ¾ ½ ¿ÀµÁ Â Ã¹Ä Å Æ¯Ç È#É ÊËµÌ Í Î¹Ï Ð Ñ Ò&Ó Ô ÕÖ ×Ø Ó Ò ÙÚ Ø ÒÛ ÝÞµß à á¹â ãåä æç èé ê ë ìí î ï ìð è

(9)

Results

Example. 4 Jobs Running.

ñóò

ô#õ ö÷ùø ú ûýü þÿ

JID AID TID DM SM GSM STIME ETIME CPUTIME XFRTIME EXIT TEMPLATE HOST

0 0 0 zomb done -- 13:55:13 13:55:47 00:12 00:22 0 job_template cygnus.dacya.ucm.es 1 0 1 zomb done -- 13:55:13 13:55:48 00:12 00:23 0 job_template hydrus.dacya.ucm.es 2 0 2 zomb done -- 13:55:13 13:56:02 00:23 00:26 0 job_template aquila.dacya.ucm.es 3 0 3 zomb done -- 13:55:13 13:56:03 00:24 00:26 0 job_template

90seg

cepheus.dacya.ucm.es

JID AID TID DM SM GSM STIME ETIME CPUTIME XFRTIME EXIT TEMPLATE

HOST

0 0 0 zomb done -- 13:56:49 13:57:23 00:12 00:22 0 job_template cygnus.dacya.ucm.es 1 0 1 zomb done -- 13:56:49 13:57:23 00:12 00:22 0 job_template hydrus.dacya.ucm.es 2 0 2 zomb done -- 13:56:49 13:58:36 00:22 01:25 0 job_template

187seg

aristoteles.rediris.es 3 0 3 zomb done -- 13:56:49 13:58:25 00:22 01:14 0 job_template platon.rediris.es

! " # " $ " % " & " ' " ( " )" *" + " # "" # #" ,-. /10 2 . 3 04 5 -617 8 -4 9 : ;4 6 ,-6 . / <=> ?@ABCD E B FF G HI JK L M N OPI Q M RS T RI Q U N V W I Q X Y Z X ZY [ X [Y \ X \Y ] X ]Y Y X ^`_ a b c d1e f c a g h ia dkj c d l m n op qrstuv w t xx

(10)

Conclusion and Future Work

1. Extension experiment at all IRISGrid Resources.

2. Integration of Migrations Jobs.

3. Integration of Dynamic Reservation of bandwidth.

Traffic Prioritiy.

(11)

¾ ¾