• Aucun résultat trouvé

Комплексная система управления данными и задачами в гетерогенной компьютерной среде (Integrated Workload and Data Management System in a Heterogeneous Computing Infrastructure)

N/A
N/A
Protected

Academic year: 2022

Partager "Комплексная система управления данными и задачами в гетерогенной компьютерной среде (Integrated Workload and Data Management System in a Heterogeneous Computing Infrastructure)"

Copied!
8
0
0

Texte intégral

(1)

. . . . . “! ! ”,

alexei.klimentov@cern.ch Ruslan.Mashinistov@cern.ch novikov@wdcb.ru Poyda_AA@nrcki.ru itertychnyy@gmail.com

" # #$ %&

, # PanDA,

&' ( !#( &(

, #&()* '+(

* &$

- ! !, $ /, ' - ! ! !##) , (

#( '$ ' ($

!, '!)*$ $

#& $ .

- , No 14.Z50.31.0024.

“!" # ! ”

$ !% ! , ! "

WMS PanDA !

& “!" # ! ”.

"& )* '+ !$

#+(() ' '(

#&! * (. "

$ #0) #& ( %- - ! !, # - '$ ,

!##) , ! .. 1 *

#(() ( &, ( $ '$

%( -&( $

!. 1 & $ & & # '' , &! ) $ ( 2# 3 (3).

#, ( '' $ 2#

ATLAS ( #% '% &#! 3

' #& $ $* ! -# ' % ' ('*

'+ $ 1603 ) # ! !.

#& # !#(

# & (Workload Management System)

PanDA #& '+ % %

* ( $(*

- ! !! / - WLCG [1]) #

! !- !# #) !

!!.

" # #$ %&

, # PanDA,

&' ( !#( &(

, #&()* '+(

* &$

- ! !, $ /, ' - ! ! !##) , (

#( '$ ' ($

!, '!)*$ $

#& $ .

1 !

1.1 " PanDA

" ( &' $% &!)* % #%

' ' #$, #&!

!#( &( PanDA (“Production and Distributed Analysis”) [2], 2007 % !#

#&! ( '' , &

( $ 2# ATLAS,

#% 3 [3].

%̮ # PanDA '#

“#& ” '' $ #̮

̮- ! !. 1 # (

! #( ( % #&

2# $ #0̮, &!

&!) '' ! $, '#

& $ ( ( %!## -&,

#0 #& ̮ # #̮,

# XVII $% &

DAMDID/RCDL’2015 «

», , 13-16 2015

(2)

'# ̮ !# #

%' !, # ( ( ) #($ #̮ &

%!) ! ( !#(

# $.

'! PanDA '

# # '

' ATLAS # '

#($ & ( #% '%

&#! 3 (2010-2013 %%.). ! 1

# # ! &#!*$

& ATLAS !$ WLCG 2015 %.

% # PanDA ' &' %' ( # $ $%̮ '' $( $ $ $%̮.

#& # &' PanDA # ''* $ #$ )

&' #$ , #'$

'+ &

- ! !.

1.2

$ &' % #$ '*(

$ ! &!)* %

# ! 2.

.2. $ ! &' 1 # (() (:

> . #

#& & !#( $ #:

! & # & ( &

', #( $ !#

! (# 2 0 , ' '+ !

! ( '( &), 0 ( & 0& , .1. ? #()*$( & ATLAS !$ WLCG #$ 100000 2015 %.

(3)

!* ( # #

!# #( &, #0 -) $ $ &$

&$ .#.

> &(. &(

!* () $& #& #(& &

. 1 &#!) ( !$,

#&$ ( #( #& $

&, #() $ (, &!) , ') -) #) , # &!.

& ! &#! # &

# !$ !

# #(.

# $ & #&(

!* ( #&)) #(&!, &$ $

#*) $ !-% -

“-# ”. '&, 0 % !) !) !

& - ! !,

#, /-- ! !!, - ! !!

'$ , '' 0 !##) .

@* -! # $ & (( (

#% $$ $ ( & %!&

&! . & ! ') ( '(

$( (& # , & -!), !) # &( %! !- ( (*$ .

&( %! # #

#& & &$ ,

#&(( '& &( ! 0! & .

3 #' # # &( # & 2.2.

> & API '- -,

#&()* #( & !

#()* -! %.

> 1'* - %, #&()*

!- !# , $

#$ $.

1 ( &'

$ # ) $ PanDA

&)) ( !)*:

x # %!$ ! &

( $ ' # $ &.

x # %!$

$ & & ATLAS-#-$

#%$ # , , -%

% !#( .

x ' #( &

& - CERN ( $ %-$ #0 ( ' .

!* &' (( ( ! 0 #(&

#%! '#), #-! (

2# ATLAS. B % Rucio,

-( AGIS, '&# , #&!)*( $& !$

%& CERN, - # ,

#&()* ' PanDA &'

!, (& ATLAS. 3%(

& &0 ##

! ( ,

&' # '+ ' '* #$ ) #!(

, #' '+ &

- ! !.

2 '

2.1 "

- ( . 1

& (( ( !# #

#& $ &, # $ !&, 0 - &# #*

% ) $. # 2 # ' #! &! #& ( #*)

% & $ !#$

#( % % 2 , ' % '! 0 - , $

# &, !# ! '$% 1 , #0 $$

$. !)* 2 # # #(

% &# # #! &, # ) '$!) ( &#!

& -) 0 0& . ' # % ! #! - &!

'* #& ), &! &!, ' #(.

" #% (

&( !!) #%!) &)

PanDA, &

#&( #-$ ( ATLAS

# , :

&0 ! -

#& , )*$ $ - /.

& ATLAS-#-

!#( %, ( % '

&' & % &*(

$ !#( $.

0 ' # ' # #$!

#&) E3F MySQL Oracle.

2.2 (

1 & )$ # (( ( # # $ & (Pilot),

&!)*$ #) #& #(&.

&#! # &( ') -) 0 !. @

! #H ) -'$ '(, # ( !# , # % # &#! # #H ! '

& ' !.

(4)

'&, & ' ! !0 # , # '$ , #% ,

#(&( !! '&

&#!! . '&(( #(&' '$ &

! # * &0 &, &! %' (

! & # #* % (( '' )*$ ! #

&. &( (() ( 0 '&!)* ' , #!!)*

0 -, & !) ' %

!(.

J#! # $ & '$ !&$

'# ( & % # #* (‘-'

# $ &’). &#!*$

# $ & # $ &#! (%

&#! # &(, # $

& #0 .#.)

#() ( -%!$ -$

(#, #) % ). 0

# & “'!0 ” ! #& $ & .

2 Pilot * ( # $

# , - !) ( & .

# (() ( (#, ' %, 0 &), -!) ( (#, '+

!#$ !). J#!* # &

(& ( , #! !#!)

#& !) &! & '! ,

! ()*!) & # (#, # '( '+! '%

! !#$

#), ! #!!) &!

! # #(.

" #% ' &(

# $ & PanDA, ( (

#0 $ $ ' , # $

!$ 3-5.

$, # ( ! 3, (( ( , #&!

PanDA ( #( & 2# ATLAS.

1 # !( # %

&( (() (:

Mover, !* ()* &%!&!

$* $$ $, '$$ (

#( #& &, 0

%!&! % ) &! !)*$ $.

RunJob, &#)* #!)*

#& !) &! &#!)*

# !$

.

.3. $ ' # $ & !&$

!# WAN

 Monitor, #& ( (

#( & ' $ #

# % &(.

$ ' # % &(,

# ( ! 3, ##% , ( , '! #( (

#& ( &, !#

( #&( #(!) '* ( ! $* $). ( -%!(

$ ( $ , , - ! ! '$ . " 2

! # & &#! ( #(

'$ !&$ '

# :

&%!0) ( !) -!) !, '' ) ( -,

#& ( & &#! ( . 1 $ (( ( 0 &#! #$ &,

#&!)*$, #, - #

'* MPI, #& ( &

&#! ( !$, !0 $

# ! &). % # &

&! !, -( , (

#& ( & '! %

! # ! &) !&

'! (, * !#.

.4. $ ' # $ &

! !& ( ( '& !#

WAN)

(5)

$, # ( ! 4,

#& ( ' , $ ' !& ) !#

WAN. ( -%!( $ ( ' !##) . " 2 ! (

&#! # % &( '! (

#0! !&, !#)* )&

0! .

( ! $ !#

#() ( #!( #&

&, &$ ! ( !* ( (

# #!(. " &! , # &

0 && '! '+ !

# &#! #& &, #&(

&#! # &. "$

$ - #) ( & !&

# % &(, '!-&!) (.

1 $ (( (

& ! # #!(

& ( ! ( '(), 0 #( %!& !&

# $ &, ( % '! ( ' '+ !, '%

0 .

.5. $ ' # $ &

! !& ( ( !# WAN)

$, # ( ! 5, (( (

#0! 0! # !(.

" ( 0, #

$, !# WAN. M &0

#* $ &!

0! $*

#(!), # ! (

%!&! !& # $ &. 2

$( ( &0 &#! #$

&.

( & $ #$ $ (! 3) (( ( $ PanDA. " (

$ (! 4) - 0 #&! ( PanDA (

#)( !##) , # Titan [4]. (( $ (! 5) - %(

&' , #0( ' .

2.3 )& &

F( ' 2-- % & (

#& ' &'

# -, #&()* &%!0

$ - & &$ $*, #(

&#! #& &, # 0 ! $

#( ( %). F - & $ #%$ !.

# & #&

' '( '- -,

# ( !-!) '--! (

#( $ #& $ &.

!* % #' & ( (( ( ! '$ #&(

- CERN $& !$

%&.

* #& $ $ - 0! #

!& & ( ! # -.

0 !% & 2 % !( $ #0

$& # '#

% #. " -,

& ' , $ # # !# #) ( # - %. #&

!( # - -%

% '#) &0 ' %

#)( $ $(

(#, Dropbox, Google Drive ..), # !# ( PanDA.

) % #

#% ' & API ( - #%

#%(), #&()* #&

(, !# #& ( '- -, #*) HTTP &#.

'&, &' -

#&( &#! & #& (, ')* - ,

- CERN, #&

$(, #0 PanDA.

3 *

3.1 *

&! 2# !) )

&' “!

! ”, #) $ , )(:

> !##) “!

! ”. "#&

% #(

# #& ) 122,9 TFLOPS 2#! ) ('( 2011 %.

& 1280 H $ !$#$ !&, '+$ #& )

# $ '* InfiniBand DDR, !!) # !) #( 20,5 ' ! $( $ 144 ' . H $

!&$ ! #(

(6)

Linux (CentOS). $( $

# # - Lustre 2.0. F( !#( # !

# $ & #&! ( 0 ! SLURM. Q !&

# ' 1.

' 1. Q !& !##) “! ! ”

H !& #$ Intel Xeon E5450 (3,00 //, 4 ()

# !& 2 ( !& 8 1# ( #( !& (/' ) 16 1# ( #( ( (/' ) 2 V( ( #( !&

(/' )

120 1'* !& 1280 1'* # 2560 1'* ( 10240 1'*( #( #&

(TFLOPS)

122,9

!##) “!

! ” !# , ( % % #& $!,

# !) ! 5. 3 (

$ % # ! 6.

.6. F ( $ % !##

“! ! ” 0 «

# % (»

&( &#!) ( ! , !# !##)

!* ( ( # # ! SSH & !

#& % -, 0 & SLURM. F( #0

SSH- # & SLURM

#&! ( - SAGA (Simple API for Grid Applications) [5], #& ( !-

#&( &$ #$

#0! $ . F( &%!&

$$ $ & !% $*

%!& &! '' #&!) ( &.

" 0 ( #& !)

$! ' # $ & (! 3) ( ' PanDA, ! CERN (. & 3.2 “J( 2# ATLAS”).

3 ( $ #)(

!##) “! ! ” # $ ( #( &

2# ATLAS # ! 7.

.7. $ % !##)

“! ! ” ( #( &

2# ATLAS 0 « # (»

Pilot Launcher &#! # &(

# & - &

(. F( % ' #(

#& &( # '' $

2# ATLAS, '! ( #%

'#, !# #)

! - CVMFS. "$

$ &%!0) ( # Grid Storage Element Tier1 “!

! ”.

> - ! ! '$ ,

&! ( “! ! ”, (*( & 16 !& # %

$* 60 3. F( %

#& !) $! ' # (! 3) .

3.2 + - ATLAS

( ('( 2014 % #(

2# , &!

'' & 2# ATLAS. F(

2 % 2# '& ' # ( ' !#(

& 2# ATLAS PanDA,

! CERN.

# ( ' !* #! &(

(7)

PanDA CERN &, '* #

&(, &#! !$

“! ! ”. " $ 2#

# 200 #& $ & &#!

!# #( 0. 1 &

' 0$ , $

!##) (( ( !( '

# -# % & ( & ( ( &!(

#& % F

$% &!( (TRT).

! 8 # #(

& 2# ATLAS !$

“! ! ” (!0' &

%- # ,

#& - ).

.8. !# &$ &

2# ATLAS !##)

“! ! ” & !, !##)

«! ! » # ) % '!) $! 2# ATLAS. J(

# !#) 0. J ( #(

2# ' '' #( 10-15 (

#& $ &.

3.3 + &

1 & % # '

!* &#! #& $ &, (&$ B& "$ M% '

# #(()*$ #!* &' . "

#$ & ' ' &

( ' % #*)

#%% '#( Bowtie2 [6] Abyss[7].

F #% # $ # #% 2#

ATLAS #&() &#! & # 0. # (

!* () ( !##)

“! ! ”. #&

!# &! . #*) # Quast[8]

! &0 % #

# -'' $$ - - '

% #

!# #( &.

+

#$ ( &!)*( % #&() '+(

* &$

- ! ! ( #( '$

$ ' ($ !, $ B& "$ M%, '- !%$. F #$

# '% ( & '' $

%% (, -& (

&!( $ !, # H , .#.

" $ ' # & ) 2# “!

! ” ' #& ! PanDA, ' # &( %

#& #& %)

!##) “! ! ”.

" ( ( 0 #(

' 200 #& $ &. 0 '

&' , ''* & $ ' # $ &. 3 &'

& # '- - API (

&#! & % $ 0&% . $ #

# &$ # '' $

2# ATLAS #& $ &$

# '%, ' !. " ) , ' #() ( ' $ # !-

# PanDA.

/

[1] Worldwide LHC Computing Grid. # . M !. URL:

http://wlcg.web.cern.ch/

[2] Maeno, Tadashi. "PanDA: distributed production and distributed analysis system for ATLAS." Journal of Physics: Conference Series. Vol. 119. No. 6. IOP Publishing, 2008.

[3] Collaboration, A. T. L. A. S., and G. Aad. "The ATLAS experiment at the CERN large hadron collider." J. Instrum 3 (2008): S08003.

[4] T. Maeno et al. “Evolution of the ATLAS PanDA workload management system for exascale computational science” 2014 J. Phys.: Conf. Ser.

513 032062.

[5] Goodale, Tom, et al. "SAGA: A Simple API for Grid Applications. High-level application

programming on the Grid." Computational Methods in Science and Technology 12.1 (2006)

[6] % ( (

# . # . M !. URL: http://bowtie- bio.sourceforge.net/bowtie2/manual.shtml [7] Simpson, Jared T., et al. "ABySS: a parallel

assembler for short read sequence data." Genome research 19.6 (2009): 1117-1123.

(8)

[8] % ( '.

# . M !. URL:

http://bioinf.spbau.ru/ru/quast

Integrated Workload and Data Management System in a Heterogeneous Computing

Infrastructure

Alexei Klimentov, Ruslan Mashinistov, Alexandr Novikov, Alexey Poyda, Ivan Tertychnyy In this paper an approach to manage computing powers of grids, cloud platforms, supercomputers as well as university and research clusters for computational tasks in data intensive domains is described. This approach uses the experience of the workload management system PanDA (Production and distributed analysis) which was developed at CERN for the ATLAS experiment. Based on this approach, we designed and implemented Workload and Data Management System (WDMS) at NRC “Kurchatov institute”. WDMS is currently in production for High Energy and Nuclear Physics and bioinformatics applications and it can be used for other data- and compute-intensive scientific workflows.

Références

Documents relatifs

ii) La mise en page de la classification : La classification est une extension du MDD qui donne une hiérarchie complète des classes de données. Chaque classe contient un en- semble

When all processors have equal speed, it turns out that a pure cyclic columnwise allocation provides the best solution among all possible distributions of tiles to processors

The plots of the average response

There are many complex problems within the area of cloud computing that can be addressed by solutions that are found through optimization. For example, a) the problems of

Планирование с дублированием задач допускает многократное выполнение некоторых за- дач КП на различных ресурсах, для сокращения времени передачи

By offering various service and deployment models [1], cloud computing provides easy ways to host and manage infrastructures and applications, while reducing deployment and

As SpeQuloS uses the libcloud API as a common interface to interact with various types of Cloud services, we decided to use the Grid5000 API[10] to implement a new

From the perspective of energy efficiency, a cloud computing data center can be defined as a pool of computing and communication resources organized in the way to transform the