D ESIGN OF COOPERATIVE VISUALIZATION ENVIRONMENT WITH INTENSIVE DATA - PowerPoint PPT Presentation

D ESIGN OF COOPERATIVE VISUALIZATION ENVIRONMENT WITH INTENSIVE DATA MANAGEMENT IN PROJECT LIFECYCLE Kenji ONO and Jorji Nonaka High-performance Computing Team Integrated Simulation of Living Matter Group Computational Science Research Program RIKEN Ultrascale Vis Workshop 2008

TOC  Supercomputer development project  Computer Hardware development  Grand Challenge in Life Science  Post-process system  Current status of system design 2

N EXT - GENERATION SUPERCOMPUTER �  Development  Under initiative of MEXT  Riken is responsible for dev. in collaboration with vendors  Hybrid System composed of Scalar and Vector Units  LINPACK 10PFLOPS Control unit Processing unit Front end computer Scalar Unit Vector Unit (Integrated OS) Data base, Vector processor homology search Scalar Unit Vector Unit based on the Earth Similarity with PC Simulator - - e - - I 3 e - • high speed low • New generation - I power CPU low-power vector Solar cell • New strong processor with network for Homology optical network search Climate change CFD enormous Nono device parallelism Plazma Physics Disk unit 3 Local file Local file 共有ファイル Global file

E XPECTED M AJOR A PPLICATIONS Grand challenges • Applica'ons • Benchmark • Grand challenge • Focusing on life science 4

I SSUES TO BE CONSIDERED  Large-scale nature of data  computational space, time-varying, multivariate  Expected data size is order of 1PB (Peak)  Data Tsunami  How to do vis., data processing, analysis?  Depend on each researcher  Various scenarios  Complex hardware  Various configurations 5

C URRENT STATUS AND GOAL  System design  To develop user-friendly post-processing system  Derive useful information from Large-scale dataset  Easy to use  Enhance productivity  To assist scientific knowledge and understanding physical phenomena  Operating this post-process system to assist research 6

I SSUES ON POST - PROCESS OF LARGE - SCALE DATASET  Size of dataset  Space, Temporal data, multivariate  Distribution of dataset  Distributed parallel, GRID  Complexity of HW system  Heterogeneous environment, File system, Network  High cost of data copy, movement  MMU/HDD capacity  can not move any large-scale data  need appropriate tools to access large-scale data  Many cumbersome procedures  File handling, preparation of process  need an environment to focus on “THINKING” 7

O RIENTATION OF POST - PROCESS  do note move data  beyond that…  Sharing data, collaboration  Remote, virtual organization  Data intensive service  Data access, visualization, analysis, processing  Sharing data, results, knowledge, resource  Data repository for group  Data browse, search, break down  Comprehension of phenomena 8  Sharing information, knowledge

W HAT WE WANT TO DO 9

SWEETS � S CIENTIFIC K NOWLEDGE D ISCOVERY T OOLS  Script based loosely coupling  script glues modules in sub-systems  scalability, flexibility and sustainability  Sub-system  Visualization  Data / project management  Automation of a routine task using a workflow  Data sharing  ...  module works independently, but provide full capability when working together  E ffi ciently extract useful information from simulated results 10 Tools to obtain knowledge, then to magnify value of simulations

R EQUIRED TASKS FOR RESEARCH Thinking Cumbersome File access Parameter edit Visualiza'on Data copy/move Data/project mgmt Visualiza'on Submit a batch job Automa'on Data analysis Program launch Data sharing ... Analysis Service Task Low level 11 (sub‐system) procedure

A N EXAMPLE OF HARDWARE CONFIGURATION !899:!1 !$%6$%1;< =/$%1> !899:!1 =/$%1< >-0$+)1;< 8'%C ! 'D '+ !$%6$% 8'%C ! 'D !899:! '+ *7$+) !$%6$%1 >-0$+) =/$%1* 30/?1*## !),)0/)0&1*## 12 @'A1B,+,7$($+) *+,-./0/121 50-$1!$%6$% !"#$%&'(#")$% 30/",-04,)0'+

A CONCEPT MODEL OF DATA ENTITY 0 1 0 1 !"#$%&' (#) *+,% 0 1 1 1 1 1 1 0 0 0 -.%" /#.' 13

D ATA ENTITY OF SWEETS SYSTEM !"#$%&'()* +#,()* -./%()01#2 !"#$%&'()*+% !#)*$+&',- -./%()*+% ,#++%)' <#=()*+% 1)2',- -.%/0 ) ,#++%)' ?%*/()*+%(#2(*( ! /% 1'*"'(#2('3%(0*4 N !*'3 0)/&',- 5)0(#2('3%(0*4 1'*'9> A."%&'#"4 6).7(8"#9: ./$#',- -./%('4:% !*'3 0)/&',- 1.B% ;%'*(0*'* <#=(&#++*)0 @)#0% !"#$%&'(#)*$+&',- <#=(0."%&'#"4 C.+%(#2(&"%*'.#) ?%*/(@A(#2(*($#= C.+%(#2(9:0*'% ;%'*(0*'* !%"+.>>.#) D"#9: EF)%" G.)H('# @E ,#++%)' 34%"()01#2 5%'6(76'6()* 6)H)#F)( " *8 !"%I.%F(:*'3 6>%"()*+% ;%'*(0*'*()*+% ;%'*(0*'* A#+*.)(#2(%7:%"'.>% C*"8%' ?%/*'%0( ! /%> ;*./(*00"%>> ,*'%8#"4 ,#)'%)'(#2(*( ! /% ,#++%)' C4:% J9'3#".'4 ;.)K(I*/9% G#8.)(@A ;*7(I*/9% !%">#)*/(>%''.)8 J/'%")*'.I% !"#$%&'(@A G.)H*=/% 8#4'()* G.)H(:"#8"*+ -/*8 M#>'()*+% 57'%)>.#)(2#"(*9'#+*'.&( 14 ,/9>'%"()*+% "%8.>'"*'.#) @!(*00"%>> L%4(2#"(*9'#+*'.&("%8.>'"*'.#) G#8.)(@A !*>>F#"0 M#+%(0."%&'#"4

P REREQUISITE OF VIS . SUB - SYSTEM Various scenarios for each researcher’s approach  Remote or local visualization  Provides unified environment by a common client  Real-time or post visualization  Basically, file based visualization  Interactive or batch visualization  Software or hardware rendering  Large-scale data handling  Parallel rendering  Platform 15  Linux, Windows, Max OSX, PC cluster, Supercomputer

2- WAY SCENARIO  1st step : Visualization on supercomputer  Sever client system  with basic visualization function  Data reduction, ROI  Operation is only batch job by its policy  2nd step : Visualization on CPU/GPU cluster  Capability of interactive visualization by GPU  Relatively small MMU (Data reduction is necessary)  Reduce risk of development  Reduce utility time of supercomputer  can use existing software, COTS 16

S TRATEGY : INTERACTIVITY AND SCALABILITY File I/O LOD Rendering Data reduc8on Time Mul8‐Step Image Mul8core rendering / Composi8on HW rendering Image composi8ng 17 # of Cores

B INARY -S WAP I MAGE C OMPOSITION Composi8on Nodes (n) 1 2 3 4 n‐3 n‐2 n‐1 n Stage . . . 1 Stage . . . 2 . . . . . . Stage log 2 (n) . . . Network conten8on Image . . . Collect 18

M ULTI -S TEP I MAGE C OMPOSITION Composi8on Nodes: n = p * m 1 . . . p 1 2 3 m 1 2 3 m . . . Step ... ... 1 1 n Binary‐Swap Binary‐Swap 2 p 1 Step . . . Local root nodes 2 Binary‐Swap / Binary‐Tree / Direct Send 1 Global root node 19 Final composited image

MSIC BG/L (RIKEN) BlueGene / L Dual core x 1024 nodes 20

MSIC - T2K (U NIV . OF T OKYO ) AMD Opteron Quad Core x 4 Sockets 21

OTHER SUB - SYSTEMS  Workflow  Kepler  Data base  RDB ? XML feature  Scripting  python ... plays glue  Analysis  R  user program  Other useful existing tools... 22

C ONCLUDING REMARKS  Design and current development status  to assist discovery and understanding  Database centered structure  sub-systems take part in the system  Visualization  Workflow  Other programs  resource mgmt and access control provide data sharing  reuse of existing useful software  script base  Supercomputer will be fully operational on 2011  Combining existing software, enhancing capability 23