Open TS: an Advanced Tool for Parallel and Distributed Computing

1
Open TS an Advanced Tool forParallel and
Distributed Computing

• PSI RAS Moscow State University,Sergey M.
  Abramov, 23/02/05(presentation for Microsoft
  Research Ltd., Cambridge)

2
Proposal Motivation

• Parallel architecture multitude multi-core CPUs,
  SMP, clusters, PC farms, supercomputers,
  distributed systems, GRID
• Parallel programming is still too complex
• Challenge
• Parallel programming should be easier to be
  widely used
• Software should be portable and efficient
• Proposal addresses this for Windows platform

3
Presentation Outline

• Short self-introduction
• Open TS architecture applications
• Proposal
• Open TS
• Web service synthesis
• Approach testing (applications)

4
1. Short Self-Introduction
5
Program Systems Institute

• PSI RAS was established in 1984 by a decree
  issued by the USSR government
• Goals to foster the development of computer
  science in the country

The first buildings of the Institute
The first directorof the InstituteProfessor
A.Aylamazyan
6
2005 PSI RAS, Pereslavl-Zalesski
7
2005 Institute Structure

• Artificial Intelligence Research Centre
• Medical Informatics Research Centre
• Research Center for Multiprocessor Systems
• System Analysis Research Centre
• Control Processes Research Centre
• Scientific and educational centre
• International Childrens Computer Centre named
  after A. Aylamazyan
• Kindergarten and Primary school Pochemuchka

8
University of Pereslavl-Zalesski
9
SKIF Supercomputing Project

• Joint project of the Russian Federation and the
  Republic of Belarus
• 2000-2004
• 10 10 organizations
• PSI RAS is the leading organization from the
  Russian Federation
• Hardware and Software

10
Flagship SKIF ?-1000

• Peak performance2,5 Tflops
• Linpack-performance2,0 Tflops
• Efficiency factor80.1

• The most powerful supercomputer in ex-USSR
• Ranked No98 in Top500 (November 2004)

11
Moscow State University
2005 MSU celebrates 250th anniversary
12
2. Proposal BackgroundOpen TS Architecture
Applications
13
T-System History

• Mid-80-iesBasic ideas of T-System
• 1990-iesFirst implementation of T-System
• 2001-2002, SKIF GRACE Graph Reduction
  Applied to Cluster Environment
• 2003-current, SKIF Open TS Open T-system

14
Open TS an Outline

• High-performance computing
• Automatic dynamic parallelization
• Combining functional and imperative approaches,
  high-level parallel programming
• ? language Parallel dialect of C an
  approach popular in 90-ies

15
Open TS Environment
16
?-Approach

• Pure functions (tfunc) invocations produce
  grains of parallelism
• T-Program is
• Functional on higher level
• Imperative on low level (optimization)
• C-compatible execution model
• Seamless C-extension (or Fortran-extension)
• Multiple assignment

17
Comparison T-System and MPI
Sequential
Parallel
18
? Seven Keywords C

• tfun ?-function
• tval ?-variable
• tptr ?-pointer
• tout Output parameter (like )
• tdrop Make ready
• twait Wait for readiness
• tct ?-context

19
Sample Program

• include ltstdio.hgt
• tfun int fib (int n)
• return n lt 2 ? n fib(n-1)fib(n-2)
• tfun int main (int argc, char argv)
• if (argc ! 2) printf("Usage fib ltngt\n")
  return 1
• int n atoi(argv1)
• printf("fib(d) d\n", n, (int)fib(n))
• return 0

20
Open TS Runtime

• Three-tiered architecture (?, M, S)
• Design microkernel , 10 extensions currently
• Supermemory
• Lightweight threads
• DMPI Dynamic MPI
• auto selection of MPI implementation
• dynamic loading and linking

21
Supermemory

• Object-Oriented Distributed shared memory (OO
  DSM)
• Global address space
• Cell versioning

22
Multithreading Communications

• Lightweight threads
• PIXELS (1 000 000 threads)
• Asynchronous communications
• A thread A asks non-ready value (or new job)
• Asynchronous request sent Active messages
  Signals delivery over network to stimulate data
  transfer to the thread A
• Context switches (including a quant for
  communications)
• Latency Hiding for node-node exchange

23
DMPI

• Dynamic MPI
• auto selection of MPI implementation
• dynamic loading and linking
• Seven implementations of MPI supported now
• LAM
• MPICH
• SCALI MPI
• MVAPICH
• IMPI
• MPICH-G2
• PACX-MPI
• even PVM can be used instead of MPI

24
Debugging WAD, LTDB
25
Statistics Gathering
26
Message Tracing
27
Tests NASA CG, NASA EP, FIB
28
Open TS Applying to Distributed Computing

• Meta-cluster messaging support(MPICH-G2, IMPI,
  PACX-MPI)
• Customizable scheduling strategies(network
  topology information used)

29
Out of Scope ofthis Presentation

• Other T-languages T-Refal, T-Fortan
• Memoization
• Automatically choosing between call-style and
  fork-style of function invocation
• Checkpointing
• Heartbeat mechanism
• Flavours of data references normal, glue and
  magnetic lazy, eager and ultra-eager
  (speculative) data transfer

30
T-System in Comparison
Related work Open TS differentiator
Charm FP-based approach
mpC Implicit parallelism
Glasgow Parallel Haskell Allows C/C based low-level optimization
OMPC Provides both language and C templates library
Cilk Supports SMP, MPI, PVM, and GRID platforms
31
Out of Comparison Scope

• Parallel Programming Using C (Scientific and
  Engineering Computation) by Gregory V. Wilson
  (Editor), Paul Lu (Editor)ABC, Amelia, CC,
  CHAOS, COOL, C//, ICC, Mentat, MPC,
  MPI, pC, POOMA, TAU, UC

32
Open TS Current Status

• Tested for quality under SKIF project
• Approximately 20 applications (some notes below)
• Supported by grants of Russian Academy of Sciences

33
?-Applications

• MultiGen biological activity estimation
• Remote sensing applications
• Plasma modeling
• Protein simulation
• Aeromechanics
• Query engine for XML
• AI-applications
• etc.

34
MultiGenChelyabinsk State University
?0
Level 0
Level 1
?11
?12
Level 2
?22
?21
Multi-conformation model
35
MultiGen Speedup
National Cancer Institute USA Reg.No.
NCI-609067 (AIDS drug lead)
National Cancer Institute USA Reg.No.
NCI-641295 (AIDS drug lead)
TOSLAB company (Russia-Belgium) Reg.No. TOSLAB
A2-0261 (antiphlogistic drug lead)
Substance Atom number Rotations number Conformers Exectution time (min.?) Exectution time (min.?) Exectution time (min.?)
Substance Atom number Rotations number Conformers 1 node 4 nodes 16 nodes
NCI-609067 28 4 13 933 321 122
TOSLAB A2-0261 82 18 49 11527 3923 1609
NCI-641295 126 25 74 26619 9557 3448
36
AeromechanicsInstitute of Mechanics, MSU
37
Belocerkovskis approach
flow presented asa collection of
smallelementary whirlwind(colours
clockwiseand contra-clockwiserotation)
38
Aeromechanics Speedup
500 steps 1 node 2 nodes 4 nodes 8 nodes
C 160
MPI 160 87 56 41
T 214 130 81 61
39
Creating Images fromSpace-Born Radar Holograms
40
Simulating Broadband Radar Signal

• Graphical User Interface
• Non-PSI RAS development team (Space research
  institute of Khrunichev corp.)

41
LANDSAT Image Classification

• Computational web-service

42
Further Development

• Distributed computing
• Schedulers use information of the network
  topology and connectivity between nodes
• DMPI add new transports
• Interface to web-services
• Fault-tolerance
• Optimizing for modern CPUs
• Algorithmic skeletons, patterns and high level
  parallel libraries

Supportedby Russian Academy of Sciences
43
3. The ProposalOpen TS an Advanced Tool for
Parallel and Distributed Computing
44
Open TS Proposal

• Open TS migration to MS Windows
• Web service
• Approach illustrated by applications
• Artificial intelligence
• Molecular modeling

45
Migration Approach

• Goal provide Open TS on Windows clusters and
  SMPs (Win 32 platform)
• Migration issues
• MPI available
• Linux-specific memory management
• Compilation of T-programs

46
Web-Service Synthesis

• T-functions are pure easy to expose as
  web-services
• Ease creation of computational services
• Easy integration of computational service with
• Data
• Other services
• Build end-to-end data processing chains

47
Approach demos and presentations

• Current importance of selected applications
• Two applications
• Artificial Intelligence
• Molecular modeling (QM/MM approach)
• Development of T-programs by experts in
  application areas, not in T-programming
• Testing of T-System usability

48
AKTIS Text Categorization System

• The goal classification documents into a
  predefined categories
• The categories are defined by a training set a
  set of category-labeled documents
• Features
• Machine learning techniques will be used
• One-word and multi-word terms will be used
• Extraction of multi-word terms will be based on
  partial syntactic analysis
• Different types of term occurrence will be
  taking into account

49
AKTIS High Performance Computing is Needed

• Complicated algorithms and a big incoming data
  flow require a powerful computational facility
• Input texts of varying size and complexity
  require dynamical load balancing
• Open-TS will be used to solve the issues

50
Moscow State University Team

• Department of Chemistry,M. V. Lomonosov Moscow
  State University
• Molecular modeling.QM/MM multi-scale technique

51
The aim is to study mechanisms of chemical
reactions in complex molecular environment by
considering multidimensional potential energy
surfaces in the QM/MM approach E EQM EMM (
EQM/MM)
52
Running ProjectCleavage of Insulinby
Serine-Carboxyl Peptidases
E EQM EMM ( EQM/MM)
QM
Catalytic triad Asp84
Glu80 Ser287
Substrate Insulin
53
Project Plan
5. Web-service tool Development and testing 6.
AI Initial implementation 7. AI Testing and
performance tuning on Windows 8. QM/MM Initial
implementation 9. QM/MM Testing and performance
tuning on the Windows

• 1. Open TS Preliminary porting design
• 2. Open TS Migrating to the Windows platform
• 3. Open TS bugfixing and performance tuning
• 4. Web-service tool Design of web-service
  synthesis tool

2005
2006
Q1
Q2
Q3
Q4
Q1
1
2
3
4
5
6
7
8
9
54
THANKS

• QUESTIONS ARE WELCOME