Outline - - PDF document

1

1

Self Adapting Numerical Software and Update on NetSolve

• 2

Outline

• Current status

!"#$

• BiCG

%&'(&)*+(&,

• Middleware for ease of use on clusters

)

• !.)!*#/)).

2 /#")0#!"//)* #1!/*)/"#/ 2!3/#")3!)! "/4#"#

• ')&#"**

(/*##"

• !5*/14!#/!")#"#$!*
• )#)*!#")#

1!0"#")#

TUNING SYSTEM Different Algorithms, Segment Sizes Data Structure Best Algorithm, Segment Size Data Structure

Self-Adapting Numerical Software (SANS) Effort

4

Software Generation Strategy - ATLAS BLAS

• 1678#**3

)9373:;<

• =>#!)

"#/"## //)/!#/ !*"# "#$

• !&/
• *"/)
• !.)&/#")
• !*!

*&!/& /!. ).3 #/3/3 ")3 .3/)! <*)3*3?

• %#*!
• #*)")

/!34!/)* 1"#/"#

• *!#*

"#/*

• %/1.!

).

• )9//#*)")

"#$@

• <//
• 9//*
• (%**
• #/
• *
• "!##$

3

5

500 1000 1500 2000 2500 3000 3500 4000 4500 100 200 300 400 500 600 700 800 900 1000 Size Mflop/s

Intel P4 1.5 GHz 32-bit SSE2 Intel P4 1.5 GHz 64-bit SSE2 Intel P4 1.5 GHz

ATLAS Matrix Multiply Intel Pentium 4 at 1.5 GHz – using SSE2

%A;7.)"*7 %ABA.)"*7 %ABA.)"

6C9888#DEC8F84 )"GG

6

Solving Large Sparse Non-Symmetric Systems of Linear Equations Using BiCG-Stab

0#")"#$

• &#.7/"

9)"

• #")!0
• #)=>

)"

4

7

Optimization of BiCG-Stab 10% - 20% Improvement

• !"

# !""

• $"#

$" $!!!"# $!!!" %& 8

Split ADI Method

))#!"".)# )*#!*#$!.)/1 H))/!!DE#! HDD9+,I 9 D+J,9 "/!@ 9 D+7, 9/"35*/ D70I &.7#0/"K

5

9

Performance Of L1Cache A2x Kernel

!7/)) L * .)/1/"* "1 98M#"#. #")"#$

100 200 300 400 500 M flop/s 30 60 90 120 Size

Alpha EV/67 Matrix Vector Multiply

ATLAS Opt by Hand

10

LAPACK For Clusters

!#1**/)* )!#0".)# #")/)/))%&' 1!//#/* .!*N".)#! #

##//.*! "#$.#)* .*!"/! /))/*) % ).*)) ")#

6

11

Big Picture…

Application Library (e.g. LAPACK, ScaLAPACK, PETSc,…) User has problem to solve (e.g. Ax = b) with the assistance of a numerical library

Data (A,b) Answer (x)

12

User has problem to solve (e.g. Ax = b)

Natural Data (A,b)

Middleware Application Library (e.g. LAPACK, ScaLAPACK, PETSc,…)

Friendly intervention…

Natural Answer (x) Structured Data (A’,b’) Structured Answer (x’)

7

13

LAPACK For Clusters - Process

!!1 /))(&* (&!#*/!"" (&*"#/#!))/ =.>*/ (&)!!*."/ (&1/%&'

14

Resource Selector

H.*)!)* #/).)*

7#/+.3),;+)!3/"*3##).),

• !!#/)).).*
• O

O

• O
• O

O O

• O

O

• O

O

• O
• O

O O

• O

O O

• O

O O

• O

O O

• O

O

Bandwidth Latency Load CPU Performance Memory

8

15

Middleware Function

Middleware User d/dt T() = 0

1. User invokes help of middleware to generate data set in remote memory depot, IBP 2. Middleware makes NWS query of available clusters

• btaining current, relevant system/network information
• 3. Resources are selected based on the application specific

performance model and information from step 2.

• The data handle (IBP capability) is passed over the network

to the lead compute node and the solve invoked (here the structured mapping of the data is imparted in the application routine)

(clusters)

1 2 3 4

16

Pentium III 933 MHz Ax = b LAPACK / ScaLAPACK / LFC

LAPACK routine

ScaLAPACK and LFC Routines No loading of data measured

9

17

Pentium III 933 MHz Ax = b LAPACK/ScaLAPACK/LFC

LAPACK routine

ScaLAPACK routine LFC Routine || load of data LFC Routine seq load of data

18

LAPACK For Clusters (LFC)

(&))*# #*/!/ &)* #"! .#)*

• !" !#/

#

• /#")0

&)*!-!/ !!)" )!".)

• 2!/!!)"!

)!*#! !

)"./ */* /#"*)// "")/!

• &)*!-!

#

• /1)#"

!)"#!

%)!3 &)13P3 ##/)*3 !##/ )*

10

19

NetSolve - Grid Enabled Server

)0#")-!.! !4 <!#%/!*&))#!).* ?

*/!/3!#/".)#)/".)3)! .)/3*))///3/*3?

*"#* 0#")#! ("

NetSolve: The Big Picture

AGENT(s)

A C

S1 S2 S3 S4

Client

Matlab Mathematica C, Fortran Web Schedule Database

No knowledge of the grid required, RPC like.

IBP Depot

11

NetSolve: The Big Picture

AGENT(s)

A C

S1 S2 S3 S4

Client

Matlab Mathematica C, Fortran Web Schedule Database

No knowledge of the grid required, RPC like. A, B

IBP Depot

NetSolve: The Big Picture

AGENT(s)

A C

S1 S2 S3 S4

Client

Matlab Mathematica C, Fortran Web Schedule Database

No knowledge of the grid required, RPC like.

H a n d l e b a c k

IBP Depot

12

NetSolve: The Big Picture

AGENT(s)

A C

S1 S2 S3 S4

Client Answer (C)

S2 ! Request

Op(C, A, B)

Matlab Mathematica C, Fortran Web Schedule Database

No knowledge of the grid required, RPC like. A, B OP, handle

IBP Depot

24

Basic Usage Scenarios

• !.!*#/)).

*

• !N

).#/3%&'3 *"3/%&'3%/3Q&3 %&'

• 1#"")/
• =%))")))>0/*
• %#*!
• #"")/0/*
• &#")"")/*

"/"*"#! /*"*

• =<)*&))>-!<!

&#"*

• 5*!"1)!

1"##

• )0"

#*

• /*"3=*>

5*!

• *!3*"788

R/*

• &")*-).*3&!3

(3?

13

25

Generating New Services in NetSolve

!!!!)*/)

• /./+*#,
• ""
• ))

Java GUI NetSolve Parser/ Compiler

@PROBLEM degsv @DESCRIPTION This is a linear solver for dense matrices from the LAPACK

• Library. Solves Ax=b.

@INPUT 2 @OBJECT MATRIX DOUBLE A Double precision matrix @OBJECT VECTOR DOUBLE b Right hand side @OUTPUT 1 @OBJECT VECTOR DOUBLE x …

Server

Service Service Service Service New Service

New Service Added!

26

Task Farming - Multiple Requests To Single Problem Used Early on with MCell

)*@

• (#)*@
• 5*="*

"#> !"/!*))# *)"#"3!!"!) ")))"")/

14

27

IPARS

• !%)))//*#*)
• & 3*3/3 H)
• %.)! !*)#*)
• !))!3.)/1)3/#")3))##3

/"/?

• !#)#*)
• #!).)/1)3)!3/#"
• ;)#*)")"
• 0#*)
• (#1#")!!)"#
• %!)3!)))3.))1*"
• %!"4""/3*)$
• (*))%//*))
• %/@
• &3(3 ).3 #/3!H.

28

Web Server NetSolve Client IPARS-enabled Servers Web Interface

)""/*)$ %.)!*)$ *..

15

29

NetSolve

C Fortran Matlab Mathematica SciRUN Globus NetSolve servers Ninf servers NetSolve servers Condor NetSolve servers

Globus proxy NetSolve proxy Ninf proxy Condor proxy

PSE front-ends Grid back-ends Web Interface Grid middleware

Remote procedure call Resource Discovery System Management Resource Scheduling Fault Tolerance

NetSolve: A Plug into the Grid

Scilab// Custom

30

Things Not Touched On

• /*
• '.LF*/
• "&//
• #")#2!!
• 2/
• /).)/*
• )1
• /1!#*
• 1*
• 1H/
• </1")%/)
• !!)#!*#
• (*))/
• L)1*#"N %/
• /)4-).)&*
• #/*
• *#!!")#)/
• #/!#)#.!#*!

**".)#

16

31

Contributors

• L/S1*3'
• % *$/$13'
• '/3'
• &)H)3(

) #A@88

• *! )3'
• 2&3&
• /)) ))3'
• Q 3'
• L!3'

(!!)#?

"@44).4)4 "@44/)/*1!*4)4 "@44/*1!*46!4 )!!)

""* *"