Mathworks PARALLEL COMPUTING TOOLBOX RELEASE NOTES

Download

Page 1

Parallel Computing Toolbox™ Release Notes

Page 2

How to Contact The MathWorks

www.mathworks. comp.soft-sys.matlab Newsgroup www.mathworks.com/contact_TS.html Technical Support

suggest@mathworks.com Product enhancement suggestions

bugs@mathwo doc@mathworks.com Documentation error reports service@mathworks.com Order status, license renewals, passcodes

info@mathwo

com

rks.com

Web

Bug reports

Sales, prici

ng, and general information

508-647-7000 (Phone)

508-647-7001 (Fax)

The MathWorks, Inc. 3 Apple Hill Drive Natick, MA 01760-2098

For contact information about worldwide offices, see the MathWorks Web site.

Parallel Computing Toolbox™ Release Notes

The software described in this document is furnished under a license agreement. The software may be used or copied only under the terms of the license agreement. No part of this manual may be photocopied or reproduced in any form without prior written consent from The MathW orks, Inc.

FEDERAL ACQUISITION: This provision applies to all acquisitions of the Program and Documentation by, for, or through the federal government of the United States. By accepting delivery of the Program or Documentation, the government hereby agrees that this software or documentation qualifies as commercial computer software or commercial computer software documentation as such terms are used or defined in FAR 12.212, DFARS Part 227.72, and DFARS 252.227-7014. Accordingly, the terms and conditions of this Agreement and only those rights specified in this Agreement, shall pertain to and govern theuse,modification,reproduction,release,performance,display,anddisclosureoftheProgramand Documentation by the federal government (or other entity acquiring for or through the federal government) and shall supersede any conflicting contractual terms or conditions. If this License fails to meet the government’s needs or is inconsistent in any respect with federal procurement law, the government agrees to return the Program and Docu mentation, unused, to The MathWorks, Inc.

Trademarks

MATLAB and Simulink are registered trademarks of The MathWorks, Inc. See

www.mathworks.com/trademarks for a list of additional trademarks. Other product or brand

names may be trademarks or registered trademarks of their respective holders.

Patents

The MathWorks products are protected by one or more U.S. patents. Please see

www.mathworks.com/patents for more information.

Page 3

Summary by Version ............................... 1

Version 4.3 (R2010a) Parallel Computing Toolbox

Software

Version 4.2 (R2009b) Parallel Computing Toolbox

Software

Version 4.1 (R2009a) Parallel Computing Toolbox

Software

Version 4.0 (R2008b) Parallel Computing Toolbox

Software

Version 3.3 (R2008a) Parallel Computing Toolbox

Software

........................................ 3

........................................ 7

........................................ 13

........................................ 16

........................................ 20

Contents

Version 3.2 (R2007b) Distributed Computing Toolbox

Software

Version 3.1 (R2007a) Distributed Computing Toolbox

Software

Version 3.0 (R2006b) Distributed Computing Toolbox

Software

Compatibility Summary for Parallel Computing

Toolbox Software

........................................ 24

........................................ 27

........................................ 32

................................ 37

iii

Page 4

iv Contents

Page 5

SummarybyVersion

This table provides quick access to what is new in each version. For clarification, see “Using Release Notes” on page 1 .

Parallel Computing Toolbox™ Release Notes

Version (Release)

Latest Versi V4.3 (R2010a

V4.2 (R2009b)

V4.1 (R2009a)

V4.0 (R2

V3.3 (R2008a)

V3.2 (R2007b)

V3.1 (

V3.0 (R2006b)

008b)

R2007a)

New Features and Changes

Yes

)

Details

Yes Details

Yes Detai

Yes Details

Version Compatibilit Consideratio

Yes Summary

Yes

Summa

Yes Summary

Fixed Bugs and Known Problems

Bug Reports Includes fix

Bug Reports Includes fixes

Bug Repor Includes

Bug Reports Includes fixes

Bug Rep Inclu

Bug Reports Includes fixes

fixes

orts

des fixes

Related Documentation at Web Site

Printable R elease Notes: PDF

Current product documentation

Using Release Notes

Use release notes when upgrading to a newer version to learn about:

• New features

• Changes

• Potential impact on your existing files and practices

Page 6

Parallel Computing Toolbox™ Release Notes

Review the release notes for other MathWorks™ products required for this product (for example, MATLAB

or Simulink®). Determine if enhancements,

bugs, or compatibility considerations in other products impact you.

If you are upgrading from a software version other than the m ost recent one, review the current release notes and all interim versions. For example, when you upg rade from V1.0 to V1.2, review the release notes for V1.1 and V1.2.

What Is in the Release Notes

New Features and Changes

• New functionality

• Changes to existing functionality

Version Compatibility Con si derations

When a new feature or change introduces a reported incompatibility between versions, the Compatibility Considerations subsection explains the impact.

Compatibility issues reported after the product release appear under Bug Reports at The MathWorks™ Web site. Bug fixes can sometimes result in incompatibilities, so review the fixed bugs in Bug Reports for any compatibility impact.

Fixed Bugs and Known Problems

The MathWorks offers a user-searchable Bug Reports database so you can view Bug Reports. The development team updates this database at release time and as more information becomes available. Bug Reports include provisions for any known workarounds or file replacem ents. Information is available for bugs existing in or fixed in Release 14SP2 or later. Information is not avail able for all bugs in earlier releases.

Access Bug Reports using y our MathWorks Account.

Page 7

Version 4.3 (R2010a) Parallel Computing Toolbox™ Software

Version 4.3 (R2010a) Parallel Computing Toolbox Software

New Features and Changes

Yes Details below

This table summ

Version Compatibility Considerations

Yes — Details as Compatibi

Considerat

below. See a Summary.

New features and changes introduced in this version are

• “New Save and Load Abilities for Distributed Arrays” on page 3

• “Enhanced Functions for Distributed Arrays” on page 4

• “Importing Configurations Programmatically” on page 4

• “Enhanced 2-D Block-Cyclic Array Distribution” on page 4

• “New Remote Startup of mdce Process” on page 4

• “Obtaining mdce Process Version” on page 4

• “Demo Updates” on page 5

arizes what is new in Version 4.3 (R2010a):

labeled

lity

ions,

lso

Fixed Bugs an d Known Problems

Bug Reports Includes fixes

Related Documentation at Web Site

Printable Re Notes: PDF

Current pro documentat

lease

duct

ion

• “taskFinish File for MATLAB Pool” on page 5

• “Upgrade Parallel Computing Products Together” on page 5

New Save and Load Abilities for Distributed Arrays

You now have the ability to save distributed arrays from the client to a single MAT-file. Subsequen tly, in the client y ou can load a distributed array from that file and have it automatically distributed to the MATLAB pool workers. Thepoolsizeanddistributionschemeofthearraydonothavetobethesame when you load the array as they were when you saved it.

Page 8

Parallel Computing Toolbox™ Release Notes

You also can now l oa d data directly into distributed arrays, even if the originally saved arrays were not distributed.

For more information, see the

dsave and dload reference pages.

Enhanced Functions for Distributed Arrays

The svd function now supports single-precision and complex data in distributed arrays. Other functions enhanced to support single-precision distributed arrays are

In addition to their original support for 1-D distribution by columns, the enhanced by rows or when the distribution dimension is greater than 2, and 2-D block-cyclic (

tril and triu functions now support arrays with 1-D distribution

'2dbc') distributed arrays.

chol, lu, mldivide,andeig.

Importing Configurations Programmatically

A new function allows you to programmatically import parallel configurations. For m ore information, see the

importParallelConfig reference page.

Enhanced 2-D Block-Cyclic Array Distribution

2-D block-cyclic ('2dbc') array distribution now supports column orientation of the lab grid. The for its orientat ion argument, which is reflected in the codistributor object’s

Orientation property. For information on '2dbc' distribution and using lab

grids, see “2-Dimensional Distribution”.

codistributor2dbc function now accepts the value 'col'

New Remote Startup of mdce Process

New command-line functionality allows you to remotely start up MATLAB Distributed Computing Server proc esses on cluster machines from the desktop computer. For more information, see the

remotemdce reference page.

Obtaining mdce Process Version

An enhancement to the mdc e command lets you get the command version by executing

Page 9

Version 4.3 (R2010a) Parallel Computing Toolbox™ Software

mdce -version

For more information on this command, see the mdce reference page.

Demo Updates

Product demos ar e available in the Demos node under Parallel Computing

Toolbox

Benchmarking A\b

This new demo benchmarks Parallel Computing Toolbox performance with the

BER Performance of Equalizer Types

Demos of BER Performance of Several Equalizer Types have been removed from Parallel Computing Toolbox, because Communications Toolbox now incorporates support for Parallel Computing Toolbox. See “New Demos” in the Communications Toolbox release notes.

in the help broswer.

mldivide function.

taskFinish File for MATLAB Pool

The taskFinish file (taskFinish.m) for a MATLAB pool now executes when the pool closes.

Compatibility Considerations

In previous versions of the software, the taskFinish file ex ecuted when a MATLAB pool opened. Beginning with this release, it runs when the pool closes.

Upgrade Parallel Computing Products Together

This version of Parallel Computing Toolbox™ software is accompanied by a corresponding new version of MATLAB software.

Distributed Computing Server™

Page 10

Parallel Computing Toolbox™ Release Notes

Compatibility Considerations

As with every new release, you must upgrade both Parallel Computing Toolbox and MATLAB Distributed Computing Server products together. These products must be the same version to interact properly with each other.

Jobs created in one version of Parallel Computing Toolbox software will not run in a different version of MATLAB Distributed Computing Server software, and might not be readable in different versions of the toolbox software.

Page 11

Version 4.2 (R2009b) Parallel Computing Toolbox™ Software

Version 4.2 (R2009b) Parallel Computing Toolbox Software

New Features and Changes

Yes Details below

This table summ

Version Compatibility Considerations

Yes — Details as Compatibi

Considerat

below. See a Summary.

New features and changes introduced in this version are

• “New Distributed Arrays” on page 7

• “Renamed codistributor Functions” on page 8

• “Enhanceme nts to Admin Center” o n page 10

• “Adding or Updating File Dependencies in an Open MATLAB Pool” on

page 10

• “Updated globalIndices Function” on page 10

• “Support for Job Templates and Description Fil es with HPC Server 2008”

on page 11

arizes what is new in Version 4.2 (R2009b):

labeled

lity

ions,

lso

Fixed Bugs an d Known Problems

Bug Reports Includes fixes

Related Documentation at Web Site

• “HPC Challenge Benchmarks” on page 11

• “pctconfig Enhanced to Support Rang e of Ports” on page 11

• “Random Number G enerator on Client Versus Workers” on page 12

New Distributed Arrays

A new form of distributed arrays provides direct access from the client to data stored on the workers in a MATLAB pool. Distributed arrays have the same appearance and rules of indexing as regular arrays.

Page 12

Parallel Computing Toolbox™ Release Notes

You can distribute an existing array from the client workspace with the command

D = distributed(X)

where X is an array in the client, and D is a distributed array with its data on the workers in the MATLAB pool. Distributing an array is performed outside an

spmd statement, but a MATLAB pool must be open.

Codistributed arrays that you create on the workers within are accessible on the client as distributed arrays.

The following new functions and methods s upport distributed arrays.

Function Name Description

distributed

Distribute existing array from client workspace to workers

distributed.rand, distributed.ones,etc.

Create distributed array consistent with indicated method, constructing on workers only

gather

Transfer data from MATLAB pool workers to client

isdistributed

C(x,y)

True for distribu t ed array

Indexing into distributed array C on client to access data stored as codistributed arrays on workers

Renamed codistributor Functions

As part of the general enhancements for distributed arrays, several changes to the codistributed interface appear in this release.

spmd statements

Compatibility Considerations

The following table summarizes the changesinfunctionnamesrelatingto codistributed arrays.

Page 13

Version 4.2 (R2009b) Parallel Computing Toolbox™ Software

Old Function Name

codcolon codistributed.colon

codistributed(..., 'convert') codistributed(...)

codistributed(...) without 'convert'

New Function Name

codistributed.build

option

codistributed(L, D) using distribution

scheme of

codistributor('1d', ...)

codistributor('2d', ...) codistributor('2dbc', ...) or

codistributor(arrayname) getCodistributor

defaultLabGrid codistributor2dbc.defaultLabGrid

defaultPartition codistributor1d.defaultPartition

isa(X, 'codistributed')

localPart getLocalPart

redistribute(D) using default distribution

D to define that of L

codistributed.build(L, getCodistributor(D))

Still available, but can also use

codistributor1d

codistributor2dbc

Still available, but can also use

iscodistributed(X)

redistribute(D, codistributor())

scheme

redistribute(D1, D2) using distribution

scheme of

D2 to define that of D1

redistribute(D1, getCodistributor(D2))

Some object methods are now properties:

OldMethodName

blockSize(codistObj) codistObj.BlockSize

defaultBlockSize codistributor2dbc.defaultBlockSize

distributionDimension(codistObj) codistObj.Dimension

distributionPartition(codistObj) codistObj.Partition

labGrid(codistObj) codistObj.LabGrid

New Property Name

Page 14

Parallel Computing Toolbox™ Release Notes

Enhancements to

Admin Center has located menu cho items, etc.

several small enhancements, including more conveniently

ices, m odified dialog boxes, properties dia log box es for listed

Adding or Upda

Admin Center

ting File Dependencies in an Open

MATLAB Pool

Enhancement dependencie

matlabpool('addfiledependencies', filedepCell) matlabpool updatefiledependencies

where file use when ad pool. The changes t

Updated

The glob distrib must pre

ution as its second argument. Because this argument is required, it

stothe

s in a running MATLAB pool. The new forms of the command are

depCell

ding file dependencies to a job or when you open a MATLAB

pdatefiledependencies

oallthelabsinthepool.

globalIndices Function

alIndices

cede the optional argument specifying the lab.

matlabpool command let you add or update file

is a cell array of strings, identical in form to those you

option replicates any file dependency

function now requires that you specify the dimension of

Compatibility Considerations

vious toolbox versions, the

In pre argum the di argum

ent before the dimension argument, and both w ere optional. Now

mension argument is required, and it must precede the optional lab

ent.

globalIndices function accepted the lab

Page 15

Version 4.2 (R2009b) Parallel Computing Toolbox™ Software

Support for Job Templates and Description Files with HPC Server 2008

Using job templates and job description files with Windows HPC Server 2008 lets you specify nodes and other scheduler properties for evaluating your jobs. To support these features, the ccssche du ler object has new properties:

•

ClusterVersion — A string set to 'CCS' or 'HPCServer2008'

• JobTemplate — A string set to the name of the job template to use for

all jobs

•

JobDescriptionFile — A string set to the name of the XML file defining a

base state for job creation

Compatibility Considerations

CCS is now just one of multiple versions of HPC Server. While 'ccs' is still acceptable as a type of scheduler for the use

'hpcserver' for this purpose. In the Configurations Manager, the new

scheduler type is available by selecting File>New>hpcserver(ccs).

findResource function, you can also

HPC Challenge Benchmarks

Several new MATLAB files are availab le to demonstrate HPC Challenge benchmark performance. You can find the files in the folder

matlabroot/toolbox/distcomp/examples/benchmark/hpcchallenge.Each

file is self-documented with explanatory comments.

pctconfig Enhanced to Support Range of Ports

The pctconfig function no w lets you specify a range of ports for the Parallel Computing Toolbox client session to use. This range also includes ports used for a pmode sessio n.

Compatibility Considerations

You now specify the range of pctconfig ports with the 'por tran ge' property; you no longer use the ports of the

'portrange' setting, you no longer use the 'pmodeport' property.

'port' property. As any client pmode session uses tho se

Page 16

Parallel Computing Toolbox™ Release Notes

Random Number Ge

nerator on Client Versus

Workers

The random numbe different seed the client hav

Compatibility Considerations

In past relea number strea the workers a of the clien

ses, while all the workers running a job had s eparate random

r generator of the MATLAB workers now use a slightly

from previous releases, so that all the MATLAB workers and

e separate random number streams.

ms, the client had the same stream as one of the workers. Now

ll have unique random number streams different from that

Page 17

Version 4.1 (R2009a) Parallel Computing Toolbox™ Software

Version 4.1 (R2009a) Parallel Computing Toolbox Software

New Features and Changes

Yes Details below

This table summ

Version Compatibility Considerations

Yes — Details as Compatibi

Considerat

below. See a Summary.

New features and changes introduced in this version are

• “Number of Local Workers Increased to Eight” on page 13

• “Admin Center Allows Controlling of Cluster Resources” on page 1 4

• “Support for Microsoft Windows HPC Server 2008 (CCS v2)” on page 14

• “New Benchmarking Demos” on page 14

• “Pre-R2008b Distributed Array Syntax Now Generates Error” on page 15

• “LSF Support on Mac OS X 10.5.x” on page 15

arizes what is new in Version 4.1 (R2009a):

labeled

lity

ions,

lso

Fixed Bugs an d Known Problems

Bug Reports Includes fixes

Related Documentation at Web Site

Number of Local Workers Increased to Eight

You can now run up to eight local workers on your MATLA B client machine. If you do not specify the number of local workers in a command or configuration, the default number of local workers is determined by the value of the local scheduler’s computational cores on the client machine.

ClusterSize property, which by default is equal to the number of

patibility Considerations

Com

In previous versions, the default number of local workers was four, regardless of the number of cores. If you want to run more local workers than cores (for example, four workers with only one or two cores), you must set the value of

ClusterSize equal to or greater than the number of w orkers you n eed. Then

Page 18

Parallel Computing Toolbox™ Release Notes

you can specify the increased number of workers in the appropriate command or configuration, or let your of workers.

Admin Center Allows Controlling of Cluster Resources

When using the MathW orks job manager, the Admin Center GU I now allow s you to start, stop, and otherwise control job managers and MATLAB workers on your cluster nodes. For more information about Admin Center, see “Admin Center” in the MATLAB Distributed Computing Server documentation.

Compatibility Considerations

You can no longer start Admin Center from the MATLAB Desktop Parallel pull-down m enu. You must start Admin Center from outside MATLAB by executing the following:

ClusterSize setting control the default number

•

matlabroot/toolbox/distcomp/bin/admincenter (on UNIX

systems)

•

matlabroot\toolbox\distcomp\bin\admincenter.bat (on Microsoft

Windows®operating systems)

operating

Support for Microsoft Windows HPC Server 2008 (CCS v2)

The parallel computing products now support Microsoft Windows HPC Server 2008 (CCS v2), including service-oriented architecture (SOA) job submissions. There is no change to the programming interface for options, other than the addition of a new CCS scheduler object property,

UseSOAJobSubmission. For implications to the installation of the MATLAB

Distributed Computing Server, see the online installation instructions at

http://www.mathworks.com/distconfig.

ccs

New Benchmarking Demos

New benchmarking demos for Parallel Computing Toolbox can help you understand and evaluate performance of the parallel computing products. You can access these demos in the Help Browser under the

Toolbox

node: expand the nodes for Demos then Benchmarks.

Parallel Computing

Page 19

Version 4.1 (R2009a) Parallel Computing Toolbox™ Software

Pre-R2008b Dist Generates Error

In R2008b, distr that release, t Now in R2009a, error. For a su for Codistrib

LSF Support o

For availabi contact Pla

http://www

If Platform Parallel C Server can

ibuted array syntax was updated for codistributed arrays. In

he old form of the syntax still worked, but generated a warning.

the old forms of the syntax no longer work and generate an

mmary of the syntax updates, see “Changed Function Names

uted Arrays” on page 19.

nMacOSX10.5.x

lity of Platform LSF

tform Computing Corporation via their Web site at

.platform.com/Products/platform-lsf/technical-information

Computing does not support LSF on Mac OS X 10.5.x, then

omputing Toolbox and MATLAB Distributed Computing

not support this combination.

ributed Array Syntax Now

support on Macintosh®OS X 10.5.x,

Page 20

Parallel Computing Toolbox™ Release Notes

Version 4.0 (R2008b) Parallel Computing Toolbox Software

New Features and Changes

Yes Details below

This table summ

Version Compatibility Considerations

Yes — Details as Compatibi

Considerat

below. See a Summary.

New features and changes introduced in this version are

• “MATLAB

Applications” on page 16

• “spmd Construct” on page 17

• “Composite Objects” on page 18

• “Configuration Validation” on page 18

• “Rerunning Failed Tasks” on page 18

• “Enhanced Job Control with Generic Scheduler Interface” on page 18

• “Changed Function Names for Codistributed Arrays” on page 19

arizes what is new in Version 4.0 (R2008b):

Fixed Bugs an d Known Problems

labeled

lity

ions,

lso

Compiler Product Support for Parallel Computing Toolbox

Bug Reports Includes fixes

Related Documentation at Web Site

• “Determining if a MATLAB Pool is Open” on page 19

MATLAB Compiler Product Support for Parallel Computing Toolbox Applications

This release offers the ability to convert Parallel Computing Toolbox applications, using MATLAB libraries that can access MATLAB Distributed Computing Server. For information on this update to MATLAB Compiler, see “Applications Created with Parallel Computing Toolbox Can Be Compiled”.

Compiler™, into executables and shared

Page 21

Version 4.0 (R2008b) Parallel Computing Toolbox™ Software

Limitations

• MATLAB Compiler does not support configurations that use the local

scheduler or local workers (i.e., workers that run locally on the desktop machine running the MATLAB client session).

• Compiled Parallel Computing Toolbox applications do not support Simulink

software. For a list of other unsupported products, see the Web page http://www.mathworks.com/products/ineligible_programs/.

• When workers are running a task from compiled code, they can execute

only compiled code and toolbox code. They cannot execute functions contained in the current directory. Batch and MATL AB pool jobs attempt to change the worker working directory to the client working directory. When noncompiled files in the current directory conflict with compiled versions (for example, files with different extensions), an error is thrown.

spmd Construct

A new single program multiple data (spmd) language construct allows enhanced interleaving of serial and parallel programming, with interlab communication.

The general form of an

spmd

end

The block of code represented by <statements> executes in parallel o n workers in the MATLAB p oo l. Data on the labs is available for access from the client via Composite objects. For more information, see the page and “Using Distributed Arrays, spmd, and Composites”.

spmd statement is:

spmd reference

Compatibility Considerations

Because spmd is a new keyword, it will conflict with any user-defined functions or variables of the same name. If you have any code with functions or variables named

spmd,youmustrenamethem.

Page 22

Parallel Computing Toolbox™ Release Notes

Composite Objec

Composite objec to data that is st assigned insid a MATLAB pool is client using t spmd, and Comp

Configurati

The Configur configurat by clicking “Validatin

Rerunning

When usin failures properti rerun at

•

• Attemp

, it can attempt to rerun up to a specified number of times. New

es of a task object to control reruns and access information about

tempts are:

Maximum

ts provide direct access from the client (desktop) program

ored on labs in the MATLAB pool. The data of variables

e an spmd b lock is av ai la ble via Composites in the cl ie nt. When

open, you can also create Composites directly from the

Composite function. See also “Using Distribute d Arrays,

osites”.

on Validation

ations Manager is enhanced with the capability for validating

ions. Open the Configurations Manager on the MATLAB Desktop

Parallel > Manage Configurations. For more information, see

g Configurations”.

Failed Tasks

g a job manager, if a task does not complete due to certain system

NumberOfRetries

tedNumberOfRetries

• Failed

Enhan Inter

The ge task gene

•

GetJ

• Des

• Ca

AttemptInformation

ced Job Control with Generic Scheduler

face

neric scheduler interface now allows you to cancel and destroy jobs and

s and to investigate the state of a job. The following new properties of the

ric scheduler object facilitate these features:

obStateFcn

troyJobFcn

troyTaskFcn

ncelJobFcn

Page 23

Version 4.0 (R2008b) Parallel Computing Toolbox™ Software

• CancelTaskFcn

New toolbox functions to accommodate this ability are:

•

getJobSchedulerData

• setJobSchedulerData

For more information on this new functionality, see “Managing Jobs”.

Changed Function Names for Codistributed Arrays

What was kn own in previous releases as distributed arrays are henceforth called codistributed arrays. Some functions related to constructing and accessing codistributed arrays ha ve changed names in this release.

Compatibility Considerations

The following table summarizes the changesinfunctionnamesrelatingto codistributed a rrays. The first three functionsbehaveexactlythesamewith no change in operation, arguments, etc. The

'codistributed' in addition to the array in question.

isa function takes the argument

Old Function Name

distributed codistributed

distributor codistributor

dcolon codcolon

isdistributed isa(X, 'codistributed')

New Function Name

Determining if a MATLAB Pool is Open

The function matlabpool now allows you to discover if a pool of workers is already open. The form of the command is:

matlabpool size

For more information about this option and others, see the matlabpool reference page.

Page 24

Parallel Computing Toolbox™ Release Notes

Version 3.3 (R2008a) Parallel Computing Toolbox Software

New Features and Changes

Yes Details below

This table summ

Version Compatibility Considerations

Yes — Details as Compatibi

Considerat

below. See a Summary.

New features and changes introduced in this version are

• “Renamed Functions for Product Name Changes” on page 20

• “New batch Function” on page 21

• “Enhanced Job Creation Functions” on page 21

• “Increased Data Size Transfers” on page 21

• “Changed Function Names for Distributed Arrays” on page 21

• “Support for PBS Pro and TORQUE Schedulers” on page 22

• “findResource Now Sets Properties According to Configuration” on page 22

arizes what is new in Version 3.3 (R2008a):

labeled

lity

ions,

lso

Fixed Bugs an d Known Problems

Bug Reports Includes fixes

Related Documentation at Web Site

• “parfor Syntax Has Single Usage” on page 23

• “dfeval Now D estroy s Its Job When Finished” on page 23

Renamed Functions for Product Name Changes

As of result of the product name changes, some function names are changing in this release.

patibility Considerations

Com

Two function names are changed to correspond to the new product names:

Page 25

Version 3.3 (R2008a) Parallel Computing Toolbox™ Software

• dctconfig has been renamed pctconfig.

dctRunOnAll has been renamed pctRunOnAll.

•

New batch Function

The new batch function allows you to offload work from the client to one or more workers. The batch submission ca n run scripts that can include jobs that distribute work to other workers. For more information, see the reference page, and “Getting Started” in the Parallel Computing Toolbox User’s Guide.

New Matlabpool Job

The batch functionality is implemented using the new MATLABpool job feature. A MATLAB pool job uses one worker to distribute a job to other workers, thereby freeing the client from the burden of tracking and job’s progress and manipulating data. For more information, see the

createMatlabPoolJob reference page.

batch

Enhanced Job Creation Functions

The createJob and createParallelJo b functions have been enhanced to run without requiring a scheduler object as an argument. This is also true for the new

createMatlabPoolJob function. When a scheduler is not specified, the

function uses the scheduler identified in the applicable parallel configuration. For details, see the reference page for each function.

Increased Data Size Transfers

The default size limitation on data transfers between clients and workers has been significantly increased. In previous releases the default limitation imposed by the JVM memory allocation was approximately 50 MB. The new higher limits are approximately 600 MB on 32-bit systems and 2 GB on 64-bit systems. See “Object Data Size Limitations”.

Changed Fun ction Names for Distributed Arrays

Several functions related to distributed arrays have changed names in this release.

Page 26

Parallel Computing Toolbox™ Release Notes

Compatibility Considerations

The following table summarizes the changesinfunctionnamesrelatingto distributed arrays.

Old Function Name

darray distributed, distributor

distribute distributed

dcolonpartition defaultPartition

distribdim distributionDimension

isdarray isdistributed

labgrid labGrid

local localPart

partition distributionPartition

localspan globalIndices

New Function Name

Support for PBS Pro and TORQUE Schedulers

Parallel Computing Toolbox software now fully supports PBS Pro®and TORQUE schedulers. These schedulers are integrated into parallel configurations and scheduler-related functions like

Note If you do not have a shared file system between client and cluster machines, or if you cannot submit jobs directly to the scheduler from the client machine, any use of third-party schedulers for parallel jobs (including

pmode, matlabpool,andparfor) requires that you use the generic scheduler

interface.

findResource.

findResource Now Sets Properties According to Configuration

The findResouce function now sets the properties on the object it creates according to the configuration identified in the function call.

Page 27

Version 3.3 (R2008a) Parallel Computing Toolbox™ Software

Compatibility Considerations

In past releases, findResource could use a configuration to identify a scheduler, but did not apply the configuration settings to the scheduler object properties. If your code uses separate statements to find an object then set properties, this still works, but is not necessary any more.

parfor Syntax Has Single Usage

The parfor statement is now recognized only for parallel for-loops, not for loops over a distributed range in parallel jobs.

Compatibility Considerations

In R2007b, the pre-existing form of parfor was replaced by for i=

(drange) parfor has only one context, so parfor statements used in parallel jobs in

code fo r versions prior to R2007a must be modified to use

Limitations

, but both forms of syntax were recognized in that release. Now

for (drange).

P-Code Scripts. You can call P-code script files from w ithin a parfor-loop, but P-code script cannot contain a

sim Inside parfor-Loops. Running simulations in a

sim comman d at the top level of the loop is n ot allowed in this release. A sim

command visible in a parfor-loop generates an error, although you can call

sim inside a function that is called from the loop. Be sure that the various labs

running simulations do not have the same working directory, as interference can occur with the simulation data.

parfor-loop.

parfor-loop with the

dfeval Now Destroys Its Job When F inished

When finishe d performing its distributed evaluation, the dfeval function now destroys the job it created.

Compatibility Considerations

If you have any scripts that rely on a job and its data still existing after the completion of no longer work.

dfeval, or that destroy the job after dfeval, these scripts will

Page 28

Parallel Computing Toolbox™ Release Notes

Version 3.2 (R2007b) Distributed Computing Toolbox Software

New Features and Changes

Yes Details below

This table summ

Version Compatibility Considerations

Yes — Details as Compatibi

Considerat

below. See a Summary.

New features and changes introduced in this version are

• “New Parallel for-Loops (parfor-Loops)” on page 24

• “Configurations Manager and Dialogs” on page 25

• “Default Configuration” on page 26

• “Parallel Profiler” on page 26

• “MDCE Script for Red Hat Removed” on page 26

arizes what is new in Version 3.2 (R2007b):

labeled

lity

ions,

lso

Fixed Bugs an d Known Problems

Bug Reports Includes fixes

Related Documentation at Web Site

New Parallel for-Loops (parfor-Loops)

New parallel for-loop (parfor -loop) functionality automatically executes a loop body in parallel on dynamically allocated cluster resources, allowing interleaved serial and parallel code. For details of new parfor functionality, see “Parallel for-Loops (parfor)” in the Distributed Computing Toolbox™ documentation.

itations

Lim

ode Scripts. You can call P-code script files from within a

P-C

P-code script cannot contain a

but

parfor-loop.

parfor-loop,

Page 29

Version 3.2 (R2007b) Distributed Computing Toolbox™ Software

Compatibility Considerations

In past releases, parfor was a different function. The new parfor uses parentheses in defining its range to distinguish it from the old

New

parfor:

parfor (ii = 1:N); <body of code>; end;

Old parfor:

parfor ii = 1:N; <body of code>; end;

For this release, the old form of parfor without parentheses is still supported, although it generates a warning. You can read more about the new form of this existing functionality in “Using a for-Loop Over a Distributed Range (for-drange)”. You should update your existing form of

for-loops over a distributed range (for-drange), thus,

for ii = drange(1:N); <body of code>; end;

parfor code to use the new

parfor.

Configurations Manager and Dialogs

This release introduces a new graphical user interface for creating and modifying user configurations, and for desig n ating the default configuration used by some toolbox functions. For details about the configurations manager, see “Programming with User Configurations” in the Distributed Computing Toolbox documentation.

Compatibility Considerations

This new feature has no impact on how configurations are used in a program, only on how configurations are created and shared among users. In previous versions of the product, you modified your configurations by editing the file

matlabroot/toolbox/distcomp/user/distcompUserConfig.m.Nowthe

configuration data is stored as part of your MATLAB software preferences.

The new configurations manager cannot directly import old-style configurations that were defined in the However, a utility called

importDistcompUserConfig, available on

the MATLAB Central Web site, allows you to convert and import your existing configurations into the new configurations manager.

distcompUserConfig.m file.

Page 30

Parallel Computing Toolbox™ Release Notes

Visit http://www.mathworks.com/matlabcentral and search for

importDistcompUserConfig.

Default Configuration

This version of the toolbox enables you to select a user configuration to use as the default. Thus, commands such as default configuration without y our having to specify it each time you run the command. You can set the default configuration using the configurations graphical interface, or programmatically with the function.

Parallel Profiler

A new parallel profiler graphical user interface generates reports on lab computation and communication times during execution of parallel jobs. For details about this new feature, see “Using the P arallel Profiler”.

MDCE Script for Red Hat Removed

The MDCE script rh_mdce, specific to Red Hat Linux®,hasbeenremoved from

matlabroot/toolbox/distcomp/util/bin.

pmode and matlabpool will use the

defaultParallelConfig

Compatibility Considerations

If you make use of this script, you must replace it w ith its more generic equivalent,

matlabroot/toolbox/distcomp/bin/mdce.

Page 31

Version 3.1 (R2007a) Distributed Computing Toolbox™ Software

Version 3.1 (R2007a) Distributed Computing Toolbox Software

New Features and Changes

Yes Details below

This table summ

Version Compatibility Considerations

Yes — Details as Compatibi

Considerat

below. See a Summary.

New features and changes introduced in this version are

• “Local Scheduler and Workers” on page 27

• “New pmode Interface” on page 28

• “New Default Scheduler for pmode” on page 28

• “Vectorized Task Creation” on page 28

• “Additional Submit and Decode Scripts” on page 29

• “Jobs Property of Job Manager Sorts Jobs by ID” on page 29

• “New Object Display Format” on page 30

arizes what is new in Version 3.1 (R2007a):

labeled

lity

ions,

lso

Fixed Bugs an d Known Problems

Bug Reports Includes fixes

Related Documentation at Web Site

• “Enhanced MATLAB Functions” on page 30

• “darray Function Replaces distributor Function” on page 30

• “rand Seeding Unique for Each Task or Lab” on page 31

• “Single-Threaded Computations on Workers” on page 31

Local Scheduler and Workers

A local scheduler allows you to schedule jobs and run up to four workers or labs on a single MATLAB client machine without requiring engine licenses. These workers/labs can run distributed jobs or parallel jobs, including pmode sessions, for all products for which the MATLAB client is licensed. This

Page 32

Parallel Computing Toolbox™ Release Notes

local scheduler and its workers do not require a job manager or third-party scheduler.

New pmode Interface

The interactive parallel mode (pmode) has a new interface. The pmode command input and displays of the lab outputs are provided in a user interface that you can separate from the MATLAB client Command Window.

Compatibility Considerations

In previous versions of Distributed Computing Toolbox, the pmode interface used the MATLAB Command Window, with the pmode input using a different prompt. The output from the labs was intermingled with the MATLAB client output.

New D e fault Scheduler for pmode

If you start pmode without specifying a configuration,

pmode start

pmode automatically starts a parallel job using the local scheduler with labs running on the client machine. For more information about running pmode, see “Interactive Parallel Computation with pmode” in the Distributed Computing Toolbox d ocumentation.

Compatibility Considerations

In the previous version of the toolbox, when pm ode was started without specifying a configuration, it searched the n etwork for the first available job manager to use as a scheduler.

Vectorized Task Creation

The createTask function can now create a vector of tasks in a single call when you provide a cell array of cell arrays for input arguments. For full details, see the

createTask reference page.

Page 33

Version 3.1 (R2007a) Distributed Computing Toolbox™ Software

Compatibility Considerations

In previous versions of the distribut ed computing products, if your task function had an input argument that was a cell array of cell arrays, your code will need to be modified to run the same way in this release.

For example, your old code may have been written as follows so that the function

In this new version, the same code will produce four tasks. To get the old functionality, you must wrap the four cell arrays in another cell array, so that

createTask knows to create only one task.

myfun gets four cell array input argum ents:

createTask(j, @myfun, 1, {{C1} {C2} {C3} {C4}})

createTask(j, @myfun, 1, { {{C1} {C2} {C3} {C4}} })

Additional Submit and Decode Scripts

There are several submit and decode functions provided with the to olbox for your use with the generic scheduler interface. These files are in the directory

matlabroot/toolbox/distcomp/examples/integration

This version of the toolbox includes new subdirectories for Platform LSF and PBS, to support network configurations in which the client and worker computers do not share a file system. For more information, see “Supplied Submit and Decode Functions” in the Distributed Computing Toolbox documentation.

Jobs Property of Job Manager Sorts Jobs by ID

The Jobs property of a job manager o bject now contains the jobs in the order in which they were created, as indicated by the Similarly, the otherwise specified. This change makes job manager behavior consistent with the behavior of third-party schedulers.

findJob function returns jobs sequenced by their ID,unless

Compatibility Considerations

In previous versions of the distributed computing products, when using a job manager, jobs were arranged in the the status of the job.

Jobs property or by findJob according to

ID property of each job.

Page 34

Parallel Computing Toolbox™ Release Notes

New Object Displ

When you create d without a semico displayed in a n

display functi

command line.

istributed computing objects (scheduler, job, or task)

lon at the end of the command, the object information is

ew format. This new format is also shown when you use the

on to view an object or simply type the object name at the

ay Format

Compatibility Considerations

With this enh has changed.

Enhanced MA

Several MA

•

cat

• find

• horzcat

• subsind

• vertcat

For a co

buted arrays, see “Using MATLAB Functions on Codistributed Arrays”

distri in the D

ancement, the output format shown when creating an object

TLAB Functions

TLAB functions have been enhanced to work on distributed arrays:

mplete list of MATLAB functions that are enhanced to work on

istributed Computing Toolbox documentation.

darra

The fu in a p

y Function Replaces distributor Function

nction

arallel job.

darray now defines how an array is distributed among the labs

Compatibility Considerations

he previous version of the toolbox, the

In t

ine how an array was distributed. In many cases, you can replace a call to

def

tributor

dis

hout arguments as an input to an array constructor,

wit

with a call to darray. For example, if you used distributor

distributor function was used to

Page 35

Version 3.1 (R2007a) Distributed Computing Toolbox™ Software

rand(m, n, distributor());

youcanupdatethecodetoread,

rand(m, n, darray());

rand Seeding Unique for Each Task or Lab

The random generator seed is now initialized based on the task ID for distributed jobs, or the ensures that the set of random numbers generated for each task or lab within a job is unique, even when you have more than 82 tasks or labs.

Compatibility Considerations

In the previous version of the distributed computing products, the rand function could by default generate the same set of numbers for some tasks or labs when these exceeded 82 for a job.

labindex for parallel jobs (including pmode). This

Single-Threaded Computations on Workers

Despite the ability in MATLAB software to perform multithreaded computations on multiple-CPU machines, the workers and labs running distributed and parallel jobs perform only single-threaded computations, so that multiprocessor cluster machines can better accommodate multiple workers or labs.

Page 36

Parallel Computing Toolbox™ Release Notes

Version 3.0 (R2006b) Distributed Computing Toolbox Software

New Features and Changes

Yes Details below

This table summ

Version Compatibility Considerations

Yes — Details as Compatibi

Considerat

below. See a Summary.

New features and changes introduced in this version are

• “Support for Windows Compute Cluster Server (CCS)” on page 32

• “Windows 64 Support” on page 33

• “Parallel Job Enhancements” on page 33

• “Distributed Arrays” on page 33

• “Interactive Parallel Mode (pmode)” on page 34

• “Moved MDCE Control Scripts” on page 34

• “rand Seeding Unique for Each Task or Lab” on page 35

arizes what is new in Version 3.0 (R2006b):

labeled

lity

ions,

lso

Fixed Bugs an d Known Problems

Bug Reports Includes fixes

Related Documentation at Web Site

• “Task ID Property Now Same as labindex” on page 36

Support for Windows Compute Cluster Server (CCS)

Distributed Computing Toolbox software and MATLAB®Distributed Computing Engine™ software now let you program jobs and run them on a Windows Compute Cluster Server. For information about programming in the toolbox to use Windows Compute Cluster Server (CCS) as your scheduler, see the Server Scheduler”.

findResource reference page, and see also “Find a Windows HPC

Page 37

Version 3.0 (R2006b) Distributed Computing Toolbox™ Software

Windows 64 Suppo

The distributed both MATLAB clie

Parallel Job E

computing products now support Windows 64 (Win64) for

nt and MATLAB worker machines.

nhancements

Parallel Jobs Support Any Scheduler

Support for p releases, on supported p by a job mana LSF, mpiexe informatio

arallel jobs now extends to any type of scheduler. In previous ly the M athW orks job manager and mpiexec scheduler object arallel jobs. You can now run parallel jobs on clusters scheduled

ger, Windows Compute Cluster S erver (CCS), Platform

c, or using the generic scheduler interface. For programming

n, see “Programming Parallel Jobs”.

New labSendReceive Function

The labSe performs reduces t simulta see the

neously rather than by separate statements. For more information,

ndReceive

thesamethingsasboth

he risk of deadlock, because the send and receive happen

abSendReceive

function is introduced in this release. This function

labSend and labReceive,butgreatly

reference page.

Improved Error Detection

This re labs ru detec

lease o ffers improved error detection for miscommunication between nning parallel jobs. Most notable among the improvements are error

tion of mismatched

labSend and labReceive statements.

Distr

Dist in th to wo and p esp Cod

ibuted Arrays

ributed arrays are partitioned into segments, with each segment residing

e workspace of a different lab, so that each lab has its o wn array segment

rk with. Reducing the size of the array that each lab has to store rocess means a more efficient use of memory and faster processing,

ecially for large data sets. For more information, see “Working with

istributed Arrays”.

Page 38

Parallel Computing Toolbox™ Release Notes

There are many new a nd enhanced MATLAB functions to work with distributed arrays in parallel jobs. For a listing of these functions and their reference pages, see “Job Management”.

parfor: Parallel for-Loops

Parallel for-loops let you run a for-loop across your labs simultaneously. For more information, see “Using a for-Loop Over a Distributed Range (for-drange)” or the

Interactive Parallel Mode (pmode)

The interactive parallel mode (pmode) lets you work interactively with a parallel job running simultaneously on a number of labs. Commands you type at the pmode command line are executed on all labs at the same time. Each lab executes the commands in its own workspace on its own local variables or segments of distributed arrays. For more information, see “Getting Started with pmode”.

parfor reference page.

Moved MDCE Control Scripts

To provide greater consistency across all platforms, the MDCE control scripts for Windows have moved and those for UNIX and Macintosh have new names.

Compatibility Considerations

Windows Utilities Moved. In previous versions of the distributed

computing products, the MDCE utilities for Windows computers were located in

matlabroot\toolbox\distcomp\bin\win32

The utilities are now located in

matlabroot\toolbox\distcomp\bin

The files that have moved are

nodestatus mdce startjobmanager

Page 39

Version 3.0 (R2006b) Distributed Computing Toolbox™ Software

stopjobmanager startworker stopworker mdce_def.bat

UNIX and Macintosh Utilities Renamed. In previous versions of the distributed computing products, the MDCE utilities for UNIX and Macintosh computers were called by

nodestatus.sh startjobmanager.sh stopjobmanager.sh startworker.sh stopworker.sh

You can now call these with the following commands:

nodestatus startjobmanager stopjobmanager startworker stopworker

Note For UNIX and Macintosh, mdce and mdce_def.sh have not been moved or renamed.

rand Seeding Unique for Each Task or Lab

The random generator seed is now initialized based on the task ID for distributed jobs, or the ensures that the random numbers generated for each task or lab are unique within a job.

Compatibility Considerations

In previous versions of the distributed computing products, the rand function would by default generate the same set of numbers on each worker.

labindex for parallel jobs (including pmode). This

Page 40

Parallel Computing Toolbox™ Release Notes

Task ID Property

Although you cre task for each w or four workers (l The first task i lab whose and

labi

labindex f

the sequence o corresponds

Tasks proper

ate only one task for a parallel job, the system copies this

ker that runs the job. For example, if a parallel job runs on abs), the nthejob’s

ndex

or the lab that ran that task have the same value. Therefore,

f results returned by the

to the value of

ty.

Now Same as labindex

Tasks property of the job contains four task objects.

Tasks property corresponds to the task run by the

is 1, and so on, so that the ID property for the task object

labindex and to the order of tasks in the job’s

Compatibility Considerations

In past rele

ID propert

ases, there was no correlation between

getAllOutputArguments function

labindex and the task

Page 41

Compatibility Summary for Parallel Computing Toolbox™ Software

Compatibility Summary for Parallel Computing Toolbox Software

This table summarizes new features and changes that might cause incompatibilities when you upgrade from an earlier version, or wh en you use files on multiple versions. Details are provided with the description of the new feature or change.

Version (Rele ase) New Features and Ch a n g es with Version

Compatibility Impact

Latest Version V4.3 (R2010a)

V4.2 (R2009b) See the Compatibility Considerations

See the Compatibility Considerations subheading for this change:

• “taskFinish File for MATLAB Pool” on page 5

• “Upgrade Parallel Computing Products

Together” on page 5

subheading for each of these new features or changes:

• “Renamed codistributor Functions” on page 8

• “Updated globalIndices Function” on page 10

• “Support for Job Templates and Description

Files with HPC Server 2008” on page 11

• “pctconfig Enhanced to Support Range of

Ports” on page 11

• “Random Number Generator on Client Versus

Workers” on page 12

Page 42

Parallel Computing Toolbox™ Release Notes

Version (Rele ase) New Features and Ch a n g es with Version

V4.1 (R2009a) See the Compatibility Considerations

V4.0 (R2008b) See the Compatibility Considerations

Compatibility Impact

subheading for each of these new features or changes:

• “Number of Local Workers Increased to Eight”

on page 13

• “Admin Center Allows Controllin g of Cluster

Resources” on page 14

subheading for each of these new features or changes:

• “spmd Construct” on page 17

• “Changed Function Names for Codistributed

Arrays” on page 19

V3.3 (R2008a) See the Compatibility Considerations

subheading for each of these new features or changes:

• “Renamed Functions for Product Name

Changes” on page 20

• “Changed Function Names for Distributed

Arrays” on page 21

• “findResource Now Sets Properties According

to Configuration” on page 22

• “parfor Syntax Has Single Usage” on page 23

• “dfeval Now Destroys Its Job When Finished”

on page 23

Page 43

Compatibility Summary for Parallel Computing Toolbox™ Software

Version (Rele ase) New Features and Ch a n g es with Version

Compatibility Impact

V3.2 (R2007b) See the Compatibility Considerations

subheading for each of these new features or changes:

• “New Parallel for-Loops (parfor-Loops)” on

page 24

• “Configurations Manager and Dialogs” on

page 25

• “MDCE Script for Red Hat Removed” on page

V3.1 (R2007a) See the Compatibility Considerations

subheading for each of these new features or changes:

• “New pmode Interface” on page 28

• “New Default Scheduler for pmode” on page

• “Vectorized Task Creation” on page 28

• “Jobs Property of Job Manager Sorts Jobs by

ID” on page 29

• “New Object Display Format” on page 30

• “darray Function Replaces distributor

Function” on page 30

• “rand Seeding Unique for Each Task or Lab”

on page 31

V3.0 (R2006b) See the Compatibility Considerations

subheading for each of these new features or changes:

• “Moved MDCE Control Scripts” on page 34

• “rand Seeding Unique for Each Task or Lab”

on page 35