The Windows HPC Team Blog : Calvin's notes from the field

Note to the following posts

MichaelCole — Wed, 16 Jan 2008 14:35:00 GMT

The following posts are exact copies of Calvin's entries on the previous Windowshpc portal site. I edited the posting dates to reflect the original date of the entry.

Michael Cole
Project Manager/Site Administrator

Cleaning up CCS perf counter entries

MichaelCole — Fri, 12 Oct 2007 10:40:00 GMT

Just a quick note. When you uninstall CCP V1, the performance counter categories "Compute Cluster" and "Compute Nodes" aren't removed. (This is a bug which will be resolved in V2.) This means that if for some reason these entries get corrupted, they can be hard to clean up. I encountered this because of an unrelated bug where the performance counter names for the Japanese CCP were localized.

Corrupt entries can cause errors like:

C:\>cluscfg view
Counter 'Total number of jobs' does not exist in the specified Category.

Another symptom is a dialog when trying to submit a job from the job scheduler that says: "Total processor number is zero. Cannot submit job."

In any case, the following PowerShell script can be used to clean up these entries after uninstalling the CCP. (Don't run it while the CCP is still installed! In V1, you can technically stop the Microsoft Compute Cluster Scheduler Service, run this script and then restart the service and it will re-populate the counters, but if you have corruption and don't know how it get there, it might be best to just re-install the CCP.)

$objperfcats = [System.Diagnostics.PerformanceCounterCategory]
$objperfcats::GetCategories() | where-object {$_.CategoryName -eq "Compute Cluster"
-or $_.CategoryName -eq "Compute Nodes" } | foreach { $objperfcats::Delete($_.CategoryName) }

Check out Ryan Waite's interview on the "How Software Is Built" blog.

MichaelCole — Mon, 06 Aug 2007 10:00:00 GMT

http://howsoftwareisbuilt.com/2007/07/22/interview-with-ryan-waite/

In this interview, Ryan Waite, Group Program Manager for High-Performance Computing at Microsoft talks about:

o Microsoft’s entry into the High Performance Computing area.

o Microsoft’s support of the open source Message Passing Interface (MPI).

o MPI has become bloated, and there’s no usability data (like there is with Microsoft Office) to indicate what should be deprecated.

o At Microsoft, select customers help shape the features for the next product release.

o Microsoft has always focused on ease of use, simplified administration, and uncomplicated setup. HPC is no different.

o Instrumentation helps find reliability issues. If partners are affecting reliability (with unstable drivers, for example) certification helps raise the bar.

o Parallel programming is now the only way to do more computation in the same amount of time. Processors just are getting faster like they used to.

o It’s easier for open source to be incremental. Proprietary software has to make radical changes.

o Hard drives + overnight shipping = really high bandwidth.

Using "clusrun" to deploy PowerShell, VC runtimes, or other goodies to your compute nodes

MichaelCole — Fri, 06 Jul 2007 13:30:00 GMT

Normally, deploying applications to a large number of Windows machines requires substantial planning and infrastructure. Microsoft ships enterprise-quality solutions for handling software deployments, such as Systems Management Server (SMS). It is also possible to deploy software packages using Active Directory Group Policy. Critical security updates can be handled using Windows Service Update Services (WSUS). As your compute cluster infrastructure becomes larger and more complex, these solutions can and should be considered. However, for the majority of test clusters and even some larger installations, there is a quicker way to get software to your compute nodes, if you can figure out how to do an unattended setup of the software from the command line.

Windows PowerShell is an excellent example of a program which can be deployed easily using a command-line setup. If you download the PowerShell 1.0 x64 version (which is appropriate for Windows Compute Clusters), you will get an executable named something like "WindowsServer2003.WindowsXP-KB926139-x64-ENU.exe". If you run:

WindowsServer2003.WindowsXP-KB926139-x64-ENU.exe /?

you will get a dialog with a usage message. The key thing to note is that this executable supports a "/quiet" flag which allows you to install without generating an interactive UI. So to install PowerShell on all of your compute nodes automatically, you can simply copy the setup program above to a share that is accessible to the compute nodes (for example, on the head node "head", share name "share" and run:

C:\> clusrun \\head\share\WindowsServer2003.WindowsXP-KB926139-x64-ENU.exe /quiet

That's it!

Some applications require extra steps to get to the unattended install, so you have to do some digging. For example, let's say you have an application built with SP1 of Visual Studio 2005. This uses a newer version of the C Runtime and you'll have to get this on all of your compute nodes in order for your application to run. Otherwise you'll get cryptic errors. You can download the setup program for this update C Runtime, and you'll get something named "vcredist_x64.exe". If you run "vcredist_x64.exe /?" you'll get the options that you'll need to perform an intermediate step before performing your remote deployment. The key is to use the /T: and /C options:

C:\> vcredist_x64.exe /T:\\head\share\vcredist /C

This will create a directory \\head\share\vcredist with a single file in it named something like "VCREDI~2.EXE". So there is a second indirection, but we are almost done! Do:

C:\> \\head\share\vcredist\VCREDI~2.EXE /T:\\head\share\vcredist2 /C

This will extract the final setup files into \\head\share\vcredist2

Within this directory you will find a file named "vcredist.msi". This is a Windows Installer Setup file. It can be invoked via "msiexec" (the default program for the .msi extension.). The trick to performing a quiet MSI install is to use the /qn option. So the final command to get this VC Runtime version out to your compute nodes would look like this:

C:\> clusrun msiexec /qn /i \\head\share\vcredist2\vcredist.msi

Good luck!

An Introduction

MichaelCole — Sat, 07 Apr 2007 12:00:00 GMT

I've been at Microsoft since 1996 in various roles on product teams and in consulting. These days I work pretty much full time assisting HPC ISVs who are trying to get their applications onto Windows Compute Cluster Server. Recently, I've also begun to work with some solution integrators to assist them in preparing and testing clusters for deployment. The Microsoft HPC product team has graciously invited me to share some of my experiences in this blog. I'll try to keep the short bits in the main forums and the more long-winded stuff here.

Microsoft's entry into this space is recent, but the business of high performance computing has a long history. I met a consultant recently whose business card boasted: "34 years of HPC experience." I found it simultaneously amusing and daunting. In some sense, that 34-year claim certainly can be valid, but the landscape of what we call high performance computing has changed dramatically in 34 years.

My goal is to share some common threads, challenges, and hopefully a few solutions based that I'm seeing in the field, so I probably won't mention something unless I think it's important to quite a few of our partners. I look forward to your feedback!

How to detect CCS

MichaelCole — Fri, 06 Apr 2007 13:28:00 GMT

Because ISVs have to support solutions from multiple vendors, I frequently get asked: "How do I know if I am running on CCS?" It's a good question. Also, as future versions of the product get shipped, it will be increasingly important to be able to tell which version of CCS is installed, so that setup programs can be act accordingly.

There are various "signs" that the Windows Compute Cluster Pack might be installed. Certain binaries might be in place, environment variables might be set, there are registry keys you can find. However, I discourage developers from depending on these things, not necessarily because they might change in future versions, but because looking for those things is specific to just this one product. Proper setup programs should use reusable code to detect the presence, absence, or version information for features. If you use the techniques I'm outlining here, they can easily be modified to detect dependencies on other products which use a Microsoft Installer compatible installation, which consistutes the vast majority of Windows applications. (In fact, it's a Windows Logo requirement. See http://www.microsoft.com/windowsserver2003/partners/isvs/appspec.mspx for details.)

In this posting I'm going to introduce the use of Microsoft Installer APIs that you can use get information about your CCS installation. MSI defines the concepts of "Programs", "Features", and "Components." Perhaps the best way to get started is to download the Windows SDK at http://www.microsoft.com/downloads/details.aspx?FamilyId=C2B1E300-F358-4523-B479-F53D234CDCCF&displaylang=en. This includes a VB script named "WiLstPrd.vbs" which you can use to see how MSI APIs can be called to determine what's installed. (It will get installed in C:\Program Files\Microsoft SDKs\Windows\v6.0\Samples\SysMgmt\MSI\scripts. If you just want to try the script but don't want to wait for the SDK to dowload, you can get it and the code mentioned here in the files repository from this site: http://windowshpc.net/files/17/scripting/entry1138.aspx.)

For example, if I run this on the head node in a test cluster, I get:

C:\Documents and Settings\caclark\test>cscript WiLstPrd.vbs
Microsoft (R) Windows Script Host Version 5.6
Copyright (C) Microsoft Corporation 1996-2001. All rights reserved.

{3F7A6441-44A4-41C0-9A0A-294629D49208} = Microsoft Office Live Meeting 2005 7.9.2107
{5ADF3AE1-035F-4366-A554-27EF55449BBB} = Debugging Tools for Windows 64-bit 6.7.4
{A9CF9052-F4A0-475D-A00F-A8388C62DD63} = MSXML 4.0 SP2 (KB925672) 4.20.9839
{E95625F4-36FD-4BE0-8018-438BA5AC50AC} = SMS Advanced Client 4.0.5571
{37477865-A3F1-4772-AD43-AAFC6BCFF99F} = MSXML 4.0 SP2 (KB927978) 4.20.9841
{E09B48B5-E141-427A-AB0C-D3605127224A} = Microsoft SQL Server Desktop Engine 8.0.761
{0D8EA686-5DE3-4816-BEAE-02331D1D5DA2} = MSXML 6.0 Parser (KB927977) 6.0.3890
{E33A28C7-5017-4F83-8B19-1A3A480392DE} = F1 Profiler 8.0.0
{B6EC01E7-431D-4D29-B9D4-E1D74CAF0AB0} = Microsoft .NET Framework 2.0 (x64) 2.0.50727
{9A0ED01E-FD18-457A-AB9C-0835DCDB17BB} = Microsoft Platform SDK (R2) (3790.2075) 5.2.3790
{01493E6E-2473-4DE5-963B-BF17BACC21C3} = Microsoft Compute Cluster Pack 1.0.676
{1BBC7AAE-2DA9-40C2-B903-8E7A5EFEECF2} = CA eTrust Antivirus 7.1.192

Note that each product has a unique GUID. From this I can tell that the Microsoft Compute Cluster Pack is installed, but I know that there are many pieces to CCP, such as the binaries required for a compute node, the features for a headnode such as the MSDE database, and the client tools which include the Compute Cluster Administrator and Compute Cluster Job Scheduler UIs. With this script, it's possible to drill down further into a particluar product:

C:\Documents and Settings\caclark\test>cscript WiLstPrd.vbs /?
Microsoft (R) Windows Script Host Version 5.6
Copyright (C) Microsoft Corporation 1996-2001. All rights reserved.

Windows Installer utility to list registered products and product information
Lists all installed and advertised products if no arguments are specified
Else 1st argument is a product name (case-insensitive) or product ID (GUID)
If 2nd argument is missing or contains 'p', then product properties are listed
If 2nd argument contains 'f', features, parents, & installed states are listed
If 2nd argument contains 'c', installed components for that product are listed
If 2nd argument contains 'd', HKLM "SharedDlls" count for key files are listed

C:\Documents and Settings\caclark\test>cscript WiLstPrd.vbs "Microsoft Compute Cluster Pack" fcd
Microsoft (R) Windows Script Host Version 5.6
Copyright (C) Microsoft Corporation 1996-2001. All rights reserved.

---Features in product {01493E6E-2473-4DE5-963B-BF17BACC21C3}---
HeadNode = Local
ClientUtils = Local
ComputeNode = Uninstalled
---Components in product {01493E6E-2473-4DE5-963B-BF17BACC21C3}---
{7D708050-87A3-4F80-8965-EDD137C27EE0} = C:\Program Files\Microsoft Compute Cluster Pack\Bin\Microsoft.Ccp.Sdm.Core.dll
{6763FD90-2629-4796-A273-14B1C49AFD02} = 20:\Microsoft.ComputeCluster.Cluster\
{21B17FE0-7C7E-4B63-BD86-B15C958B8BD7} = C:\Program Files\Microsoft Compute Cluster Pack\Bin\CcpConsole.exe
{05ED0D11-C198-4374-A814-F7BB1C5A5D54} = 02:\SOFTWARE\Microsoft\CCP\
{043E5A31-7BA6-4A35-9221-9B5000967C05} = C:\Program Files\Microsoft Compute Cluster Pack\Bin\CcpConfiguration.exe
{9DD2E181-54EA-4135-8DD2-4C7CBAF9028D} = C:\Program Files\Microsoft Compute Cluster Pack\LogFiles\
{4421EA91-45D2-40DF-8E5C-FDD617B905A0} = C:\Program Files\Microsoft Compute Cluster Pack\Bin\Microsoft.Ccp.ComputerModel.dll
{FB3362D1-9E40-4522-8D86-4B72BCD4D808} = C:\Program Files\Microsoft Compute Cluster Pack\Bin\CcpSchedulerCore.dll
{C42CD672-C04B-417C-BF7C-B0E9BB988404} = C:\Program Files\Microsoft Compute Cluster Pack\Bin\CcpActivate.dll
{A7F4CF72-0F7F-4405-8BA0-84555D06713F} = C:\Program Files\Microsoft Compute Cluster Pack\Bin\Microsoft.WizardFramework.dll
{EA44E123-CE4B-406C-BAD2-0DCC9079EF6A} = C:\Program Files\Microsoft Compute Cluster Pack\Bin\mpiexec.exe
{99109423-B2E0-4B15-9C46-F2A38A69AAE4} = C:\Program Files\Microsoft Compute Cluster Pack\Bin\Microsoft.Ccp.ManagementTasks.dll
{F105F063-18B2-430C-8C2F-FAC332F92573} = C:\Program Files\Microsoft Compute Cluster Pack\Bin\Microsoft.Ccp.RisModel.dll
{83B24CC3-D0F0-4504-92D2-29ABEF715C2C} = C:\Program Files\Microsoft Compute Cluster Pack\Bin\CcpManagement.msc
{A2447534-39F0-442F-8E43-1C15C933C6D5} = C:\Program Files\Microsoft Compute Cluster Pack\Conf\Microsoft.Ccp.ComputerModel.sdmDocument
{B18ECDB4-00CF-4896-96E2-ABAA30DF71F6} = C:\Program Files\Microsoft Compute Cluster Pack\Bin\SdmL.exe
{DAECCA05-F483-4BED-93BA-148D3C44179F} = C:\Program Files\Microsoft Compute Cluster Pack\Bin\CcpSdm.exe
{0F3D19F5-D8DE-46A2-BF78-F01E1F6295DA} = C:\Program Files\Microsoft Compute Cluster Pack\Bin\Microsoft.Ccp.NetworkModel.dll
{5BBB32A6-C08F-490F-A4FC-42F5155AE7AF} = C:\Program Files\Microsoft Compute Cluster Pack\Bin\Microsoft.Ccp.ClusterModel.dll
{5A7F74D9-56D4-4B61-B013-FD420754B92B} = 02:\SYSTEM\CurrentControlSet\Services\CcpScheduler\Parameters\spooldir
{8968955B-0047-42F6-9BE4-6AA453E55269} = C:\Program Files\Microsoft Compute Cluster Pack\Bin\job.vbs
{78CE729B-3479-4BBB-ABA8-433B8D9D2C84} = C:\Program Files\Microsoft Compute Cluster Pack\Bin\
{20DC0BCB-7DA4-4459-BAAF-4622B71345A8} =
{FEF27D3C-34C6-4BE8-BC5F-77DE20F0394B} = C:\Program Files\Microsoft Compute Cluster Pack\Bin\
{02819C4C-6DCB-4751-A410-410C6605DEE1} = C:\Program Files\Microsoft Compute Cluster Pack\Bin\CcpScheduler.exe
{28E9CF5C-540A-4916-B42B-15A0F6020278} = C:\Program Files\Microsoft Compute Cluster Pack\Bin\
{A1D0DD7C-ADEE-4993-A8D9-33C8EC832D25} = C:\Program Files\Microsoft Compute Cluster Pack\support.txt
{8C34B12D-FB2C-4A17-9C92-63BCEAC5F38F} = 22:\SOFTWARE\Microsoft\MMC\SnapIns\FX:{f0c64d10-d537-431e-bd72-8c55912a5611}\
{9362823D-E33C-4A95-91D5-9DF6888CE2FA} =
{D790FA5D-4E38-488C-B850-C255760799C5} = C:\Program Files\Microsoft Compute Cluster Pack\Bin\Microsoft.Ccp.Sdm.SystemModel.dll
{61BD4B5D-B4E3-4DDD-92AA-A89C95AA9A7D} = C:\Program Files\Microsoft Compute Cluster Pack\Bin\Microsoft.Ccp.Sdm.Store.dll
{E77254AD-CAE5-43BB-ADBD-ACCBA9A41EA0} = C:\Program Files\Microsoft Compute Cluster Pack\Bin\Microsoft.Ccp.ClusterManagementInterfaces.dll
{9C8609ED-CEF4-4AD6-99ED-6AC398B5A99F} = C:\WINDOWS\assembly\GAC_MSIL\CcpAPI\1.0.2.3__31bf3856ad364e35\CcpAPI.dll
{1230911E-DF5F-4B01-AFB6-21DC9DFA1DCD} = C:\Program Files\Microsoft Compute Cluster Pack\InstallShare\RISInstall.cmd
{C338DD3E-E493-4D86-93B5-46EE5ADDE1F5} = C:\Program Files\Microsoft Compute Cluster Pack\Bin\Microsoft.Ccp.Sdm.Manager.dll
{89061D2F-CCB5-47D3-A1A9-125979A7E1EB} = C:\Program Files\Microsoft Compute Cluster Pack\License\EULA.RTF
{F55F8B7F-34CB-4742-8909-80803A5BF052} = C:\Program Files\Microsoft Compute Cluster Pack\Bin\ToDoList.exe
---Shared DLL counts for key files of {01493E6E-2473-4DE5-963B-BF17BACC21C3}---
0 = C:\Program Files\Microsoft Compute Cluster Pack\Bin\Microsoft.Ccp.Sdm.Core.dll
0 = C:\Program Files\Microsoft Compute Cluster Pack\Bin\CcpConsole.exe
0 = C:\Program Files\Microsoft Compute Cluster Pack\Bin\CcpConfiguration.exe
0 = C:\Program Files\Microsoft Compute Cluster Pack\LogFiles\
0 = C:\Program Files\Microsoft Compute Cluster Pack\Bin\Microsoft.Ccp.ComputerModel.dll
0 = C:\Program Files\Microsoft Compute Cluster Pack\Bin\CcpSchedulerCore.dll
0 = C:\Program Files\Microsoft Compute Cluster Pack\Bin\CcpActivate.dll
0 = C:\Program Files\Microsoft Compute Cluster Pack\Bin\Microsoft.WizardFramework.dll
0 = C:\Program Files\Microsoft Compute Cluster Pack\Bin\mpiexec.exe
0 = C:\Program Files\Microsoft Compute Cluster Pack\Bin\Microsoft.Ccp.ManagementTasks.dll
0 = C:\Program Files\Microsoft Compute Cluster Pack\Bin\Microsoft.Ccp.RisModel.dll
0 = C:\Program Files\Microsoft Compute Cluster Pack\Bin\CcpManagement.msc
0 = C:\Program Files\Microsoft Compute Cluster Pack\Conf\Microsoft.Ccp.ComputerModel.sdmDocument
0 = C:\Program Files\Microsoft Compute Cluster Pack\Bin\SdmL.exe
0 = C:\Program Files\Microsoft Compute Cluster Pack\Bin\CcpSdm.exe
0 = C:\Program Files\Microsoft Compute Cluster Pack\Bin\Microsoft.Ccp.NetworkModel.dll
0 = C:\Program Files\Microsoft Compute Cluster Pack\Bin\Microsoft.Ccp.ClusterModel.dll
0 = C:\Program Files\Microsoft Compute Cluster Pack\Bin\job.vbs
0 = C:\Program Files\Microsoft Compute Cluster Pack\Bin\
0 = C:\Program Files\Microsoft Compute Cluster Pack\Bin\
0 = C:\Program Files\Microsoft Compute Cluster Pack\Bin\CcpScheduler.exe
0 = C:\Program Files\Microsoft Compute Cluster Pack\Bin\
0 = C:\Program Files\Microsoft Compute Cluster Pack\support.txt
0 = C:\Program Files\Microsoft Compute Cluster Pack\Bin\Microsoft.Ccp.Sdm.SystemModel.dll
0 = C:\Program Files\Microsoft Compute Cluster Pack\Bin\Microsoft.Ccp.Sdm.Store.dll
0 = C:\Program Files\Microsoft Compute Cluster Pack\Bin\Microsoft.Ccp.ClusterManagementInterfaces.dll
0 = C:\WINDOWS\assembly\GAC_MSIL\CcpAPI\1.0.2.3__31bf3856ad364e35\CcpAPI.dll
0 = C:\Program Files\Microsoft Compute Cluster Pack\InstallShare\RISInstall.cmd
0 = C:\Program Files\Microsoft Compute Cluster Pack\Bin\Microsoft.Ccp.Sdm.Manager.dll
0 = C:\Program Files\Microsoft Compute Cluster Pack\License\EULA.RTF
0 = C:\Program Files\Microsoft Compute Cluster Pack\Bin\ToDoList.exe

So you can see that the MSI APIs are pretty powerful, but if you have an exsiting C++ program, it can take some time to translate the appropriate pieces of code from the VBScript in WiLstPrd.vbs into C++, especially if you are not used to writing COM code on Windows. If you are only doing basic checks, you can use the MSI APIs described on

MSDN here:

http://msdn2.microsoft.com/en-us/library/aa369426.aspx

Here is a short sample in C++ which shows how you get some basic information:

// cccver.cpp: Get basic info on CCS components installed on local machine.

#define UNICODE 1

#include <windows.h>

#include <msi.h>

#include <stdio.h>

#include <assert.h>

#define CCP_PRODUCT_CODE L"{01493E6E-2473-4DE5-963B-BF17BACC21C3}" // GUID valid Microsoft Compute Cluster Pack

int wmain(int argc, wchar_t **argv)

{

WCHAR wszValueBuf[BUFSIZ];

DWORD dwBufsize = BUFSIZ;

/* It it possible enumerate all of the products on the system with MsiEnumProducts() and check for those with the name

"Microsoft Compute Cluster Pack" but this is not required. */

UINT iRet = MsiGetProductInfo(CCP_PRODUCT_CODE,

INSTALLPROPERTY_VERSIONSTRING,

wszValueBuf,

&dwBufsize);

switch(iRet) {

case ERROR_SUCCESS:

wprintf(L"Microsoft Compute Cluster Pack Version %s is installed\n", wszValueBuf);

dwBufsize = BUFSIZ;

/* We now know that the current machine has at least some components installed, but which ones?

to find out, query the features. */

WCHAR wszFeatureBuf[MAX_FEATURE_CHARS+1];

WCHAR wszParentBuf[MAX_FEATURE_CHARS+1];

for (DWORD i = 0; !MsiEnumFeatures(CCP_PRODUCT_CODE, i, wszFeatureBuf, wszParentBuf); i++) {

INSTALLSTATE featurestate = MsiQueryFeatureState(CCP_PRODUCT_CODE, wszFeatureBuf);

wprintf(L"%s:", wszFeatureBuf);

switch(featurestate) {

case INSTALLSTATE_ABSENT:

wprintf(L" absent\n");

break;

case INSTALLSTATE_ADVERTISED:

wprintf(L" advertised\n");

break;

case INSTALLSTATE_SOURCE:

wprintf(L" installed to run from source\n");

break;

case INSTALLSTATE_UNKNOWN:

wprintf(L" unknown\n");

break;

case INSTALLSTATE_LOCAL:

wprintf(L" installed locally\n");

break;

case INSTALLSTATE_INVALIDARG:

assert(0);

break;

}

break;

case ERROR_UNKNOWN_PRODUCT:

wprintf(L"Microsoft Compute Cluster Pack is not installed\n");

break;

default:

wprintf(L"Error 0x%x\n", iRet);

return 1;

}

/* Check for version of MSMPI.DLL component if present */

WCHAR wszVersion[BUFSIZ];

DWORD dwVersize = BUFSIZ;

WCHAR wszSystem[MAX_PATH+1];

WCHAR wszMSMPIPath[MAX_PATH+1];

GetSystemDirectory(wszSystem, MAX_PATH);

swprintf(wszMSMPIPath, L"%s\\msmpi.dll", wszSystem);

if (ERROR_SUCCESS == MsiGetFileVersion(wszMSMPIPath, wszVersion, &dwVersize, 0, 0)) {

wprintf(L"msmpi.dll version %s\n", wszVersion);

}

return 0;

}

To compile this is easy. You just have to link with msi.lib. For example, you could just run the follow from an SDK build window:

C:\temp>cl /I "c:\Program Files\Microsoft SDKs\Windows\v6.0\Include" ccsver.cpp
"c:\Program Files\Microsoft SDKs\Windows\v6.0\Lib\Msi.Lib"

Here's the output if I run this program on all nodes in a cluster using "clusrun"

C:\Documents and Settings\caclark\test>clusrun /nodes:hpca3cn2001 \\hpca3hn02\test\ccsver

-------------------------- HPCA3CN2001 returns 0 --------------------------

Microsoft Compute Cluster Pack Version 1.0.0676.0 is installed

ComputeNode: installed locally

HeadNode: absent

ClientUtils: absent

msmpi.dll version 1.0.676.0

-------------------------- HPCA3CN2002 returns 0 --------------------------

Microsoft Compute Cluster Pack Version 1.0.0676.0 is installed

ComputeNode: installed locally

HeadNode: absent

ClientUtils: absent

msmpi.dll version 1.0.676.0

-------------------------- HPCA3CN2003 returns 0 --------------------------

Microsoft Compute Cluster Pack Version 1.0.0676.0 is installed

ComputeNode: installed locally

HeadNode: absent

ClientUtils: absent

msmpi.dll version 1.0.676.0

-------------------------- HPCA3CN2004 returns 0 --------------------------

Microsoft Compute Cluster Pack Version 1.0.0676.0 is installed

ComputeNode: installed locally

HeadNode: absent

ClientUtils: absent

msmpi.dll version 1.0.676.0

On the head node (which is not configured as a compute node in this cluster), the output is like this:

C:\Documents and Settings\caclark\test>ccsver
Microsoft Compute Cluster Pack Version 1.0.0676.0 is installed
HeadNode: installed locally
ClientUtils: installed locally
ComputeNode: absent

Note that in this case the headnode does not have the "ComputeNode" feature, but it does have the HeadNode feature and the ClientTools feature, whereas the compute node will only have the ComputeNode feature. If I had installed the client tools on some of the compute nodes, it would show that.

No one will this blog entry unless I mention how to find msmpi.dll. On a compute node, this will actually be installed in the system directory, so it's in the path; a normal LoadLibary() call will find it and any programs linked with the msmpi.lib import libary in the CCS SDK will work just fine.

In general, I think it's worth the trouble to add some bits of code to your setup to detect CCS correctly, because some other methods are less reliable. Also, this is perhaps the best way to insure your versions are correct.

I hope this is helpful,

-Calvin

Comments (2)