If you are carrying an HPC deployment where you deploy the operating system by your own means (wds, ghost, true image, etc.) and will run the HPC pack on the nodes, it’s actually quite easy to install the HPC pack on the node w/o logging in at all. I’m assuming your nodes are already part of the domain where the cluster will run.

First, download psexec and place it on c:\windows\system32\ on your headnode. Then, open a command prompt and enter:

psexec -s \\[NODE-NAME] \\hpc-hn\REMINST\setup.exe -unattend -computenode:[CLUSTER-NAME]

Make sure to replace [NODE-NAME] & [CLUSTER-NAME] with your values. That command will take about 1 minute to run and sooner than later, you will have the node showing up on cluster manager waiting for a template, no need to log on the node at all!

As I am documenting a lab, I’m playing with the RC version of Microsoft Windows HPC Server 2008 R2, which you can download from connect.microsoft.com. As I am documenting some of the features, I have noticed a lot of things have changed in the GUI, and thought about documenting them here since these minor details will probably will not in the lab itself.

The Icon

“Pfft, the icon?” you may ask, but this is a welcomed addition. I for one, could never tell the icon from Server Manager and Cluster Manager, they were too much alike. The new icon sports a modern look:

Node View Improvements

The Node View displays a wealth of information just by looking at the regular node tab. v2 of Windows HPC Server had the following tabs: Properties, Network, Metrics, Operations, and Provisioning Log (click on any image for a better look):

v3 has the following tabs: Node Health, Properties, Network, Metrics, Operations, and Provisioning Log. The Node Health tab is incredibly useful feature, as you can now get a general idea of the node state by looking at this tab. The pivoting links provide a quick access to other related areas of that have to do with this node:

The Heat Map

The Heat Map has gone a mayor overhaul, starting with the introduction of Tab Actions. This feature allows you to have multiple heat maps (each still can have up to 3 metrics), but by having as many Node Map Action tabs, you can have as many Heat Maps with different metrics as you’d like, which is incredibly useful IMHO:

One of my major gripes in v2 was the lack more metrics than the ones that were available for monitoring. v3 is not shy on the metrics that are now available, and you can see it in when selecting the metrics for List Views:

I’ll keep posting small posts on more things that I find differently. So far, V3 feels like a great improvement. Windows HPC Server has come a long way, and it’s nice to have seen its evolution since the days when it had the “Compute Cluster 2003” name.

HPC Pack SP2 was released a while back, but if you try and find it, you won’t be able to – what gives? After searching a bit in the Windows HPC Server Job Submission and Scheduling forums, I found the answer:

A minor issue was found in SP2 that could affect users who have applications that interact with the HPC job scheduler via COM. While we prepare a fix for the issue we have disabled the downloads of SP2.

Last Friday marked the end of a 3-day event in México City around High Performance Computing for IT Professionals. The attendees learned everything they need to know regarding Windows HPC Server 2008 in terms of setup, scheduling jobs, PowerShell, command line, network topologies, SOA, job templates, node templates, and node groups. The crowd was great to work with, highly-skilled individuals form different backgrounds, some wanting to get their feet wet with this technology, and other experienced folks who wanted to learn more about advanced topics.

This was my first time in México City and it’s unfortunate that I could not stay a bit longer to get to know the city a little better. At least we got to try authentic Mexican food and ate at a legitimate “Taquería” close to the event venue, the “Borrego Viudo” while watching the Mexican soccer team beat Italy in the World Cup:

As a follow-up to our attendees, I would like to communicate the following:

1. Your diplomas for the course completion can be obtained from KED starting June 7th (contact person is Liliana Gutiérrez)
2. If you would like a copy of the training material, please drop me an e-mail and I’ll make sure you get a copy (csaborio at scorpiotek dot com)

Now, off to TechEd NA 2010!

Note: This script will work with PowerShell v1.0 and is slow.  If you would like a faster version and you are willing to upgrade your head-node to v2, check this post instead.

I’ve been working in creating a series of demos for a couple of HPC trainings we are delivering this month.  One of the things that I’ve noticed with the heat map, is that when restoring the nodes from a snapshot (they are virtual nodes) the heat map fail to display the info.  This is due to some services on the nodes not starting and failing to communicate with the head node.  The solution is to go through every node and restart this service.  Good luck doing that if you are running a 1,000 node cluster!

You might think “well, why not just use clusrun to stop and start the service?”  Thought of that and then realized that one of the services that need to be restarted is the one that is used for clusrun, so as soon as you shut it down, your clusrun session is over.

Once again, psexec and PowerShell save the day.  If you would like to restore heat map functionality from the head node to all nodes on your cluster, follow these steps:

1. Download and drop a copy of psexec on c:\Windows\System32 on your head node
2. Copy the PowerShell code below and save it in a file, say “C:\FixHeatMap.ps1″

   $HEAD_NODE = "HPC-HN"
   Write-Host "Setting Head Node To:" $HEAD_NODE
   Set-Content Env:CCP_SCHEDULER $HEAD_NODE
   
   $nodes = Get-HpcNode
   $services = "HpcManagement", "HpcNodeManager"
   
   foreach ($node in $nodes)  {
   foreach ($service in $services) {
   
   $serviceStart = "net stop $service"
   $serviceStop = "net start $service"
   $commandBytesStart = [System.Text.Encoding]::Unicode.GetBytes($serviceStart)
   $commandBytesStop = [System.Text.Encoding]::Unicode.GetBytes($serviceStop)
   $encodedStartCommand = [Convert]::ToBase64String($commandBytesStart)
   $encodedStopCommand = [Convert]::ToBase64String($commandBytesStop)
   $commandLine += $encodedCommand
   
   $computer = $node.NetBiosName
   $computer = "\\$computer"
   
   psexec $computer cmd /c "echo . | powershell -EncodedCommand $encodedStartCommand"
   psexec $computer cmd /c "echo . | powershell -EncodedCommand $encodedStopCommand"
   }
   }
   #Start the service that retrieves info from CN to heat-map
   net stop HpcSdm
   net start HpcSdm
   

3. Edit the text of the file so that $HEAD_NODE = “HPC-HN” matches the name of your head node
4. Open an HPC PowerShell command prompt and launch the script, here is the script in action with 4 nodes:

One of the things you will notice is that it can be slow as 3 services are started and stopped for every node via a remote call with psexec.  I know the process can be long but I don’t have time to optimize it.  If there are PowerShell gurus that would like to chip in, please be my guest!

Sources

• http://www.leeholmes.com/blog/UsingPowerShellAndPsExecToInvokeExpressionsOnRemoteComputers.aspx

Last week was Supercomputing 09 in Portland Oregon. We had the great opportunity to assist with setup and demos, which all turned out quite good.

Microsoft’s booth was huge this year, we had about 12 demo booths and people showed lots of interest in partner demos and Microsoft demos alike. A partner demo that particularly called my attention was Adaptive Computing (formerly Cluster Resources), which had an on-demand solution for provisioning an HPC cluster with either Linux or Windows machines.

Unfortunately, my camera decided to die at the event and it was until the last day that I managed to get a replacement. I took the following pics of the Microsoft booth way early in the a.m., so there no one was at the show.

BTW, the whole booth was created by SOEST, who have been building amazing trade show exhibits for quite some time, and I think the pictures speak for themselves:

The booths looked quite futuristic

The theme was taking flight with HPC Server 2008

A flight simulator was placed at the middle of the booth, pretty sweet.

Meeting place in front of the flight simulator

This was Eric Lantz’s booth, which had a real time rendering of car that was sent to 15 partner booths across the show floor.

Information booth, the machine at the right was a demo Cray (about $35K for one of those puppies)

Front of the booth.

This was the theatre where live HPC presentations took place througout the day.

Kudos to both Amys, Fred, and Doug, who made the event a complete success and whom are always a pleasure to work with.