OK, so a possble Deadlock condition is what exactly and should I see these error messages while Windows is running....errrrr....no apparantely not according to Microsoft....
OK, so this error means that something in the kernal is out of memory or you have a leak...so lets go right to it and get this sorted...right lets get an action plan sorted people:
A Performance monitors
B. Windows Debugging
C. Registry Optimisation
D. Paging File?
A. Performance Monitoring
Setup counters on all clusters servers so follow these instructions to complete this action:
1. Open Perfmon (locally or remote, if remote, specify the target servername before adding counters)
2. Expand 'Performance logs and alerts in the left window pane, select 'Counter logs'
3. Right click in the right window and select 'New log settings'
4. Name the log and add the following Objects:
Memory
Logical Disk
Paging file
Physical disk
Process
Processor
Server
System
Terminal Services objects (if server is a Terminal server)
5. Set the Interval to 30 seconds
6. Accept the settings and close out the dialog box.
7. You should see the new log setting turn Green. Close performance monitor and allow the system to run until the next occurrence of the problem.
B. Windows Debugging
Your cluster servers WILL need internet access to allow this to work with downloads enabled
1. Download and install the Debugging Tools for Windows package from Microsoft's web site
here
2. Select the link for "Install Debugging Tools for Windows 32-bit Version (or 64-bit if on a 64-bit OS)
3. Download and install the latest version listed into the C:Debuggers folder
4. Launch Windbg from the c:debuggers folder
5. Click on ‘File’ and select ‘Symbol File Path...’ and enter the following information and accept the changes: "SRV*c:symbols*http://msdl.microsoft.com/download/symbols"
6. Click on ‘File’ and select ‘Kernel Debug’ then select the ‘Local’ tab and accept the changes
7. Once the debugger loads, click on “Edit†and select Open/Close log file, name the log file then run the following commands:
!vm (Shows Virtual Memory Sizes)
!poolused 2 (Sort by non-paged pool processes)
!poolused 4 (Sort by paged then kernel process)
!process 0 0 (Shows processes with leaks)
C. Registry Keys
TCP/IP Chimney, what is this???
Windows TCP/IP Chimney Offload architecture has been developed to maximize host CPU utilization for application workloads on multi-Gigabit networks (this can cause excessive network issues if enabled and the hadrware does not support the technology)
These registry keys will be set to "1" in SP2 of Server 2003 whcih can cause issues for networking cards....
HKLMSystemCurrentControlSetServicesTcpipParameters
"EnableTCPChimney"=dword:00000000
"EnableTCPA"=dword:00000000
"EnableRSS"=dword:00000000
To enable crash on demand you need to set the registry key:
HKLMSystemCurrentControlSetServicesi8042prtParameters
"EnableCrashOnScroll"=dword:00000001
D. Paging File
If you have a cluster with 3.5GB of memory then you usually need 1.5x times that as a paging file up to a maximum of 4096MB so here we need to set the Minimum and maximum to 4096MB
WARNING : DO NOT stagger the values so one is 2046MB and the other if 4096MB if you do your systems will hang on allocating more paging file this is not good a cluster, if you are not sure about how to complete the paging file change then leave it to a Windows Professional!!!!!
This will not fix the issue but it will make issues more evident....
Finally, Microsoft do not support mail files in a cluster configuration as PST files (the better ones linked to Exchnage) as they hold hard locks open to the files which affects the clusters operation....to see if this is the issue using Network Monitor and filter by user and you will see the PST files using nmost of the resources......
OK, so this error means that something in the kernal is out of memory or you have a leak...so lets go right to it and get this sorted...right lets get an action plan sorted people:
A Performance monitors
B. Windows Debugging
C. Registry Optimisation
D. Paging File?
A. Performance Monitoring
Setup counters on all clusters servers so follow these instructions to complete this action:
1. Open Perfmon (locally or remote, if remote, specify the target servername before adding counters)
2. Expand 'Performance logs and alerts in the left window pane, select 'Counter logs'
3. Right click in the right window and select 'New log settings'
4. Name the log and add the following Objects:
Memory
Logical Disk
Paging file
Physical disk
Process
Processor
Server
System
Terminal Services objects (if server is a Terminal server)
5. Set the Interval to 30 seconds
6. Accept the settings and close out the dialog box.
7. You should see the new log setting turn Green. Close performance monitor and allow the system to run until the next occurrence of the problem.
B. Windows Debugging
Your cluster servers WILL need internet access to allow this to work with downloads enabled
1. Download and install the Debugging Tools for Windows package from Microsoft's web site
here
2. Select the link for "Install Debugging Tools for Windows 32-bit Version (or 64-bit if on a 64-bit OS)
3. Download and install the latest version listed into the C:Debuggers folder
4. Launch Windbg from the c:debuggers folder
5. Click on ‘File’ and select ‘Symbol File Path...’ and enter the following information and accept the changes: "SRV*c:symbols*http://msdl.microsoft.com/download/symbols"
6. Click on ‘File’ and select ‘Kernel Debug’ then select the ‘Local’ tab and accept the changes
7. Once the debugger loads, click on “Edit†and select Open/Close log file, name the log file then run the following commands:
!vm (Shows Virtual Memory Sizes)
!poolused 2 (Sort by non-paged pool processes)
!poolused 4 (Sort by paged then kernel process)
!process 0 0 (Shows processes with leaks)
C. Registry Keys
TCP/IP Chimney, what is this???
Windows TCP/IP Chimney Offload architecture has been developed to maximize host CPU utilization for application workloads on multi-Gigabit networks (this can cause excessive network issues if enabled and the hadrware does not support the technology)
These registry keys will be set to "1" in SP2 of Server 2003 whcih can cause issues for networking cards....
HKLMSystemCurrentControlSetServicesTcpipParameters
"EnableTCPChimney"=dword:00000000
"EnableTCPA"=dword:00000000
"EnableRSS"=dword:00000000
To enable crash on demand you need to set the registry key:
HKLMSystemCurrentControlSetServicesi8042prtParameters
"EnableCrashOnScroll"=dword:00000001
D. Paging File
If you have a cluster with 3.5GB of memory then you usually need 1.5x times that as a paging file up to a maximum of 4096MB so here we need to set the Minimum and maximum to 4096MB
WARNING : DO NOT stagger the values so one is 2046MB and the other if 4096MB if you do your systems will hang on allocating more paging file this is not good a cluster, if you are not sure about how to complete the paging file change then leave it to a Windows Professional!!!!!
This will not fix the issue but it will make issues more evident....
Finally, Microsoft do not support mail files in a cluster configuration as PST files (the better ones linked to Exchnage) as they hold hard locks open to the files which affects the clusters operation....to see if this is the issue using Network Monitor and filter by user and you will see the PST files using nmost of the resources......
Tags
Clustering