• Management Pack:  HPC Server
  • MP Version:  3.1.3266.0 for HPC Server 2008 R2
  • Released:  2/14/2011
  • Publisher:  Microsoft

Network Direct Monitor

  • ID:  Microsoft.HPC.2008R2.Monitor.Network.Configuration.NetworkDirect
  • Description:  Network Direct configuration monitor for HPC 2008 R2 network
  • Target:  HPC 2008 R2 Cluster Network
  • Enabled:  Yes

Operational States

Name State Description
Success Success  
Failure Error  

Overridable Parameters

Parameter Name Default Value Description Override
Timeout Seconds 300  
Interval Seconds 3600  
Sync Time    

Alert Details

Monitor State Message Priority Severity Auto Resolution
Failure (Error) Network Direct configuration is not correct Medium Critical Yes

Run As Profiles

Name
HPC Server Admin Action Acount

Monitor Knowledgebase

Summary

This monitor checks the NetworkDirect registration with the Windows operating system. The monitor will enter the Critical state if:

  • NetworkDirect is disabled on the application network when NetworkDirect-capable drivers have been installed on the cluster.

  • NetworkDirect is enabled on the enterprise network.

NetworkDirect is a Remote Direct Memory Access (RDMA) networking interface that offers low latency and high throughput performance for Message Passing Interface (MPI) traffic, but the use of this interface requires RDMA-capable networking hardware and drivers. The usual configuration is to enable NetworkDirect on the application network for MPI traffic.

Causes

If NetworkDirect is disabled on the application network, the most likely cause is that the NetworkDirect driver (also called a “provider”) has not been registered with the Windows operating system in which the provider would be added to the Winsock Catalog.

If NetworkDirect is enabled on the enterprise network, the network cards of a system were likely swapped out with non-RDMA-capable cards after the prior network interface card drivers were registered with the Winsock Catalog.

Resolutions

The registration of the NetworkDirect provider with the operating system is unique to each hardware vendor. Some hardware vendors may include this step in their driver installer (.msi), and some may use a separate step. The InfiniBand providers accomplish this step using a separate utility that can register (ndinstall –i), de-register (ndinstall –r), and list (ndinstall –l) the networking providers on a system. For detailed instructions about using NetworkDirect-enabled drivers, refer to the instructions provided by your hardware vendor.

External References

InfiniBand device drivers with NetworkDirect support can be installed and registered at the same time that you deploy the compute nodes in your HPC cluster. This is accomplished by deploying the compute nodes using a node template that has been specially configured for this purpose. For more information, see http://go.microsoft.com/fwlink/?LinkId=130612.

See Also for HPC Server Management Pack


Downloads for HPC Server Management Pack

AZURE OPTIMIZATION ASSESSMENT GET STARTED
MIGRATION TO AZURE GET STARTED
SYSTEM CENTER MIGRATION TO AZURE GET STARTED
MIGRATION TO AZURE FOR SQL AND WINDOWS 2008 GET STARTED