HP SVA V2.0 System Administration Guide

3 Common System Administration Tasks
This chapter describes additional tasks system administrators commonly do on the Scalable
Visualization Array (SVA). Many of the tasks described in this chapter typically require root
privileges and knowledge of the hardware components of the SVA and its software configuration.
3.1 Accessing Application Data Files
Data set sizes for visualization applications can range from less than 1GB to more than 100GB.
For example, a seismic data sets can be 1GB to 128GB, and a medical data set can be 1GB to 50GB.
Applications can access data files using NFS or the HP Scalable File Share (HP SFS) product,
which is based on the Lustre file system. When visualization nodes are integrated into a cluster
with an HP SFS, they access this file system using the System Interconnect (SI). When the HP
SFS is in a separate cluster and not accessible by the SI, visualization nodes access the file system
using GigE. Applications can also copy data files to all the visualization nodes.
Depending on the size of the data sets and the configuration of the SVA and related systems,
you have several options to implement access to the data:
• Copy the data files to every visualization node. Local disk access is fast, but requires
considerable disk space locally. If the data files change frequently, this method is not
recommended. Another disadvantage of this technique is that data is lost whenever the
node is re-imaged, an occasional occurrence in an HP XC cluster environment.
The /tmp directory is a reasonable location for locally stored data.
• Copy the data files to the head node, export the file share from the head node, and mount
the share on all the other nodes using the SI. You can place directories designated to contain
the data under the /var directory, which is then served using automount.
You can take advantage of the SI's high bandwidth potential if the SVA configuration uses
an interconnect such as InfiniBand® or Myrinet®. However, disk space on the head node
is limited. Additionally, network traffic congestion can occur on the SI between the file data
flow and the I/O traffic generated by the visualization application.
• Copy the data files to the head node as in the previous option. Use the Administrative
Network (GigE) rather than the SI to share the data files to all other nodes. This option is
useful if the application loads the SI, for example; with MPI traffic.
• Access the data files stored on an external file system such as HP SFS. You can access the
data directly from the visualization nodes using GigE NICs in the nodes. It is also possible
to make use of the SI for the data traffic. Configuring parallel data flows from the file system
using the SI to the visualization nodes offers the fastest method of data distribution.
Mounting external and internal file systems is described in the HP XC documentation How To
section. A link to the HP XC documentation is available from the online SVA Documentation
Library.
3.2 Managing Licenses
This section addresses the following topics:
• A description of the license manager and the license file.
• A means to determine whether the license manager is running.
• Instructions on how to start and stop the license manager.
3.1 Accessing Application Data Files 51