HP Scalable Visualization Array Software - Manual (Page 13)

Final images can also be transmitted to a remote workstation display over a network external to
the cluster. This lets users interact with applications running on the cluster from their offices.
Optionally, you can use HP Remote Graphics Software (RGS) to accomplish this more easily.
Figure 1-2 also shows a master application node communicating with the other visualization
nodes over the SI. The SI carries file I/O and application communications; for example, MPI
traffic. The user interface for a visualization application can run on a master application node
and communicate with the visualization nodes over the SI, sending control information such as
changes in point of view, data, or OpenGL commands.
1.3 Displays
Display devices are not necessarily provided as part of the SVA. For example, your site can use
projector display systems or immersive displays provided by third party vendors.
Displays fall into a number of categories, including immersive CAVE displays, single monitors,
multiheaded monitors, large wall displays, multiheaded desktops, flat panels, and projector
displays used in theaters. SVA hardware and software deliver images to digital or analog standard
interfaces. The SVA depends on the graphics cards to drive the image output. This means the
wide range of display devices that the graphics cards support are available for use.
See Section 3.5 for more information.
1.4 SVA Functional Attributes
The key to SVA scalability and flexibility is its combination of cluster technology with high-speed
graphics cards and networks to transfer data. The SVA enables scaling up the number of nodes
working on a problem in parallel to handle larger dataset sizes, to increase frame rates, and to
display at higher image resolutions.
1.4.1 Scalability
There are a number of ways that applications can be designed and implemented to take advantage
of an SVA for effective scaling:
• Performance scaling: Render image data on separate nodes in the SVA. In effect, the work
is divided up among nodes working in parallel. Larger datasets can be accommodated by
more render nodes. The system design can scale from four to forty visualization nodes. This
count does not include the required head node.
The parallel attributes of the rendering pipeline removes a key performance bottleneck of
a conventional hardware accelerated graphics architecture, which feeds data sequentially
to a centralized pipeline.
In addition, the choice of a network that transmits data among the visualization nodes with
adequately low latency and high speed maintains interactive frame rates for delivery to the
display devices.
• Resolution scaling: Parallel rendering, combined with the parallel display of multiple tiles
makes such scaling possible. You can display high-resolution data and use large display
surfaces, including immersive displays and display walls.
In general, adding nodes to a dataset of fixed size provides good scaling up of the frame rate,
although speed-up is not linear because of the inevitable overhead due to portions of an
application's code that cannot be made parallel. However, a strength of SVA as a cluster
visualization platform is that scalability is nearly linear when the dataset size and node count
are both increased. For example, doubling the node count from four to eight makes it possible
to double the distributed dataset size with virtually no loss of frame rate. To achieve such gains
in frame rate, an application must be a true parallel application to efficiently distribute data and
to load balance across cluster nodes.
1.3 Displays 13