Chapter 2. Infrastructure planning 33
2.4.3 Web application server
The representative web application solution are the IBM WebSphere, the Red Hat JBoss and
the Oracle WebLogic. These solutions have almost the same I/O pattern with a web server.
Consider the following parameters for GPFS random I/O:
pagepool: The default pagepool parameter size is typically not sufficient for random I/O or
workloads involving a large number of small files. In certain cases, allocating 4 GB, 8 GB
or more memory can improve workload performance.
prefetchPct: This parameter defaults to 20% of pagepool. GPFS uses this value as a
guideline to limit how much pagepool space will be used for prefetch or write-behind
buffers in the case of active sequential streams. The default works well for many
applications. Alternatively, if the workload is mostly sequential (video serving and ingest)
with little caching of small files or random I/O, increase this number up to its 60%
maximum so that each stream can have more buffers available for prefetch and
worker1Threads: This parameter controls the maximum number of concurrent file
operations at any one instant. If there are more requests than that, the excess waits until a
previous request has finished. This attribute is primarily used for random read or write
requests that cannot be pre-fetched, random I/O requests, or small file activity. The default
value is 48. The minimum value is 1. The maximum value of prefetchThreads plus
worker1Threads is 550 on 64-bit kernels.
blocksize: A smaller block size is likely to offer better performance for small file, small
random read and write, and metadata-intensive workloads.
2.4.4 Media contents delivery
This section provides details about media delivery contents in a GPFS cluster environment.
3D animation development infrastructure
How do you design and build this type of system? If you start with no studio and want to
create a full-length feature animation studio, what kind of infrastructure will you need?
Note: This is the pagepool on the GPFS client and not the GPFS server.
34 Implementing the IBM General Parallel File System (GPFS) in a Cross-Platform Environment
Figure 2-16 shows the basic architecture for a 3D animation development studio.
Figure 2-16 Basic architecture for an animation studio
The workflow infrastructure consists of four subsystems:
Content creation, editing workstations
Workstations are high-performance desktops, and in some cases, mobile computers. As
mentioned, the process of creating rich digital content, whether it is video for TV, cinematic
film, or graphics for interactive entertainment (games) can be divided into two basic
activities, content creation and content editing:
Content creation is the process of creating 2D or 3D images, animations, backgrounds,
textures, and so forth.
Content editing is the processing and sequencing of motion sequences into a final
form. Both are I/O-intensive graphics processes.
render farm is an industry term for what computer scientists might call a loosely coupled
supercomputer. Render farm technology arose as a way to meet the needs of
organizations attempting to produce compelling content in a reasonable amount of time.
The basics are fairly simple to grasp. Special effects and animation are collections of
individual pictures or frames. The process of turning a model into a fully lit 2D or 3D
picture is called rendering, which is a highly compute-intensive task. Although this area
usually runs on a Linux platform, 3D media content creation is usually a Windows
platform. Previously, NFS was used for sharing data. GPFS is now a good solution
because the GPFS cluster provides a mixed platform solution with Linux and Windows
Scalable network backbone
Linkage to other