Hardware Reference
In-Depth Information
9.3.1 Usage Examples
Two iconic examples of the many scientific and commercial application
domains in which GPFS is used include weather modeling and digital media.
GPFS has a long history of working with the weather forecasting commu-
nity around the world. Currently their performance requirements are typically
not as extreme as some other HPC workloads, but their availability require-
ments are much higher|some systems are designed to not miss a forecast for
several decades. To ensure applications can always retrieve and store their
data, this requirement demands not just a stable file system code base, but
the ability to seamlessly recover from failures and the ability to incrementally
add compute power and storage capacity over time.
Media workloads have a lot in common with HPC workloads. High band-
width requirements are combined with a rich, real-time, and heterogeneous
workflow. A single project, such as commercial or film post-processing, con-
sists of numerous teams, each working on different platforms and with different
data requirements, to generate and share data. Furthermore, digital media is
pushing the boundaries of modern data management as frame resolution, bit
depth, and frame rate continue to increase.
In many GPFS media deployments, a single file system is accessed by appli-
cations running on Linux, Windows, and Mac OSX platforms from both Fibre
Channel and Ethernet networks. Linux and Windows clients use the native
GPFS client for high-bandwidth applications, and NFS/SMB for lower band-
width applications. OSX clients use NFS to access file data through GPFS
CNFS servers. Users make extensive use of filesets and ILM policies to scope
projects and dynamically manage data placement in a tiered storage system.
9.4 Conclusion
For 15 years, GPFS has continued to grow and adapt to new application
domains and cluster architectures. For most applications, the scale-out and
distributed design allows GPFS to offer POSIX semantics with minimal, if any
overhead. Further, GPFS has proven its ability to adapt to emerging applica-
tion domains while innovating new features such as policy-based management,
disaster recovery support, data tiering, global data caching, and much more.
There are still many ongoing challenges|adapting to new workloads,
meeting ever-increasing performance requirements, and keeping up with
emerging network and storage technologies remains a continuous challenge
with every new GPFS release. Simplifying the management complexity of
GPFS's rich set of features, many of which are targeted at a specic appli-
cation domain, represents an opportunity for improvement. One goal is for
GPFS to become a more dynamic, self-configuring, and self-tuning system
 
Search WWH ::




Custom Search