- What is a Compute Cluster?
- Different Types of Compute Jobs
- Building a Compute Cluster
- Computing Resources Needed
- Price Per CPU
- Optimal Solution Economics
- Beowulf Solution
- Beowulf Cluster on SPARC Hardware
- SUN Supported Beowulf Cluster
- How To Build Your Compute Cluster
- Advantages of a Sun Based Cluster
- Grid Computing
- Compute Cluster Software
Advantages of a Sun Based Cluster
The advantages of building a compute cluster based on Sun products instead of the Beowulf PC cluster components are:
Uniform software environment across all nodes in the cluster, from the smallest blade to the largest SMP. This uniformity enables both horizontal scaling (adding more nodes) and vertical scaling (adding more resources per node).
Full 64-bit support of both hardware and software processes requiring very large memoriesmore than 500 Gbytes of physical memory for a single process.
Solaris supports very large CPU counts; more than 100 CPUs with a single operating system image and good scaling.
Available Sun workstations can be used in the cluster.
Good administrative tools are available.
A very large set of optimized, commercial software packages is available.
All components are tightly integrated and supported by a single vendor.
The PrismTM multiprocess debugger is available at no cost to develop MPI codes. No free equivalent exists for a PC-based Beowulf cluster.
The availability of source code for software tools is sometimes cited as an advantage of the pure Beowulf solution. However, the source code for the Sun Grid Engine software and Sun HPC ClusterTools software, including Sun MPI software and the Prism debugger, are available at no cost for non-commercial use.
The addition of a few larger nodes to the mix allows the cluster to handle capability computing in addition to the capacity computing that has been traditionally associated with Beowulf clusters, while maintaining the cost effectiveness of the capacity solution.
The availability of readily ported, certified, supported applications, ISV codes with Sun MPI.
These advantages mean that you can build a compute cluster with nodes as simple as the NetraTM X1 server. This configuration allows up to 40 CPUs per rack, and is very effective in an environment running only single-threaded jobs requiring up to 2 Gbytes per process. If the memory requirements are high, nodes like the Sun Fire 280R server or the Sun Blade 1000 workstation can handle up to 8 Gbytes.
If you need to run multithreaded on more than two CPUs, you can add larger nodes such as the Sun Fire 3800, the Sun Fire 4800, or Sun Fire 6800 servers (24 CPUs with 192 Gbytes of memory) to the compute cluster.
For parallel nodes with heavy communication requirements, the performance increase from communicating over the high-bandwidth SMP backplane should more than offset the greater cost of these jobs on a per-CPU basis.
The biggest computer currently available from Sun is the 104 CPU Sun Fire' 15K server with 576 Gbytes of internal memory. Thus, you can build a cluster with mixed capabilities that is precisely tailored to the expected job mix in your environment, and meets your computing requirements in the most cost effective manner.