Workstations Work as Clusters

By Peter Varhol

The workstation-based high-performance computing cluster that Parker Aerospace uses to speed its simulation turnaround times (see “Building a Virtual Workstation-based Cluster”) was made possible by the nature of the company’s principle hardware “the HP Z800 Workstation. DE contributing editor Peter Varhol interviews HP executives to discuss the characteristics that made it ideal for clustering. Responses are provided by Tom Salomone, HP segment manager for CAD and Design Engineering Industries; Dan Bennett, an HP technical consultant for the Z Workstations series; and Mike Diehl, the HP Z800 Workstation product manager.

Q. What are the primary characteristics of the HP Z Workstations that make them adaptable for workstation-based cluster computing?

A. The HP Z series family provides great computing power, and a great interactive experience for someone sitting at the workstation. At the same time, you are using otherwise unused resources on the workstation to contribute to a distributed compute solution. The HP Z800 Workstation is especially appropriate for this type of usage pattern, in particular because it’s a dual-socket design; that is, it can have two Intel Xeon processors, each of which can have up to six cores. Another key, and a very important point in terms of the technology included in these workstations, is the use of Intel’s Xeon processors, because they support a technology Intel calls VT-d, short for Intel Virtualization Technology for Directed I/O.

Q. Why is the support for Intel VT-d so important?

  A. The way we’re implementing this workstation cluster computing model is to use Parallels Workstation Extreme, and it uniquely takes very good advantage of Intel’s VT-d technology. That allows their software to implement virtual machines on the workstations, which can be directed to use only certain parts of the workstation.

Q. The traditional way of clustering workstations is often known as cycle stealing, but that’s not what Parker Aerospace is doing, correct?

  A. Correct. Any time you’re running multiple things on a workstation, you’re competing for resources, and unless there’s an intelligent way of putting bounds on one of those tasks, so that it does not steal the resources of the workstation, it’s going to slow down the engineer interacting with the computer. And that’s what Parallels Workstation Extreme does. It takes advantage of some of the design characteristics of the HP Z800 Workstation. For example, the dual network interconnect provides a means for clustering on a fast private network, while also giving the interactive engineer a separate connection on the enterprise local network. That really helps in terms of keeping workloads separate.

Q. And that’s also important because the cluster workload is dependent upon a fast network interconnect, right?

  A. That’s correct, although it depends on the workload. Some applications and jobs have to move a lot of data, and would really benefit from this configuration. For example, the HP Z800 Workstation has enough power and available slots to be able to put two high-end graphics cards into the workstation. The engineer sitting at the workstation has access to one of those graphics cards for work, while the other card could incorporate a GPU that could be used to accelerate the execution of the cluster job, if the application supports GPU computing. That would be very useful for jobs that have a high ratio of floating point operations. And GPU companies like NVIDIA are reaching out to application companies to support GPU computing as a part of their solver engines. Combined with the HP Z800 Workstation running Parallels Workstation Extreme virtualization solution, this could be a very powerful solution.

For More Information:

Hewlett-Packard

Intel

Parallels

For the full story on how Parker Aerospace implemented its virtual workstation cluster, download the white paper at deskeng.com/workstationcluster.

Contributing Editor Peter Varhol covers the HPC and IT beat for DE. His expertise is software development, math systems, and systems management. You can reach him at [email protected].