How we put the workstations under pressure

Enter the workstation: designed for concurrent multiprocessing, workstations are rugged and reliable, with multiple, symmetric CPUs and gobs of memory to power through even the toughest workloads. You need to really load these machines down before their relative merits begin to surface, and that means generating concurrent workloads that exercise a variety of OS and application subsystems.

For this review, we did just that. I utilized one of my favorite test tools, Clarity Studio from CSA Research. Using a combination of parallel workloads — client/server database (specifically, ActiveX Data Objects), workflow (MAPI), Windows Media playback, and Windows Media encoding — I generated a hailstorm of CPU and memory activity.

I then scaled these workloads on each system, increasing the number of concurrent tasks as well as their complexity, all the while tracking the systems’ performance and health through various internal and external metrics counters.

The net result? Despite a great deal of hype, AMD’s 2.2GHz Opteron 248 CPU — as embodied in the IBM IntelliStation A Pro workstation — doesn’t fare well under heavy workloads. When compared head-to-head with last year’s Intel Xeon platform, a 3.2GHz/533MHz Front Side Bus model represented here by the MPC NetFrame 600, the Opteron fades as the workload level increases.

In fact, across the range of tests, the Opteron system took an average of 15 percent longer to complete the tasks than the Xeon. In some cases, most notably client/server workflow against a MAPI message store, the Opteron took over 30 percent longer.

An examination of OS metrics data collected by Clarity Studio showed that the Opteron was definitely struggling to juggle all those threads. One metric in particular shed additional light on the results. The Peak CPU Saturation Index, which is calculated from a sampling of the Processor Queue Length counter as exposed by the Windows Performance Data Helper libraries, showed that, on average, the Opteron system had 16 percent more waiting threads in its queue — a clear indication that the system was in fact CPU-bound and running out of processor bandwidth.

My interpretation: Hyper-threading support on the Xeon allowed it to continue to scale thanks to its ability to execute more than one instruction at a time. Once again, Intel’s simultaneous multitasking technology — where underutilized pipeline resources are shared to create a second, virtual processor image — is looking like an ace in the hole for the company’s workstation strategy.

The story gets worse for AMD when you factor in the newest Xeon processors from Intel. Preliminary results from two systems based on the new 800MHz FSB Xeon show the aforementioned average performance gap widening to nearly 50 percent (the MAPI workload, in particular, is now running 115 percent faster than Opteron), with CPU Saturation now 30 percent higher for Opteron when compared to the next-generation Xeon CPU (watch for our expanded coverage in an upcoming issue).