nLoad

A resource saturation, measurement and test orchestration package to stress test and establish safe operating regions for real-time computing platforms.

Induced Failure with Forced Saturation

Shown here is a representative "slice" through a multi-dimensional manifold from nLoad data. The surface shows the incrementally induced stress and failure (red regions) as a result of multicore cache interference in a high performance processor. Failure is measured here in terms of observed utilization, as successive cores are loaded for various values of applications' read-write mix. Several other applications' characteristics (e.g., memory size) are held constant for this slice's visualization, but were varied in the stress test. Other failure metrics, e.g., power consumption overload, can also be used. Such infrastructure "characterization curves" and an estimation of current applications' operating point within it, plus the available margins constitute valuable data used in certifying a mission-critical system.

• Computing infrastructure (e.g., processors, memory systems, caches, backplanes, buses, network cards, network protocols, operating systems, middleware etc.) have potential built-in resource contention that can be overwhelmed by applications operating on it.
• For mission-critical applications, and their certification, it is vital to know the failure bounds for each of these elements in terms of application characteristics 
  
• Given such characterization graphs, it is further useful to know where the current applications' operating point is, and how much margin is available in system before failure.
• Structural Engineers have been building material characterization graphs for ever - in terms of their application characteristics: temperature, pressure, vibration, force, impulse etc.
• nLoad offers capability to incrementally force saturate and establish failure boundaries for multicore processor caches, PCIe buses, network cards, operating system and middleware messaging stacks.
  
• nLoad data generates a multi-dimensional manifold of failure volume described in terms of application characteristics: processor and network consumption, type and mix of memory access, type and mix of network access etc.
• An application operating point can then be estimated within this multi-dimensional volume using other nHansa tools, and then to calculate available margins.

Applied Research Transition Success Story

nLoad was funded by the U. S. Air Force and developed in collaboration with Lockheed Martin. An example application to study multicore cache forced saturation is described here. nLoad has subsequently been extended to PCIe bus, network cards, and OS/Middleware message stacks.