GET A SYSTEM-LEVEL VIEW
Performance tuning often focuses on optimizing a particular process or workload. However, there are times—like during driver development, memory pressure, and storage loading—where you need to get a broader view of the entire system.
Balance Workloads & Track Process Interaction
Often tuning is focused on optimizing a single process. But, when you are tuning drivers or balancing multiple applications running on a server, you need to have visibility on the overall activity of all processes on a system.
Figure 1 shows options in Intel® VTune™ Amplifier for system-level profiling.
Tune Drivers with Kernel Visibility
To analyze driver software, you need visibility of how it performs on the whole system, versus performance on a single process. Intel VTune Amplifier can sample an entire system, including visibility of kernel activity. High resolution hardware sampling (about 1 ms versus about 10 ms for software sampling) can find hotspots in small functions with very low overhead.
Activity for the entire system can be displayed many ways. The top section displays a grid organized by package, then core, and then process. This can be resorted by module, core, process, or thread. In the bottom section, the timeline shows individual threads (see Fig. 2).
Run Longer System Profiles (Preview)
The Platform Profiler collects coarse-grained, system-level metrics for extended profiling of minutes to hours. Software architects can identify workloads or phases of workloads that use hardware inefficiently and need tuning. Infrastructure architects can see if the current hardware configuration is a good match for most workloads. You can then decide how to improve performance, such as adding memory or I/O.