The Scalasca performance toolset architecture

Abstract

SUMMARYScalasca is a performance toolset that has been specifically designed to analyze parallel application execution behavior on large-scale systems with many thousands of processors. It offers an incremental performance-analysis procedure that integrates runtime summaries with in-depth studies of concurrent behavior via event tracing, adopting a strategy of successively refined measurement configurations. Distinctive features are its ability to identify wait states in applications with very large numbers of processes and to combine these with efficiently summarized local measurements. In this article, we review the current toolset architecture, emphasizing its scalable design and the role of the different components in transforming raw measurement data into knowledge of application execution behavior. The scalability and effectiveness of Scalasca are then surveyed from experience measuring and analyzing real-world applications on a range of computer systems.

ScalaTrace: Scalable compression and replay of communication traces for high-performance computing

“…Hence, structural information and event ordering are not preserved. There are many other tools [1,26,28,12] that report aggregate information, often based on the profiling layer of MPI, as is the case with mpiP or produce terabytes of traces. None of these tools are suitable for lossless tracing and later replay in a scalable manner with a single trace file in the megabyte range.…”

Best practice regarding the three P's: profiling, portability and provenance when running HPC geoscientific applications

“…Score-P v3.1 (Knüpfer et al, 2012) and Scalasca v2.3.1 (Geimer et al, 2010;Zhukov et al, 2015), where results collected with Score-P and Scalasca can be examined using the interactive analysis report explorer Cube v4.3.5 , Allinea Performance Reports v7.0.4 (January et al, 2015), Extrae v3.4.3 (Alonso et al, 2012), Paraver v4.6.3 (Labarta et al, 2006), Intel Vectorization Advisor 2015 (Rane et al, 2015), and Darshan v3.0.0 (Carns et al, 2011) (see Table A1 10 in Appendix A for a more detailed description of each performance analysis tool listed above). The modeling chain for the profiling workflow is as follows:…”

“…Thus, it can be complemented by using event tracing, which collects performance-related events in chronological order and therefore allows to reconstruct the dynamic application behavior in detail. However, care has to be taken when using event tracing, as it is more expensive than profiling as the amount of data in the trace increases with the runtime of the application (e.g., Geimer et al, 2010;Carns et al, 2011).…”

Reducing the Overhead of Direct Application Instrumentation Using Prior Static Analysis

“…In Scalasca [7] and TAU [21], the specification of such lists is supported through utilities that examine performance data from previous runs taken under full instrumentation. Selection criteria include the ratio between a function's execution time and its number of invocations or whether the function calls MPI -directly or indirectly.…”

“…Software tools are being developed to assist application scientists in harnessing these resources efficiently and to cope with program complexity. To optimize an application for a given architecture, different performance-analysis tools are available, utilizing a wide range of performance-measurement methodologies [17,13,21,18,7]. Many performance tools used in practice today rely on direct instrumentation to record relevant events, from which performance-data structures such as profiles or traces are generated.…”