Application performance characterization and profiling as a service
An improved system and method for characterizing application performance are disclosed. The system may comprise a profiling module and a web service module. The profiling module may be configured to prompt the user to select a public mode or a private mode and specify software and hardware targets for profiling. Test results from profiling may be combined with historical data obtained from a web service module providing access to a database of historical profiling test results to create recommendations for improving performance. The profiling module may upload the test-generated data in public mode to the web service module for use in future execution cycles or keep them private if in private mode.
1. A method for characterizing application performance, the method comprising:
downloading a profiling tool;
specifying a software target and a hardware target,
performing a hardware profiling test to obtain a first set of results;
performing a software profiling test to obtain a second set of results;
downloading prior result data from an external database of prior profiling test results; and
providing recommendations based on the first set of results, the second set of results, and the prior result data;
wherein the recommendations include (i) a modification to the hardware target that would improve application performance, and/or (ii) a modification to the software target that would improve application performance;
wherein the recommendations comprise a MPI_Send bandwidth per process ID and a recommended type of interconnect between nodes.
2. The method of claim 1 , further comprising:
determining whether a public mode or private mode has been selected; and
in response to determining that the public mode has been selected, uploading the first set of results and the second set of results to the external database of prior profiling test results.
3. The method of claim 2 , further comprising:
in response to determining that the public mode has been selected, uploading the recommendations to the external database of prior profiling test results.
4. The method of claim 1 , wherein the recommendations comprise a performance spectrometer.
5. The method of claim 1 , wherein the recommendations comprise predicted execution times for selected data set sizes.
6. The method of claim 1 , wherein the recommendations comprise a graph depicting load balance with hotspots per process ID.
7. The method of claim 1 , wherein the recommendations comprise a hotspot graph identifying potential hotspots via depicting an amount of time spent on each of a plurality of processes.
8. The method of claim 1 , wherein the recommendations comprise a recommended number of GPUs or CPUs to allocate.
9. The method of claim 1 , wherein the recommendations comprise an indicator of whether a solid state storage subsystem would benefit performance.
10. The method of claim 1 , wherein the recommendations include a visual performance indicator, the modification to the hardware target that would improve application performance, and the modification to the software target that would improve application performance.
11. The method of claim 1 , wherein the recommendations include:
a performance spectrometer;
an indicator of whether a solid state storage subsystem would benefit performance; and
a set of compiler optimization settings.
12. A system for characterizing application performance, the system comprising:
a computer server including at least one processor, the computer server is configured to:
(i) prompt a user to select a public mode or a private mode,
(ii) prompt the user to specify a software target and hardware target, and
(iii) perform hardware and software profiling tests to generate a first set of data; and
the computer server receives historical profiling test results from a web service module;
wherein the computer server combines the first set of data and the historical profiling test results to create a set of recommendations; and
wherein the set of recommendations are uploaded by the computer server to the web service module if the computer server is in the public mode;
wherein the recommendations comprise a MPI_Send bandwidth per process ID and a recommended type of interconnect between nodes.
13. The system of claim 12 , wherein the set of recommendations comprise compiler optimization settings.
14. The system of claim 12 , wherein the set of recommendations comprise library recommendations.
15. The system of claim 12 , wherein the set of recommendations comprise a performance spectrograph.
16. The system of claim 12 , wherein the set of recommendations comprise an amount of memory to allocate per node.
17. The system of claim 12 , wherein the set of recommendations comprise:
a number of GPUs to allocate per node;
a number of CPUs to allocate per node; and
a recommended type of interconnect between nodes.
18. A non-transitory, computer-readable storage medium storing instructions executable by a processor of a computational device, which when executed cause the computational device to:
prompt a user to specify a software target and a hardware target;
perform a hardware profiling test on the hardware target to obtain a first set of results;
perform a software profiling test on the software target to obtain a second set of results;
download prior result data from an external database of prior profiling test results;
make recommendations based on the first set of results, the second set of results, and the prior result data;
prompt the user to select either a public mode or a private mode; and
in response to determining that the public mode has been selected, uploading the first set of results and the second set of results to the external database of prior profiling test results;
wherein the recommendations comprise a MPI_Send bandwidth per process ID and a recommended type of interconnect between nodes.