When starting a new profiling experiment, I would initiate the profiler by profiling_results = dcgmGroup.samples.GetAllSinceLastCall(None, dcgmFieldGroup).
However, when calling GetAllSinceLastCall(profiling_results, dcgmFieldGroup) for the first time after initializing it with None, the old data from previous profiling experiments are also returned, although profiling_results.EmptyValues() is called after every GetAllSinceLastCall in previous experiments. As a result, calling GetAllSinceLastCall for the first time becomes very slow. In the second call of GetAllSinceLastCall and onwards, only the data between two consecutive calls are returned, as expected.
Could you provide any insights about this behavior and how to overcome the slowness of the first call to GetAllSinceLastCall? Thanks a lot.
Hi, I am using the python bindings of DCGM.
When starting a new profiling experiment, I would initiate the profiler by
profiling_results = dcgmGroup.samples.GetAllSinceLastCall(None, dcgmFieldGroup)
.However, when calling
GetAllSinceLastCall(profiling_results, dcgmFieldGroup)
for the first time after initializing it withNone
, the old data from previous profiling experiments are also returned, althoughprofiling_results.EmptyValues()
is called after everyGetAllSinceLastCall
in previous experiments. As a result, callingGetAllSinceLastCall
for the first time becomes very slow. In the second call ofGetAllSinceLastCall
and onwards, only the data between two consecutive calls are returned, as expected.Could you provide any insights about this behavior and how to overcome the slowness of the first call to
GetAllSinceLastCall
? Thanks a lot.