Built-in timers¶
Basic timers¶
Basic built-in timers keep track of the time NEST spent for network construction and actual simulation (propagation of the network state). These timers are active in all simulations with NEST, and the measured times can be queried using GetKernelStatus
. For example:
nest.GetKernelStatus('time_simulate')
The following basic time measurements are available:
Name |
Explanation |
---|---|
|
Cumulative time NEST spent creating neurons and devices |
|
Cumulative time NEST spent creating connections |
|
Time NEST spent in the last
|
Note
While preparing the actual simulation after network construction, NEST needs to build the pre-synaptic part of the connection infrastructure, which requires MPI communication (Jordan et al. 2018). This happens only for the first call to Simulate()
unless connectivity changed in the meantime, and it may cause significant overhead by adding to time_simulate
. Therefore, the cumulative time NEST spent for building the pre-synaptic connection infrastructure is also tracked by a basic timer and available in the kernel dictionary as time_communicate_prepare
.
In the context of NEST performance monitoring, other useful kernel-dictionary items are:
Name |
Explanation |
---|---|
|
Cumulative simulated time |
|
Number of spikes emitted by the
neurons represented on this MPI
rank during the last
|
Note
nest.ResetKernel()
resets all time measurements as well as biological_time
and local_spike_counter
.
Detailed timers¶
Detailed built-in timers can be activated (and again deactivated) prior to compilation through the cmake flag -Dwith-detailed-timers=ON
. They provide further insights into the time NEST spends in different phases of the simulation cycle, but they can impact the runtime. Therefore, detailed timers are by default inactive.
If detailed timers are active, the following time measurements are available in the kernel dictionary:
Name |
Explanation |
Part of |
---|---|---|
|
Cumulative time for communicating connection information from postsynaptic to presynaptic side |
|
|
Cumulative time for core MPI communication when gathering target data |
|
|
Time for neuron update |
|
|
Time for complete spike exchange after update phase |
|
|
Time to collocate MPI send buffer from spike register |
|
|
Time for communicating spikes between compute nodes |
|
|
Time to deliver events from the MPI receive buffers to their local synaptic targets (including synaptic update, e.g. STDP synapses) and to the spike ring buffers of the corresponding postsynaptic neurons |
|