Scan speed

This section displays the results of Sardana performance tests. The measurements have been done for various reasons:

It has to be verified that the Sardana performance is satisfactory.
When setting-up a new experiment technique, the results mentioned below can be used to decide whether Sardana is sufficiently fast. The alternative is always a hardware controlled continuous scan.
In case a beamline scientist is in doubt whether the online control system is performant, he can execute scans that are described below and compare the outcome.
It is interesting to see how the Sardana numbers compare with another program.

The first series of scans are performed with a dummy motor (exp_dmy01) in order to avoid any delays caused by the motor movements. The parameters are:

exp_dmy01
  - exp_dmy01.velocity == 10000
  - exp_dmy01.acceleration == 1

It appears that the results are basically independent of the velocity and acceleration values.

Here are the results for relative scans in the range from -0.1, 0.1. The number of points is 101 and the sample time is 0.1s.

 
Scan command: 
  %dscan exp_dmy01 -0.1 0.1 100 0.1

Elapsed times for different measurement groups (3.3.2020):  
  d1_c01                       18.7s (46%) 
  d1_c01 - d1_c04              20.2s (50%) 
  d1_c01 - d1_c12              23.5s (57%)
  d1_c01, mca-2048             38.3s (74%)
  d1_c01, mca-4096             47.9s (79%)
  d1_c01, mca-8192             67.9s (85%)
  d1_c01, i-petra:             19.6s (49%)
  d1_c01 - d1_c12, mca-2048    42.2s (76%)

%dscan exp_dmy01 -0.1 0.1 200 0.1
d1_c01                       36.7s (45%) 
%dscan exp_dmy01 -0.1 0.1 400 0.1
d1_c01                       74.0s (46%)

The scan executed with a measurement group containing one only one counter (d1_c01) takes 18.7s. These are the contributions:

The total sample time is 10.1s.
The time it takes to rel-move exp_dmy01 100 times from -0.1 to 0.1 is 2.0s. This has been measured with the macro 4.6.4.
From the lines containing the results for 4 and 12 counters we derive that each counter adds about 0.5s (for 101 scan points) to the elapsed time.
Subtracting the numbers mentioned to far we conclude that we have a 'system scan overhead time' of 7s.

If we repeat the scan with 200 and 400 points, we find a linear scaling.

So far only dummy scans have been performed. The tests have also been done with a real motor. The parameters are:

 
d1_mot66 configured like haspp09, th, p09/motor/exp.05 
  - conversion    -40000
  - acceleration   25000
  - BaseRate        2000
  - slewRate       25000
  - unitBacklash     0.1

It takes 19s to rel-move d1_mot66 in 100 steps from -0.1 to 0.1.

The elapsed scan times are (3.2.2020):

 
%timedMacro "dscan d1_mot66 -0.1 0.1 100 0.1"
d1_c01                       24.4s (58%) 
d1_c01 - d1_c04              26.2s (61%) 
d1_c01 - d1_c12              29.9s (66%)
d1_c01, mca-2048             44.7s (77%)
d1_c01, mca-4096             54.1s (81%)
d1_c01, mca-8192             74.5s (87%)
d1_c01, i-petra:             25.7s (61%)
d1_c01 - d1_c12, mca-2048    49.1s (79%)

%timedMacro "dscan d1_mot66 -0.1 0.1 200 0.1"
d1_c01                       44.7s (55%) 
%timedMacro "dscan d1_mot66 -0.1 0.1 400 0.1"
d1_c01                       86.7s (54%)

Conclusions:

If we take the motor move times into account, we see that the results of the dummy and the real motor scans are roughly in agreement.
The Sardana 'system scan overhead time' is in the order of 40% for dummy scans. For real motor scans this contribution is insignificant.

Subsections