Slow acquisition when using with 13 channels (2 controllers)

[reszelaz]

Hi all,

Have any of you experienced slow acquisition time when using 13 channels (2 controllers)? Here at ALBA we have two setups, first with 7 channels (1 controller) and second one with 13 channels (2 controllers). The second one is slower by 20 - 30 %.

The comparison was made by acquiring:

• 1000 frames EXTERNAL_GATE (0.1 s of acquisition time and 0.00001 s of latency)
  • first: 102 s
  • second: 122 s
• 4800 frames EXTERNAL_GATE (0.1 s of acquisition time and 0.00001 s of latency)
  • first: 480 s
  • second: 647 s

The first one uses:

lima-core: 1.7.4
lima-tango-common: 1.7.3
lima-xspress3: 1.4.0

And the second one:

lima-camera-xspress3       1.9.0            py37h2bc3f7f_0    esrf-bcu
lima-camera-xspress3-tango 1.9.0                         0    esrf-bcu
lima-core                  1.9.14           py37h2bc3f7f_0    esrf-bcu
lima-tango-server          1.9.16                     py_0    esrf-bcu

If not, do you have any clue on where to look for the culprit?

[laurent-claustre]

We have more recent codes for this plugin with some optimizations and fixes about acquisition threading.

We will come back to you with maybe a new conda package 1.9.1 and a new sdk conda package as well. We don' t have this detector at ESRF so we will let you testing the new packages.

With new packages you must create a new empty conda environment with the proper channel list for installation of the dependencies as well, latest lima-core (1.9.16) is not compatible with lima-camera-xspress3 1.9.0 since we change the version of the python C++ wrapper SIP.

[reszelaz]

Thank you @laurent-claustre! That sounds promising. Whenever you pass us the instructions we will make the tests. Next week we have shutdown, so it would be a good moment.

[reszelaz]

Hi @laurent-claustre,

We have just tried to run the first system, the one with 7 channels (1 controller), with the same lima packages from conda, and its performance was not significantly affected. So, to my understanding it is either:

1. Problem of the lima xspress3 plugin that does not scale well with higher number of channesl (2 controllers)
2. Problem of the xspress3 hardware that we use in the second system.

I have asked Quantum Detectors for comments. I will let you know whenever we have some more information.

Maybe we could ask on the [email protected] mailing list if there are any lima users of xspress3 with such high number of channels (2 controllers)? And how the system behaves for them?

[stufisher]

I'll copy @liamoryan in here for feedback

[liamoryan]

@laurent-claustre I think that the DUBBLE beamline at ESRF has an Xspress 3 Mini. I'm looking into the problem reported by @reszelaz. My suspicion is that it is due the generation of Xspress he is using, the X3X (13 channel) is newer technology than the X3 (7 channel) and the library handles acquisition slightly differently.