search

qiboteam / qibolab

Quantum hardware module and drivers for Qibo.
https://qibo.science
Apache License 2.0
43 stars 13 forks source link

Deduplicate controllers and Octaves before registering to `QmConfig` #968

Open stavros11 opened 1 month ago

stavros11 commented 1 month ago

https://github.com/qiboteam/qibolab/pull/933#discussion_r1709791708

It will probably need to drop config from an attribute of QmController and generate it on-the-fly for every new sequence. This will probably make the code more modular, but I am not sure if it will have any serious impact in performance (most likely no).

alecandido commented 1 month ago

This will probably make the code more modular, but I am not sure if it will have any serious impact in performance (most likely no).

We have no concern with performances in Qibolab itself. The only bottlenecks we should care about are networking and the runtime performances of the devices.

stavros11 commented 1 month ago

We have no concern with performances in Qibolab itself. The only bottlenecks we should care about are networking and the runtime performances of the devices.

That's what I believe as well. Therefore we are probably not gaining much by caching the config in QmController, since every time play is called we are actually reuploading it to the instrument. We could gain something if we were able to tell the instrument to use the last config, which could be possible, but I haven't looked at it.

alecandido commented 1 month ago

We could gain something if we were able to tell the instrument to use the last config, which could be possible, but I haven't looked at it.

Yes, that's the falls within networking. If we can save uploads, then we may become more efficient. Take into account that of course we pay a time proportional to the amount of transferred bytes, but there is a fixed amount that is needed just to establish a connection. And I'm not sure if the configs are uploaded within the same connection of the rest of the experiment (since you may upload one config and run several experiments, it wouldn't be unbelievable that the two things are actually independent).