Running the code on a server by default

microsoft / Qcodes

Modular data acquisition framework

http://microsoft.github.io/Qcodes/

MIT License

339 stars 315 forks source link

Running the code on a server by default #123

Open cdickel opened 8 years ago

cdickel commented 8 years ago

I don't speak computer completely fluently but yesterday we had a bar discussion about how to run a single experiment with several computers, for example one for each ATS card one has. I agree, this sounds cumbersome but if the software was written by default to run as a server and the user interactions (function calls, scripts etc.) would be send to it, then one could run an experiment on two computers where one just passes function calls on to the other. This wouldn't necessarily be noticed by the user with a single computer, but opens up some nice possibilities. For our multi qubit experiments this might be a nice solution sometimes. Also, one could measure remotely without using TeamViewer. Data could be both sent back and stored on the measurement computer. In many cases, the fridges have a second computer that could run qcodes. Ultimately a remote computer running qcodes would not be so different from any other piece of hardware in our setups.

guenp commented 8 years ago

Yay bar discussions :) I am very much in favor of the idea of running qcodes on a server that is a completely independent machine, and then connect to that with a measurement computer (or even a laptop) to control it. data storage can then be local and available on the network s.t. the data can be retrieved from anywhere as well.

MerlinSmiles commented 8 years ago

I'd be a big fan of this!

AdriaanRol commented 8 years ago

I'm also a fan 👍 , maybe it should also be noted that QTLab already supports this, the Diamond group in Delft used this to connect their different setups used for the loophole free bell test.

giulioungaretti commented 8 years ago

@AdriaanRol @cdickel me no speak no QTLab 💔 (did it pass .py files around, or how did it do it ?)

But no it's not cumbersome, it does require some infrastructure changes (read, probably not version 1). AFAIK, this is the future :hourglass_flowing_sand: !

There's a few things to keep in mind:

servers means distributed, which means that one should pass messages around not functions (unless one wants to make it fail really bad)
how much one trusts the network
maintaining servers is not trivial ( what happens if any thing goes down ? is the researcher supposed to know how to unix? )
a set-up with a client(s) // servers (s) architecture may not be that easy to start with one click ( which is kind of what a lot of people want [maybe?]).

But I love the idea.

(and note we live in 2016 and whatsapp manages their billions message per day with one 16cores server, so this is totally doable, just my 2 :heavy_dollar_sign: )