Running KNIME from memory stick for security?


#1

Hi, I’m interested in taking KNIME with a trained model to a company, running the model on their data, processing the output. I’d like to leave them with the output but not the model. If I install KNIME on a memory stick (with model), run it on a machine they provide, give them the final output file, then we destroy the memory stick would I:
a) Avoid giving them the model or a way to recover the model?
b) Avoid them having to give me their data (other than temporarily whilst at their site to process it)?

many thanks for any thoughts on this.


#2

Hi @mark.gardner,

This should work, but you should expect that performance will be very poor due to the limitations of your average USB Stick.

To set this up, you first install KNIME on the stick as well as all needed extensions. You also need to create a workspace on the USB stick as well upon first startup. In addition, you have to set the temp directory used by KNIME to a directory on the stick (File -> Preferences -> KNIME). Note that the actual path on your machine might differ from the one on your client’s machine (stick mounted to different drive letter or other OS used). Therefore, you might have to change this setting again when at your client.

In case you get an error while starting KNIME from the stick, there is a good chance you ran into a timeout (again, things will definitely be a lot slower when compared to “standard” use). To avoid this, open the knime.ini in the installation directory and set the parameter “-Dequinox.statechange.timeout=” to a higher value. The default is 30000 Milliseconds.

Hope that helps!

Cheers,
Roland


#3

Hi Roland,

that’s very helpful. I will give it a go before I get there with a ‘KNIME naiive’ machine & stick,

very many thanks, Mark


#4

Hi Mark,

Glad I could help! Another thing that just came to mind is that you could try this with an external SSD rather than a standard USB stick. Performance will still be worse than when run directly on the machine, but it might help a little in any case.

Cheers,
Roland