A jug backend serves two tasks: it saves intermediate results and it synchronises the running processes when needed.
There are three backend available: one is based on the filesystem, the other is a redis backend and a simple in-memory backend which does not allow sharing across processes.
By default, jug will save its results in a directory called jugdata. This is done in a way that works across NFS if you are using a cluster.
Redis is a non-relational database system. I assume you have already installed it (it is easy to install from source, but it is now a part of Ubuntu, so that is even easier).
1. Run a redis server (see its docs for how to control it, but simply calling redis-server should work). 2. Now start your jug jobs with the --jugdir=redis://127.0.0.1/.
If you just want an in-memory store, use --jugdir=dict_store:filename and the results will be loaded and saved into filename (use just --jugdir=dict_store to get a run where results are not saved to file.
This is only appropriate for small projects, but has the lowest maintenance of any system.
If all your nodes share a filesystem and you don’t want to set anything up, just you the default filesystem backend. If your computations are non-trivial (in general, you should avoid breaking up your algorithm so much that each tasks takes less than a second), then this will be fast enough and very robust.
Do note that jug works well over NFS.
If your nodes do not share a filesystem then you are going to have to use redis. For some cases (if you have many outputs of computations that do not take very long), it is also faster and, if your results are small, takes up significantly less space.
So the tradeoffs are speed and space vs. convenience.