It would be nice to have some semi-scientific experiments which could help answer the following questions:
- What is the space overhead of loose vs. packed object storage for non-trivial git histories?
- What overhead does the ipfs local storage add on top of that?
- How does fetch scale, considering that a lot of (smallish) objects have to be replicated?
- What is a good number of concurrent connections to the (local) ipfs daemon?
- Consequently, what is a good number of OS threads to reserve for the remote helper?
It would be nice to have some semi-scientific experiments which could help answer the following questions: