guile profiling / speeding up derivations on slow storage

[Csepp <raingloom@riseup.net>]

I'd really like to speed up guix pull on my netbook and have been
thinking about ways to do so for a while.

First of all, I'm curious, how do other Guile developers profile code?
Could we add a profiling flag for the CLI similar to the existing
debugging flags?
On a related note, has anyone tried to write a causal profile for Guile,
or a similar Scheme implementation?

One thing that definitely needs improvements is memory usage.  That
should speed things up a lot too, since it would result in less swapping
on machines like mine.
I would love to hear some tips on doing memory profiling on Guile from
someone who has done that.  I also looked into porting the ideas from
Scalene to Guile, but haven't gotten far since I'm busy with uni and
other Guix projects.

I also thought about some possible ways of caching expensive
computations that are not built using the store.  Since Guix is mostly
purely functional in practice, I wonder if we could take some notes from
languages like Yatima.  Or just use a plain old database for some
operations that are currently done by Scheme.

I know some of this must have already been discussed, but it's pretty
hard to track it down, and I think the roadmap for performance
improvements really should be centralized somewhere.