[Monotone-devel] Re: wrapping up the changeset branch

[graydon hoare]

Asger Ottar Alstrup wrote:

> But maybe you can tell me the status of these things given the recent 
> developments:
>
> - Storing large files ~ 1 GB

it cannot do this at the moment. it requires the upgrade to sqlite 3, 
which we have not performed yet.

> - Efficient working with large files - i.e. not having them three times 
> in memory in common operations

no work has been done on this.

> - Efficient distribution of large files

I don't know what sort of efficiency gain you might be thinking of. 
perhaps something to do with shared sub-fragments, or simply avoidance 
of copying?

> - Reduce hashing time on large files

no work has been done on this.

> - Footprint on each person machine should be a function of what they 
> have checked out, not the history or what is in the repository

there has been a minor improvement here on the changeset branch, which 
is that netsync knows how to pull both "forwards" (starting from the 
oldest historical record) and also "backwards", when it realizes it is 
fetching a disconnected subgraph. this "backwards fetching" is actually 
quite a bit more efficient when it's noticed, since it is the way deltas 
in the underlying storage layer are kept.

but towards the goal you have in mind, it's theoretically feasible to 
have the "backwards" fetch stop early, say only retrieve a small-ish 
subgraph near the heads of a branch, perhaps all nodes under the least 
dominator of the current heads. I say theoretically rather than 
practically, because we haven't made any such feature yet.

-graydon