﻿id	summary	reporter	owner	description	type	status	priority	milestone	component	version	resolution	keywords	cc	launchpad_bug
707	use local storage server as encrypted cache	kpreid	nobody	"Summary:
 * Local caching of arbitrary portions of a tahoe filesystem can be implemented by adding shares to the local storage server.
 * This can be considered an aspect of server selection policy, except that the implementation is not just uploader's-choice.
 * Caching of a selected subtree could be implemented as a variant of deep-check/repair.

Condensed log:
 * [17:39] dreid: I think it's a negative user experience to require UI file operations to actually hit a tahoe node.  At the I have a folder on disk and I want to access it's data level.
 * [17:39] kpreid: you mean, due to response time?
 * [17:44] zooko: dreid: you mean you want the file contents to be cached locally, right?
 * [17:45] dreid: zooko: For most stuff yeah.
 * [17:46] kpreid: that's just a matter of proper distribution of shares :-)
 * [17:46] dreid: I obviously wouldn't want that for like my Pictures folder, because that's huge.  But for where I keep my documents, and presentations, and my bzr repos, totally want them always local.
 * [17:47] kpreid: (semi-serious: using the shares for cache is good if you want your local machine to not have plaintext when idle. you could also put a plaintext cache in an encrypted fs, but that's less trivial)
 * [17:56] dreid: kpreid: I don't think in everyday usage I would care, but I can see how some people would.
 * [17:56] kpreid: well, I run an encrypted-homedir laptop. I want my private keys and financial info to not fall into random laptop thief's hands and so on
 * [17:57] dreid: Well you're a paranoid freak. (In the nicest way possible.)
 * [17:58] dreid: My private keys live in my pocket, but yes i see your point.
 * [17:58] kpreid: so I could put my cache of hypothetical private shared documents on tahoe in that home dir, but then I'm bloating it with cache -- I'd rather keep some already-ciphered shares around
 * [17:58] kpreid: so it's just a matter of appropriate share (re)location policy!
 * [18:00] kpreid: Random strawman proposal: start with downloading shares of all files in a directory you just viewed (do it starting with 1 share parallel across all files, not k shares)
 * [18:32] zooko: kpreid: I really like your idea of using ""server selection policy"" as an implementation of ""caching"".
 * [18:32] kpreid: but...this is not *just* the uploader's choice; this is nodes proactively adding shares to their own stores
 * [18:33] zooko: Good point.
 * [18:34] kpreid: which isn't to say it's not a good thing, but just that it's a slightly different mechanism from uploader's selection
 * [18:35] zooko: Yep.
 * [18:36] kpreid: also: deep-check/repair is similar to 'pull this subtree into my cache' in what it has to do
 * [18:59] zooko: Okay, that's a cool idea.  Maybe mention it on some ticket?  Gotta run...  :-)
 

"	enhancement	new	minor	undecided	code-encoding	1.4.1		performance	kpreid vikarti@…	
