Re: [Gnu-arch-users] arch lkml

[Christopher Dale]

Eric W. Biederman wrote:

> Tom Lord <address@hidden> writes:
>  
>
> > * serendipitous archive-as-payloads
> >
> >  So we're going to have long-lived client sites that pull
> >  updates or push updates to or from some archive host.
> >
> >  In fact, we'll have _many_ such clients for our most interesting
> >  hosts.
> >
> >  One idea is that we should, archive-side, just cache the payloads 
> >  that we'll send to these clients -- so that when a client needs some
> >  "thing" we just find it on disk and stream it out.
> >
> >  It turns out, then, that if instead of "caching" we simply "memoize" 
> >  those payloads ("caching" can discard stuff -- "memoizing" holds on
> >  to it forever) that the memo we create contains, in a _reasonably_
> >  compact form, all of the information we would have in an archive.
> >
> >  So: the memo _is_ the archive.  Disregarding archive-cached
> >  revisions for the moment, an arch archive _is_ just a memo of the
> >  payloads that a client might ask for.   
> >
> >  That's dirt-simple to implement, allows arch to use "dumb servers",
> >  and is the same "big-O" space complexity as the more sophisticated
> >  formats.  
> >    
>
> I need to read through this one a little more.  As I don't have a
> clear idea by what you mean with memorizing payloads.
>  

I think the concept is that with a cache, the system (arbitrarily, from 
the user's POV) decides what objects are going to be cached (usually 
based on a most-recently-used or most-frequently-used algorithm) and 
when they are going to be 'uncached' (usually cached objects are dropped 
when they become old or unpopular, i.e. the caching algorithm in reverse).

Whereas, with the "memoizing" (no "r"!) concept, the memoized objects 
are (presumably) explicitly decided (in the arch context the aim would 
be to identify and memoize "useful" payloads) and then kept forever (or 
at least until explicitly dropped).

You would expect that the contents of an active arch archive server 
cache would pretty closely mirror the equivalent "memo-box" (?), with 
the important exception that certain infrequently requested (but 
hideously useful) archive revisions could be found in the "memo-box" but 
not in the cache.

IOW, the "memo-box" is a cache equivalent/replacement, and indeed a 
hybrid cache+memo-box (= cache with the ability to import & mark certain 
cached items as "persistant") might be the best solution of all.

Then again, I could be wrong,

yrs k, etc.

Chris

(Doh! Always try google first: 
http://search.cpan.org/dist/Memoize/Memoize.pm
it does seem to be the same basic idea though!)