https://issues.apache.org/SpamAssassin/show_bug.cgi?id=5845





--- Comment #5 from Justin Mason <[EMAIL PROTECTED]>  2008-03-15 16:57:17 PST 
---
(In reply to comment #4)
> Woo, mapreduce! :)
> 
> The issue with all this stuff, of course, is that it's written to deal with
> processing a large dataset with processing nodes in close proximity -- ie: a
> LAN w/ low latency, packet loss, high bandwidth, and probably no need for
> things like SSL, etc.
> 
> Our model is very not that. ;)

not yet!  This involves discussion of how to change that.

For example, Amazon EC2 offers a large set of processing nodes along those
lines, and I'm sure there may be other sites that would allow us to use Hadoop
infrastructure "for free" as a donation.  I don't think our current lack of
LAN-connected nodes is a blocker.

> We don't have a single dataset to run over, we have several individual (and
> unlikely to be shared) datasets.  Without sharing the data, we end up
> immediately reducing our setup to individual processing clusters, where the
> norm is to only have 1 machine (or at most a small handful of machines).

As I noted, there's a public and private set of corpora. In order to adopt
this, we have to be comfortable sharing a subset of our mass-checked corpora
online across mass-checking nodes.

I've already done this by splitting my ham corpus into "public" and "private"
subsets; private stays behind on my home server, but public is already uploaded
to the zone, and used in mc-fast and bb-jm mass-checks.  All of my spam corpora
are considered "public" in this respect.

so I'd envisage this involving creating a relatively large combined corpus on
the zone (or some other central point), containing everyone's "public" corpora.


-- 
Configure bugmail: 
https://issues.apache.org/SpamAssassin/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.

Reply via email to