On Sep 16, 2010, at 4:16 AM, George Herbert <george.herbert(a)gmail.com>
On Thu, Sep 16, 2010 at 12:58 AM, John Vandenberg
On Thu, Sep 16, 2010 at 5:40 PM, Federico Leva
John Vandenberg, 16/09/2010 03:00:
English, French, German, Italian, Polish,
Portugeuse, Swedish and
Chinese Wikipedia all appear to have some mirrors, but are any of
reliable enough to be used for disaster recovery?
Obviously not, at least Italian ones.
The smaller projects are easier to backup, as
they are smaller.
sure that with a little effort and coordination, chapters,
universities and similar organisations would be willing to
backup a subset of projects, and combined we would have multiple
current backups of all projects.
I agree. Now we have only this:
Kudos to Milos & Wikimedia Serbia!!
How many TB are needed? I don't know
what's the average, but e.g.
now my university should have about 50 TB of free disk space
not so much, after all).
The key would be to allow the mirrors to delete their mirror when
need to use their excess storage capability. If they let us know in
advance that they are reclaiming the space, another organisation with
excess storage capability can take over.
I appreciate all the enthusiasm in thread, but (speaking for myself as
an individual, and IT consultant who does things like business
continuity and disaster recovery planning consulting among other
infrastructure work) this is a core operational competency role that
the Foundation needs to ensure is handled in house as part of the
routine IT operations. And, as I understand it now, it is, though I
have only had high level discussions with some of the Foundation staff
about this and not seen the server configs myself so I can't
personally attest to the status.
Database and file backups need to be in (at least) 2 locations, and my
understanding is that there are complete redundant copies at the
Amsterdam datacenter now, and that the new main datacenter in Virginia
will continue this.
If a third location is needed, the current HQ in San Francisco is
plenty far enough away from the other 2 locations to provide excellent
DR capability. If there's need for a datacenter / fast net access
redundant copy in SF or the Bay Area, a rack or few U of a shared rack
would be enough for a fileserver, and that's available at multiple
excellently connected locations in the Bay Area
Having multiple backups (w/ private user, deleted content data tables)
within WMF at various data centers is no doubt extremely crucial &
depending on third parties would be a terrible mistake.
But also up-to-date distributed copies (sans private data, but w/ full
history & images) outside WMF is also very important. Why can't we do
both? I highly highly doubt anything bad will happen to WMF but
despite best intentions & efforts, you never know (zombies take over?
rogue sys admin?). Distributed backups beyond WMF help ensure
wikipedia goes on w/o reliance on WMF
Disaster Recovery is not something the Foundation
should attempt to
crowdsource. I recommend it be left to professionals whose job it is
and who have prior experience in the field. If you haven't watched
major services drop, datacenters burn down, software environments melt
down, and spent years working to ensure that those don't happen again,
you really don't have a good feel for the type and magnitude of the
risks and the sorts of tools to employ to try and mitigate them.
Surely there are third parties with such experience and interested in
this. Internet Archives? Bibliotecha (sp?) Alexandria? Library of
Congress? Surely google has or should have copy?, what about as a
public dataset on Amazon cloud services (thought there was
something?), universities are also good some with super data centers
(e.g. San Diego State University), etc.
If there's interest in an offline discussion on IT
disaster recovery and reliability engineering, I can do that, but it
should be offline from Foundation-L...
Maybe not foundation-l :) but I am cool with some degree of
transparency & open discussion on a list or some communications
channel dedicated to the topic.
I'm not involved in creating dumps but couldn't it be possible to
offer daily or weekly diffs of enwiki and other wikis, and have
utilities to apply diffs to the last full dump? Having regular dumps
+ regular diffs (weekly, daily, and even minutely) + Swiss army knife
utilities for handling diffs and dumps is something that openstreetmap
has managed to excel with and makes me very happy :) to know people
have up-to-date copies distributed on various places. I feel sad to
know this is not the case with wikipedia :(
-george william herbert
foundation-l mailing list