[Opensim-users] [Opensim-dev] Grid Numberging
Diva Canto
diva at metaverseink.com
Mon Nov 3 15:31:41 UTC 2008
Sure, I'll do this so not to get stuck. But this is so wrong...
Region servers should not *cache* assets. Period. It's an unnecessary
burden in two dimensions:
1) Performance.
2) Security of assets.
Any dealings between assets and the viewer should be done directly
between the viewer and asset servers. Region servers should be out of
the loop. Except, of course, for serving the objects that are inworld.
But the act of dragging an object from inventory to the world should
result in a *copy* being created (posted either from the viewer or from
the user's asset server) in the region server's asset *storage* (not
cache), and not this horrendous optimization that the LL system
currently does, and that pretty much stops the idea of scalability on
its tracks.
Is anyone here involved with LL Viewer development? I wonder how hard it
would be for them to interpret steganographic urls in uuids... or do the
right thing and use decent global resource identifiers, but I suspect
that's way off the radar.
Stefan Andersson wrote:
> Diva,
>
> there's a blindingly simple remedy; start treating the asset guid
> as something private between the region and the client, and publicly
> refer to assets by url instead.
>
> In effect, the fact that we are storing the assets untampered in the
> database is just a legacy issue. If we were to start processing the
> asset data more intelligently, things would look a lot differently.
>
> Suppose we;
>
> a) on asset upload, processed the asset and changed guids to
> urls, firstly by lookup (is this already an asset guid known to the
> region) then by url synthetization (create an asset storage url (maybe
> using the users home asset server that we already said we would
> store... ;D ) , post the binary to that and change all guid references
> in the asset to that)
>
> b) on asset download (from asset storage) processed the asset and
> changed urls to guid, again firstly by lookup (is it already known to
> the region) then, by fetching the asset by url and assigning it a guid.
>
> This would mean that the client, when it refers to an asset, actually
> requests it by the local guid that the region has assigned it - more
> as a local cache key than a public asset id.
>
> I believe this would solve the whole thing, more or less. I would love
> to see a proof of concept - it should be fairly easily thrown
> together, probably as a custom asset cache.
>
> Best regards,
> Stefan Andersson
> Tribal Media AB
>
> Join the 3d web revolution : http://tribalnet.se/
>
>
>
>
> ------------------------------------------------------------------------
>
> Date: Sat, 1 Nov 2008 07:52:33 -0700
> From: diva at metaverseink.com
> To: opensim-users at lists.berlios.de
> Subject: Re: [Opensim-users] [Opensim-dev] Grid Numberging
>
> The only reason why I'm suggesting the encoding hack as a horrible
> option not to be immediately discarded is that it is probably the
> simplest thing to do for the LL Viewer to start pulling assets from
> different places on the internet. From all I know, the LL Viewer is
> deeply rooted on UUIDs for universal identification; refactoring this
> will be a massive undertaking on the viewer. But simply giving
> semantics to UUIDs would not be that big of a deal. I'm thinking a few
> conditionals here and there:
>
> url = RetrieveUrl(uuid)
> if ItsASyntacticallyValidUrl(url)
> try to do the right thing, please
> else
> do what you do now
>
> or something to this effect.
>
> Of course, we're hitting our heads again on the pre-condition of using
> the LL Viewer...
>
> Dickson, Mike (ISS Software) wrote:
>
> We really don’t want to start encoding info into a GUID. The RFC
> that describes them pretty much details what kinds of entropy can
> be used.
>
>
>
> I’ve been thinking for a while that the handle service
> (http://www.handle.net <http://www.handle.net/>) might be a nice
> way to do a URL style mechanism to refer to assets and such not on
> a local server. It’s been a while since I looked at it but it’s
> pretty much designed for the purpose and there’s a proxy
> architecture already defined for it.
>
>
>
> Mike
>
>
>
> *From:* opensim-users-bounces at lists.berlios.de
> <mailto:opensim-users-bounces at lists.berlios.de>
> [mailto:opensim-users-bounces at lists.berlios.de] *On Behalf Of
> *Diva Canto
> *Sent:* Saturday, November 01, 2008 9:05 AM
> *To:* opensim-users at lists.berlios.de
> <mailto:opensim-users at lists.berlios.de>
> *Subject:* Re: [Opensim-users] [Opensim-dev] Grid Numberging
>
>
>
> Actually, I just realized that the problem is already much worse
> than I thought.
> The practices that we are all doing around opensim completely
> invalidate the concept of uniqueness in UUIDs. This comes from the
> fact that UUIDs are being externalized into DBs and XML
> representations, which are then being imported elsewhere. As
> people reuse entire DBs and XML files to recreate the same worlds
> in different grids, uniqueness goes down the drain. So definitely,
> forget about "Universal" Unique Identification with UUIDs;
> uris/urls are absolutely required for universal naming. The only
> way to savage UUIDs would be use part of their representation to
> deterministically code up the ip/port of where they come from.
>
> Stefan Andersson wrote:
>
> And the whole thread started with suggesting we allocate
> information bits to encode semantics. Which drastically heightens
> the risk of collisions, intentional or malicious.
>
> Just see guids as private to the trust domain, and we'll all fine.
> Grids should be publicly referred to by something riches, as an
> uri/url for example.
>
> Best regards,
> Stefan Andersson
> Tribal Media AB
>
> Join the 3d web revolution : http://tribalnet.se/
>
>
>
> ------------------------------------------------------------------------
>
>
> Date: Fri, 31 Oct 2008 20:10:41 -0700
> From: diva at metaverseink.com <mailto:diva at metaverseink.com>
> To: opensim-users at lists.berlios.de
> <mailto:opensim-users at lists.berlios.de>
> Subject: Re: [Opensim-users] [Opensim-dev] Grid Numberging
>
> Not to mention the dudes and dudettes who generate region UUIDs by
> hand... (a-hem)
> Purely random UUIDs that don't encode higher-order information are
> too fragile when there is a chance that people will get their
> hands on them. I'm not even talking about malicious manipulation,
> just simple human intervention.
>
> Frisby, Adam wrote:
>
> Yeah, the reason I said 'almost' wasn’t because of a good UUID generator, - it's the bad ones.
>
>
>
> IE, people using bad random sources, etc which dramatically increase the chance of a collision - ie if you are only getting 30 bits of randomness, you could run into a collision fairly quickly.
>
>
>
> Adam
>
>
>
>
>
> -----Original Message-----
>
> From: opensim-users-bounces at lists.berlios.de <mailto:opensim-users-bounces at lists.berlios.de> [mailto:opensim-users-
>
> bounces at lists.berlios.de <mailto:bounces at lists.berlios.de>] On Behalf Of Sean Dague
>
> Sent: Friday, 31 October 2008 4:48 PM
>
> To: opensim-users at lists.berlios.de <mailto:opensim-users at lists.berlios.de>
>
> Subject: Re: [Opensim-users] [Opensim-dev] Grid Numberging
>
>
>
> Dr Scofield wrote:
>
>
>
> Frisby, Adam wrote:
>
>
>
> Shouldn’t UUIDs by definition be unique?
>
>
>
>
>
>
>
> IE – if your generating them randomly a collision is almost
>
> guaranteed to never ever occur.
>
>
>
> "almost" being a key word here...
>
>
>
> "...after generating 1 billion UUIDs every second for the next 100
>
> years, the probability of creating just one duplicate would be about
>
> 50%. The probability of one duplicate would be about 50% if every
>
> person on earth owns 600 million UUIDs." <http://en.wikipedia.org/wiki/UuidJusttokeep>
>
> <http://en.wikipedia.org/wiki/UuidJusttokeep>
>
> http://en.wikipedia.org/wiki/Uuid <http://en.wikipedia.org/wiki/UuidJusttokeep>
>
> <http://en.wikipedia.org/wiki/UuidJusttokeep>
>
> Just to keep " <http://en.wikipedia.org/wiki/UuidJusttokeep>almost" in perspective,
>
>
>
> -Sean
>
>
>
> --
>
> Sean Dague / Neas Bade
>
> sdague at gmail.com <mailto:sdague at gmail.com>
>
> http://dague.net <http://dague.net/>
>
>
>
>
>
> _______________________________________________
>
> Opensim-users mailing list
>
> Opensim-users at lists.berlios.de <mailto:Opensim-users at lists.berlios.de>
>
> https://lists.berlios.de/mailman/listinfo/opensim-users
>
>
>
>
>
>
>
>
>
> ------------------------------------------------------------------------
>
>
>
>
>
>
> _______________________________________________
>
> Opensim-users mailing list
>
> Opensim-users at lists.berlios.de <mailto:Opensim-users at lists.berlios.de>
>
> https://lists.berlios.de/mailman/listinfo/opensim-users
>
>
>
>
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> Opensim-users mailing list
> Opensim-users at lists.berlios.de <mailto:Opensim-users at lists.berlios.de>
> https://lists.berlios.de/mailman/listinfo/opensim-users
>
>
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> Opensim-users mailing list
> Opensim-users at lists.berlios.de
> https://lists.berlios.de/mailman/listinfo/opensim-users
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://opensimulator.org/pipermail/opensim-users/attachments/20081103/33fd85cd/attachment.html>
More information about the Opensim-users
mailing list