[Opensim-dev] Suggested in-world DNS: .sim

Thu Feb 25 23:12:30 UTC 2010

Hi

P2P = http://en.wikipedia.org/wiki/Peer-to-peer
P2P as in Point-2-Point is something different. P2P describes a
decentralized network, it does NOT describe standardized server<->client
communication.
This is actually not debatable, this is a stated fact. A PC could belong to
a company or library even though PC stands for Personal Computer. Of course,
you could go into the library and deny that it is a PC... :)

Running a simulator locally to defeat lag/whatever is not feasible. Anyone
who has been programming for more than a couple of days can see the problem
with that. Synchronizing a massive amount of data is one thing, but ensuring
that client and server (or two servers) calculates the exact same thing
always in such complex environment is not something a sane person would
attempt. It is however normal to include physics inside clients to allow
local movement without verifying it with server first. And online games have
protocols optimized for movement (such as sending keystrokes instead of abs
pos). The server is still the authority, but in most cases the client-side
emulation will be in sync. If you have played an online first person shooter
you have probably come across a few glitches as a result of this, for
example being moved back into place, people not being hit even though you
shot straight at them, you being shot even though you managed to duck in
time, etc.
If you look at online games most of them have approximately zero physics
compared to the single-player version of the same game.

So you end up with a region being a region. One central server. You can run
small regions, large regions, square regions, rounded regions, upside-down
regions... but you can't get around the problem that if your foot was in the
way of that rolling prim at that exact spot for ten milliseconds while on
the other computer it rolls on by because your foot-packet had not arrived
yet, then the outcome would be different on the two. The solution would
either to wait for all clients to update each timeframe (both send and
receive), or to receive object coordinates from a centralized authority.
Either way you end up with lag.

Now if you really WANTED to run two OpenSim boxes with the same content then
you would look into other technologies such as how web-caching is done. One
master OpenSim installation and a (third-party) proxy layer that allows
visitors to enter the region without actually being registered there. (i.e.
they only exist in the proxy and the region is read-only) One could allow
only simple signals to be sent to backend such as touching specific objects
or similar. But this is far ahead in time I guess.

BR,
 Tedd

-----Original Message-----
From: opensim-dev-bounces at lists.berlios.de
[mailto:opensim-dev-bounces at lists.berlios.de] On Behalf Of Dzonatas Sol
Sent: 25. februar 2010 20:28
To: opensim-dev at lists.berlios.de
Subject: Re: [Opensim-dev] Suggested in-world DNS: .sim

Peer to peer is just simply box to box, it has nothing to do with "an 
opensim server and a viewer both running on 
the same box", yet don't doesn't mean these boxes can't run opensim and 
a viewer on each box.

It would look like this:

[ viewer <-> opensim ] <-> [ opensim <-> viewer ]

That's peer to peer.

If both opensim's simulate the exact same region, there is no need to 
send any information between the two opensim. If the "observer" and the 
"observed" don't change, that communication is unaffected, therefore the 
simulator don't need to update each other.

I can think of many examples where this has not failed.

Cristina Videira Lopes wrote:
> You're making up your own definition of p2p in the context of opensim,  
> your definition being "an opensim server and a viewer both running on  
> the same box". That's not what p2p is, generally, although one could  
> imagine a distribution that would have those as a unit. For "true" p2p  
> VWs see OpenCobalt. OpenCobalt is both a server and a browser in the  
> same piece of software, one doesn't exist without the other, it's not  
> even modeled as server-and-client, it's a unit.
>
> OpenSim is client-server. Even if a client and a server run on the  
> same box, the architecture is still client-server. Your friend's  
> viewer in his laptop needs to connect to your sim in your laptop in  
> order to visit your world; at that point, it's up your opensim server  
> to simulate your friend's agent and to serve all simulated data.  
> That's where latency plays a role --home DSLs aren't good enough to  
> support many clients sucking data from them. Also, your friend's  
> opensim server in his latptop plays no role in the simulation of your  
> friend's agent in your opensim server -- other than serving identity  
> and inventory, in the case of HG, but not simulation.
>
> So, again, not p2p as OpenCobalt. There's an interesting technical  
> discussion waiting to happen comparing all these possible  
> architectures. But let's start by not confusing things. OpenSim's  
> architecture is completely different from OpenCobalt's.
>
> On Feb 25, 2010, at 10:56 AM, Dzonatas Sol wrote:
>
>   
>> That doesn't answer my question.
>>
>> There are two opensim boxes, with each connected directly to each  
>> other,
>> peer to peer.
>>
>> Are you saying, of only these two boxes, that neither of them can a  
>> client?
>>
>> I don't see the latency issue here when the client is directly on the
>> opensim box.
>>
>>
>> Melanie wrote:
>>     
>>> I'm saying that one of them will run the sim. The other connects to
>>> it. Even if the other also runs a sim, the sims will not be
>>> connected. Much less cooperatively simulating.
>>>
>>> So, the one being connected to needs sufficient upstream bandwidth
>>> as well as an externally identifiable name.
>>>
>>> Which means an ICANN-approved TLD. Not a fantasy TLD.
>>>
>>> Melanie
>>>
>>> Dzonatas Sol wrote:
>>>
>>>       
>>>> If one opensim box connects to another opensim box, that is,
>>>> technically, peer to peer.
>>>>
>>>> So, are you saying an opensim box cannot run a client at the same  
>>>> time?
>>>>
>>>> Melanie wrote:
>>>>
>>>>         
>>>>> Hi,
>>>>>
>>>>> peer to peer simulation is not practical for many different  
>>>>> reasons.
>>>>> Latency being the chief one.
>>>>>
>>>>> OpenSim is not going to be a peer to peer system, therefore your
>>>>> suggestion is off topic. Opensim doesn't need another TLD, and it  
>>>>> is
>>>>> not what you are envisioning. OpenSim firmly embraces the concept  
>>>>> of
>>>>> SERVER SIDE simulation, therefore every sim will always have a
>>>>> central server.
>>>>>
>>>>> I believe this has gone as far as it will go and if there is any
>>>>> more name calling, well, we'll just have to moderate some people,
>>>>> won't we?
>>>>>
>>>>> Melanie
>>>>> (Core)
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>           
>>>> _______________________________________________
>>>> Opensim-dev mailing list
>>>> Opensim-dev at lists.berlios.de
>>>> https://lists.berlios.de/mailman/listinfo/opensim-dev
>>>>
>>>>
>>>>
>>>>         
>>> _______________________________________________
>>> Opensim-dev mailing list
>>> Opensim-dev at lists.berlios.de
>>> https://lists.berlios.de/mailman/listinfo/opensim-dev
>>>
>>>
>>>       
>> _______________________________________________
>> Opensim-dev mailing list
>> Opensim-dev at lists.berlios.de
>> https://lists.berlios.de/mailman/listinfo/opensim-dev
>>     
>
> _______________________________________________
> Opensim-dev mailing list
> Opensim-dev at lists.berlios.de
> https://lists.berlios.de/mailman/listinfo/opensim-dev
>
>   

_______________________________________________
Opensim-dev mailing list
Opensim-dev at lists.berlios.de
https://lists.berlios.de/mailman/listinfo/opensim-dev