[Opensim-dev] Packet Pooling - Should it work?
Diva Canto
diva at metaverseink.com
Fri Aug 16 22:31:27 UTC 2019
Both simulators and viewers need to access the grid's assets, so they
all need to know where to get them from. For the simulators, it's
simple: the addresses of all backend services are given as configuration
variables in things like this, in GridCommon, or wherever:
[AssetService]
;
; Change this to your grid-wide asset server. Do not add a slash
to the end of any of these addresses.
;
AssetServerURI = "${Const|BaseURL}:${Const|PrivatePort}"
But what about the viewers? Where do viewers get the information about
where to download textures/meshes/etc from?
This information is not hardcoded grid-wide anywhere. It's not even
hardcoded upon login. Viewers get all this information from the
simulators they connect to, as capability URLs. This is interesting
because it allows for dynamic associations -- your viewer first connects
to simulator1, it gets assets from wherever simulator1 tells it to; but
if later the viewer connects to simulator2, it gets the assets from
wherever that simulator tells it to; it may be different. The Linden
viewer is engineered to receive these capability URLs and they are,
indeed, quite flexible. (this is what allows the hypergrid to work with
the Linden viewer in the first place)
So where do we define the capability URLs that the simulators send to
the viewers? They are all in OpenSim.ini, under the section
[ClientStack.LindenCaps]. Here is that section as it currently stands in
OpenSim.ini.example:
[ClientStack.LindenCaps]
;; For the long list of capabilities, see OpenSimDefaults.ini
;; Here are the few ones you may want to change. Possible values
;; are:
;; "" -- empty, capability disabled
;; "localhost" -- capability enabled and served by the simulator
;; "<url>" -- capability enabled and served by some other server
;;
; These are enabled by default to localhost. Change if you see fit.
Cap_GetTexture = "localhost"
Cap_GetMesh = "localhost"
Cap_AvatarPickerSearch = "localhost"
Cap_GetDisplayNames = "localhost"
GetTexture and GetMesh are the two potentially most impactful services,
because they serve relatively large data. But, as the comment says,
there's many others, and they can all be redirected elsewhere. These
variables are all defaulted to "localhost" meaning that the service is
done by the simulator. If you want it to be some other server, you just
need to set it differently. For example:
Cap_GetTexture = "http://mygrid.com:8206/CAPS/textures"
The viewer will get this new URL to fetch textures from, away from the
simulator.
Now, having set that, you need to make sure that the service actually
exists at that other URL. This can be done by placing a Robust service
at that end point serving this capability URL. Here's the configuration
of a simple Robust server that serves only the textures and nothing
else, and that is simply a texture-only interface on top of the central
asset service:
[Network]
port = 8206 ; make sure to open it in the firewall
[ServiceList]
;; GetTexture
GetTextureConnector =
"OpenSim.Capabilities.Handlers.dll:GetTextureServerConnector"
[CapsService]
AssetService = "OpenSim.Services.AssetService.dll:AssetService"
[DatabaseService]
StorageProvider = "OpenSim.Data.MySQL.dll"
ConnectionString = "..."
[AssetService]
LocalServiceModule = "OpenSim.Services.AssetService.dll:AssetService"
AssetLoaderEnabled = false
There's a lot more to it, but this will be enough to configure something
that pulls texture-servicing away from the simulator. (Similar for Mesh)
I'm explaining the basic mechanism. But you need to think about the data
replication architecture that best fits your needs. There are many
options, some more complicated than others. In fact, the default
configuration (sim serves textures) is one point in the data replication
design space, where textures are replicated in every simulator that ever
served them. It increases the load in the simulators, but decreases the
load of serving textures in the central server. But, as I said, there
are many more options. This is not something that has one single / best
configuration; it depends on many things.
Nothing of this is documented, as usual. There is enough information in
this email that a capable person can trace it in the code -- search for
some of the configuration sections I mention above and trace it from there.
I won't give customized solutions, and I don't have time to write
documentation, but I'll be happy to answer more concrete questions that
benefit everyone.
Good luck!
On 8/16/2019 11:14 AM, Mike Dickson wrote:
> Thanks Diva, that would be much appreciated and I think helpful to a great many people.
>
> Sent from Mail for Windows 10
>
> From: Diva Canto
> Sent: Friday, August 16, 2019 9:53 AM
> To: opensim-dev at opensimulator.org
> Subject: Re: [Opensim-dev] Packet Pooling - Should it work?
>
> [I meant to send this to the list]
> I'll send some info on where to look later today.
>
> -------- Forwarded Message --------
> Subject: Re: [Opensim-dev] Packet Pooling - Should it work?
> Date: Thu, 15 Aug 2019 17:27:38 -0700
> From: Diva Canto <diva at metaverseink.com>
> To: Mike Dickson <mike.dickson at utopiaskye.com>
>
> Hi,
>
> You can already remove almost everything out of the simulator with the
> right configurations.
>
> I don't recommend serving the large assets, like textures and mesh, from
> one single server because that one quickly becomes the bottleneck. But
> there are may ways of serving replicating the data.
>
> I ran a large grid that had a completely different arrangement than the
> default: each group of regions that were related (i.e. an "estate") had
> its own set of asset services. Users' assets in inventory were served by
> yet a separate server. All this has been possible for a very long time.
>
> Diva
>
> On 8/15/2019 1:07 PM, Mike Dickson wrote:
>> I'm actually not so much trying to optimize network traffic (that is
>> another great topic and I personally think what core should focus on is
>> protocols rather than code). I'm concerned there is an issue with garbage
>> collection limiting scaling of the simulator. Basically I'm of the opinion
>> that really the only thing that should be running in the simulator is...
>> the simulation. So as much as possible pulling things out (Moving to AISv3
>> for inventory out of process, SSB out of process, etc would all be good
>> additional steps). It's true if you're just running a standalone thats
>> probably overkill but I'm trying to support large regions and scale. Hence
>> yes the whole stack is really the focus.
>>
>> Mike
>>
>> On Wed, Aug 14, 2019 at 10:54 PM Mister Blue <misterblue at misterblue.com>
>> wrote:
>>
>>> There have been many attempts at optimizing the network traffic from the
>>> simulator to the clients. GP optimization confuses low level networking
>>> (queuing, ..) with application level (object updates before wind updates).
>>>
>>> Make sure you're thinking of the whole virtual world stack.
>>>
>>> On Wed, Aug 14, 2019 at 4:07 PM Mike Dickson <mike.dickson at utopiaskye.com>
>>> wrote:
>>>
>>>> So after doing some research I think my fix to this is to get as many of
>>>> the big allocations out of the region server as possible and secondarily
>>> to
>>>> get the rest coming from a pool where I can. I think that translates to
>>> the
>>>> follow projects:
>>>>
>>>> 1) Move GetMesh and GetTexture out of process and into a separate server
>>>> 2) Get the rest of the UDP allocations coming from a pool.
>>>>
>>>> For #1 there was code originally done for InWorldz/Halcyon to do that. I
>>>> can try and ressurect it and interface it to the asset service (ideally
>>>> through the local asset cache) but I think alternatively a redo makes
>>> more
>>>> sense.
>>>> For #2 there is buffer pooling code in LibOMV originally in Halcyon that
>>> I
>>>> believe Cinder got upstream via Latif a while back.
>>>>
>>>> And yes I did run with -desktop mode in Mono for some time. It sort of
>>>> bandaids things a bit but when a GC pass does happen UDP stalls until the
>>>> GC completes and the protocol recovers (for reliable messages). If you
>>>> extend that to a busy region with 20, 30 or more avatars it falls apart
>>>> quickly. Especially with everyone wearing mesh. Still probably better
>>> than
>>>> the standard GC but the real fix is to stop making garbage to collect.
>>>>
>>>> Mike
>>>>
>>>> Sent from Mail for Windows 10
>>>>
>>>> From: Leal Duarte
>>>> Sent: Tuesday, August 13, 2019 1:43 PM
>>>> To: opensim-dev at opensimulator.org
>>>> Subject: Re: [Opensim-dev] Packet Pooling - Should it work?
>>>>
>>>> PacketPool.cs code has been in usage, and still is in same packets, but
>>> of
>>>> limited usefulness
>>>> in same cases it could even be GC induced pseudo memory leak.
>>>>
>>>> It was replaced by a simpler pool of memory buffers (actually libomv
>>>> UDPPacketBuffers, but not its objectpool) on most sent packets and
>>> receive
>>>> buffers
>>>> also used as temporary work buffers on a few other places.
>>>> Most send packets have nothing to reuse but the buffer.
>>>>
>>>> Try running opensim in Workstation (desktop on mono) mode.
>>>> Server mode heuristics don't seem to match opensim needs that well.
>>>>
>>>> Ubit
>>>>
>>>> On 13-Aug-19 16:04, Mike Dickson wrote:
>>>>> I've been investigating UDP stalls for a while now and at least in some
>>>>> cases I'm fairly convinced some cases occur due to GC pauses. There is
>>>>> some packet pooling code in the underlying LibOMV probably originally
>>>>> derived from work done on Halcyon to address this case. I don't see
>>> any
>>>>> attempt in the UDP comms to make use of these buffer pools.
>>>>>
>>>>> There is seperate code in PacketPool.cs to, I think reuse packet
>>> buffers
>>>>> based on a couple of buffer sizes and it looks like this should be on
>>> by
>>>>> default but I can't find any evidence by looking at status of any
>>> packet
>>>>> reuse occuring. That is it looks like there is code there but it's
>>>> either
>>>>> switched off somewhere else or just doesn't work (or the stats are
>>> wrong
>>>> :).
>>>>>
>>>>> Should this PacketPooling be functional? Alternatively has any attempt
>>>>> been made to wire in the PacketBuffer support thats already in LibOMV?
>>>>> I'm going to dig through all this as I have time but I figured a
>>> little
>>>>> information might help short circuit some paths and direct my search.
>>>>>
>>>>> Thanks!
>>>>>
>>>>> Mike
>>>>> _______________________________________________
>>>>> Opensim-dev mailing list
>>>>> Opensim-dev at opensimulator.org
>>>>> http://opensimulator.org/cgi-bin/mailman/listinfo/opensim-dev
>>>>>
>>>> _______________________________________________
>>>> Opensim-dev mailing list
>>>> Opensim-dev at opensimulator.org
>>>> http://opensimulator.org/cgi-bin/mailman/listinfo/opensim-dev
>>>>
>>>> _______________________________________________
>>>> Opensim-dev mailing list
>>>> Opensim-dev at opensimulator.org
>>>> http://opensimulator.org/cgi-bin/mailman/listinfo/opensim-dev
>>>>
>>> _______________________________________________
>>> Opensim-dev mailing list
>>> Opensim-dev at opensimulator.org
>>> http://opensimulator.org/cgi-bin/mailman/listinfo/opensim-dev
>>>
>> _______________________________________________
>> Opensim-dev mailing list
>> Opensim-dev at opensimulator.org
>> http://opensimulator.org/cgi-bin/mailman/listinfo/opensim-dev
>>
>>
>
> _______________________________________________
> Opensim-dev mailing list
> Opensim-dev at opensimulator.org
> http://opensimulator.org/cgi-bin/mailman/listinfo/opensim-dev
>
> _______________________________________________
> Opensim-dev mailing list
> Opensim-dev at opensimulator.org
> http://opensimulator.org/cgi-bin/mailman/listinfo/opensim-dev
>
>
More information about the Opensim-dev
mailing list