[Q] simulation: robotic avatars and time reference

Hello David,

On 6/10/07, David P. Reed <[hidden email]> wrote:

> The "replicated computation" model takes some getting used to.   A
> "local" file or database is properly thought of as a non-replicated file
> within the replicated computational world.   That means that it can only
> be *directly* accessed by a non-replicated portion of the computation.
> So the way to do this in Croquet is to start a non-replicated activity
> on the machine that has the file or database, and have it communicate
> events that are replicated to all replicated island instances of a
> replicated computation.
>
> That's the answer to (a).

According to 'Island Time' section in Croquet's 'System Overview' [1]

External non-replicated messages are extremely dangerous and must be avoided. They do play a role, but it is extremely rare that anyone will ever have a need to make use of this mechanism. The problem is obviously that if a non-replicated message is executed and happens to modify the state of an Island it breaks the determinism the Island shares with the other replicated copies. We do use such non-replicated message when rendering the contents of an Island, but this is extremely well controlled.

So I have to be extremely careful and probably take inspiration form the code
rendering for local participant: for example CroquetHarness>>setupLocal in the SDK
(where a non-replicated portal and overlay are created to serve the local participant)

[1] Island Time in Croquet's System Overview
    http://croquetconsortium.org/index.php/System_Overview#Island_Time

>
> (b) is straightforward - just create a mapping (any one-to-one and onto
> mapping will do) between simulation time and Croquet time.   Such a
> mapping can be at a relative rate and start time, or even a non-linear
> mapping.   Run the simulation up to the corresponding Croquet time.

This boils down to keep some global simulation counter (simulation time
initialized with island time at the start of the simulation) and use
Croquet time, aka TIsland>>time, doesn't it?

Thanks for your reply
Bye
Enrico


>

>
> Enrico Spinielli wrote:
> > Hi all,
> > I am writing a prototypical air traffic simulation application.
> > The sky is partitioned into various entities: airspaces,
> > traffic volumes, ...
> > Aircrafts will follow flight plans.
> > Flight plans are described in 4D (3D + time) thru
> > entities such as: en-route routes, points, flight levels,
> > takeoff time, relative times over (points)...
> >
> > I would like to have
> > 1. Croquet participants (robots, not "human" avatars)
> >    representing different (group of) airports
> >    instantiating flight from a local data source (file, db,...)
> > 2. instantiation should be time driven: a new flight enters
> >    into the simulation at takeoff time as specified in its
> >    flight plan
> > 3. human avatars will be able to ask for relevant aggregated
> >    information (table/graph of flight counts per hour, ...)
> >    for their geographical area of competence (aerodrome,
> >    airspace).
> >
> > Given the above, my questions for Croquet developers/experts are:
> > a. is it really possible to have a participant which creates
> >    new instances based on local machine information (file, db...)?
> >    If not, do I only have the possibility to create instances in a
> >    centralized way?
> > b. is there any possibility to define a reference simulation start
> >    time (at island level?) shared by all participants?
> >    Is it OK for example to do it in (a subclass of)
> >    CroquetParticipant?
> >
> > Any helps/ideas is welcome
> > Thanks in advance
> > Regards
> > Enrico
> >
> >
>

On Jun 12, 2007, at 3:33 PM, Enrico Spinielli wrote:

Hello David,

On 6/10/07, David P. Reed <[hidden email]> wrote:
> The "replicated computation" model takes some getting used to.   A
> "local" file or database is properly thought of as a non-replicated file
> within the replicated computational world.   That means that it can only
> be *directly* accessed by a non-replicated portion of the computation.
> So the way to do this in Croquet is to start a non-replicated activity
> on the machine that has the file or database, and have it communicate
> events that are replicated to all replicated island instances of a
> replicated computation.
>
> That's the answer to (a).
According to 'Island Time' section in Croquet's 'System Overview' [1]

External non-replicated messages are extremely dangerous and must be avoided. They do play a role, but it is extremely rare that anyone will ever have a need to make use of this mechanism. The problem is obviously that if a non-replicated message is executed and happens to modify the state of an Island it breaks the determinism the Island shares with the other replicated copies. We do use such non-replicated message when rendering the contents of an Island, but this is extremely well controlled.

So I have to be extremely careful and probably take inspiration form the code
rendering for local participant: for example CroquetHarness>>setupLocal in the SDK
(where a non-replicated portal and overlay are created to serve the local participant)

[1] Island Time in Croquet's System Overview
    <a href="http://croquetconsortium.org/index.php/System_Overview#Island_Time" target="_blank" onclick="return top.js.OpenExtLink(window,event,this)">http://croquetconsortium.org/index.php/System_Overview#Island_Time

It's only non-replicated messages sent to replicated (on-island) objects that are dangerous.  Non-replicated messages are safely sent outside of the island all the time.  For example, keyboard events  run though quite a bit of Squeak/Morphic code before they result in replicated Croquet events.  As another example, take a look at the webcam code in the SDK.  When a new video frame is grabbed, a replicated message is sent into the island with the contents of the frame (this should be quite similar to noticing that some relevant flight-plan event should occur, and sending the appropriate replicated message); this is perfectly safe.

>
> (b) is straightforward - just create a mapping (any one-to-one and onto
> mapping will do) between simulation time and Croquet time.   Such a
> mapping can be at a relative rate and start time, or even a non-linear
> mapping.   Run the simulation up to the corresponding Croquet time.

This boils down to keep some global simulation counter (simulation time
initialized with island time at the start of the simulation) and use
Croquet time, aka TIsland>>time, doesn't it?

That is one way to do it.  In fact, the mapping between simulation time and Croquet time need not be bijective; for example, it wouldn't be if you wanted to be able to pause or rewind the simulation while still being able to interact in the space.

Josh

Thanks for your reply
Bye
Enrico


>
>
> Enrico Spinielli wrote:
> > Hi all,

> > I am writing a prototypical air traffic simulation application.
> > The sky is partitioned into various entities: airspaces,
> > traffic volumes, ...
> > Aircrafts will follow flight plans.
> > Flight plans are described in 4D (3D + time) thru
> > entities such as: en-route routes, points, flight levels,
> > takeoff time, relative times over (points)...
> >
> > I would like to have
> > 1. Croquet participants (robots, not "human" avatars)
> >    representing different (group of) airports
> >    instantiating flight from a local data source (file, db,...)
> > 2. instantiation should be time driven: a new flight enters
> >    into the simulation at takeoff time as specified in its
> >    flight plan
> > 3. human avatars will be able to ask for relevant aggregated
> >    information (table/graph of flight counts per hour, ...)
> >    for their geographical area of competence (aerodrome,
> >    airspace).
> >
> > Given the above, my questions for Croquet developers/experts are:
> > a. is it really possible to have a participant which creates
> >    new instances based on local machine information (file, db...)?
> >    If not, do I only have the possibility to create instances in a
> >    centralized way?
> > b. is there any possibility to define a reference simulation start
> >    time (at island level?) shared by all participants?
> >    Is it OK for example to do it in (a subclass of)
> >    CroquetParticipant?
> >
> > Any helps/ideas is welcome
> > Thanks in advance
> > Regards
> > Enrico
> >
> >
>

I think of correct Croquet applications as having two kinds of non-
replicated computation:

I) Things That Don't Matter.  Look but don't touch.  This is how
rendering works. You can do pretty much whatever you want if you
don't actually change anything in the Island.

II) Things That Do Matter.  A system with no input is pretty boring.

The way we inject input into the system is to have non-replicated
computation (e.g., a user interface on one participant) send a
message to an object in the Island. Croquet automatically replicates
that message to all participants.  So my keyboard and mouse are not
replicated, nor are the computation that surrounds them. But this
computation eventually results in one or more replicated messages.

Your application is simply replacing the keyboard/mouse computation
with some other set of non-replicated computation that uses a
database.  That is, they're "Things That Do Matter." Inputs.  Croquet
is designed to make this fairly easy to do (e.g., it is a network
architecture that takes care of replicating the to-Island messages
for you).

By contrast, the passage you quoted can be thought of as scaring you
off from trying to do stuff that you think are "Things That Don't
Matter." I don't think that's the case here.

I sometimes LOOSELY think of the above as:
    non-replicated stuff producing replicated inputs = OK
    non-replicated output = be careful
    non-replicated side-effects to Island = NEVER

On Jun 12, 2007, at 5:33 PM, Enrico Spinielli wrote:

> Hello David,
>
> On 6/10/07, David P. Reed <[hidden email]> wrote:
> > The "replicated computation" model takes some getting used to.   A
> > "local" file or database is properly thought of as a non-
> replicated file
> > within the replicated computational world.   That means that it
> can only
> > be *directly* accessed by a non-replicated portion of the
> computation.
> > So the way to do this in Croquet is to start a non-replicated
> activity
> > on the machine that has the file or database, and have it
> communicate
> > events that are replicated to all replicated island instances of a
> > replicated computation.
> >
> > That's the answer to (a).
> According to 'Island Time' section in Croquet's 'System Overview' [1]
> External non-replicated messages are extremely dangerous and must
> be avoided. They do play a role, but it is extremely rare that
> anyone will ever have a need to make use of this mechanism. The
> problem is obviously that if a non-replicated message is executed

> and happens to modify the state of an Island it breaks the

> determinism the Island shares with the other replicated copies. We
> do use such non-replicated message when rendering the contents of
> an Island, but this is extremely well controlled.
> So I have to be extremely careful and probably take inspiration
> form the code
> rendering for local participant: for example
> CroquetHarness>>setupLocal in the SDK
> (where a non-replicated portal and overlay are created to serve the
> local participant)
>
> [1] Island Time in Croquet's System Overview
>     http://croquetconsortium.org/index.php/System_Overview#Island_Time
>
> >
> > (b) is straightforward - just create a mapping (any one-to-one
> and onto
> > mapping will do) between simulation time and Croquet time.   Such a
> > mapping can be at a relative rate and start time, or even a non-
> linear
> > mapping.   Run the simulation up to the corresponding Croquet time.
>
> This boils down to keep some global simulation counter (simulation
> time
> initialized with island time at the start of the simulation) and use
> Croquet time, aka TIsland>>time, doesn't it?
>
> Thanks for your reply
> Bye
> Enrico
>
>
> >
> >
> > Enrico Spinielli wrote:
> > > Hi all,
> > > I am writing a prototypical air traffic simulation application.
> > > The sky is partitioned into various entities: airspaces,
> > > traffic volumes, ...
> > > Aircrafts will follow flight plans.
> > > Flight plans are described in 4D (3D + time) thru
> > > entities such as: en-route routes, points, flight levels,
> > > takeoff time, relative times over (points)...
> > >
> > > I would like to have
> > > 1. Croquet participants (robots, not "human" avatars)
> > >    representing different (group of) airports
> > >    instantiating flight from a local data source (file, db,...)
> > > 2. instantiation should be time driven: a new flight enters
> > >    into the simulation at takeoff time as specified in its
> > >    flight plan
> > > 3. human avatars will be able to ask for relevant aggregated
> > >    information (table/graph of flight counts per hour, ...)
> > >    for their geographical area of competence (aerodrome,
> > >    airspace).
> > >
> > > Given the above, my questions for Croquet developers/experts are:
> > > a. is it really possible to have a participant which creates
> > >    new instances based on local machine information (file, db...)?
> > >    If not, do I only have the possibility to create instances in a
> > >    centralized way?
> > > b. is there any possibility to define a reference simulation start
> > >    time (at island level?) shared by all participants?
> > >    Is it OK for example to do it in (a subclass of)
> > >    CroquetParticipant?
> > >
> > > Any helps/ideas is welcome
> > > Thanks in advance
> > > Regards
> > > Enrico
> > >
> > >
> >