Re: still troubles

[Timothy Stack <stack@cs.utah.edu>]

> > We can't really do dynamic changes beyond toggling resource controls on
> > and off, or trashing and rebuilding the objects that reserve the
> > resources.
> >
> 
> So IF I had a working flow, let's say to transmit a video, and suddenly had 
> another video flow sending the same type of capsules and requiring the same 
> forwarding behaviour, would I get new resources or would it share the ones of 
> the previous flow of the same type ??

Yes, although it isn't too hard to mix a `user id' in to change the packet
header and cause it to be demuxed into a different flow with the default
set of resources.

> If I wanted to double the resources (to have one flow instance for both 
> videos but with twice as much resources),  I would have to destroy the 
> pre-existing flow "object" associated with that capsule type, and instantiate 
> a new one with twice as much resources ??
> 
> [ I have understood that changing the amount of allocated resources is not 
> yet possible, but theoretically, would this be the way ??] 

Well, i would prefer the above solution since its like adding another
dimension to the system, so we have the protocol type _and_ the principal
that is using it.

> > Yeah, there are only so many resources to go around, so any other stuff
> > falls back on best effort (use whats left after everyone else has done
> > their thing).
>
> That's quite curios... I thought every ANTS protocol gets associated with a 
> nodeOS flow, so what would the other stuff be that falls back to best effort ?

Active network protocols are required to work over legacy routers, so in
the event that we can't create a new flow for a protocol we can just act
like a regular IP router and do some default forwarding
behavior.  However, we don't actually do this at the moment, although
there probably aren't enough ANTS applications in the world to exhaust all
the resources.

> > > Or do they get preempted at some point ?
> >
> > hmm, not sure what you mean by this.
> 
> Well it is two questions in one:
> 1) if all the resources on a node are given, what happens with new capsules 
> that haven't been associated to a flow ? Do they get discarded ??

When a packet that belongs to a protocol that a node doesn't know
anything about, it will ask the previous node for some information about
the protocol.  Then it will try to create the flow, download the code into
it and install demux keys to capture any packets belonging to that
protocol.  Of course, while this is happening some packets might be
dropped since it can run out of buffer space before the forwarding code is
all set up.

> 2) How long does the resource assignment last ?? Only as long as the 
> forwarding routine is executed, or as long as the ANTS protocol is in use ??
> If an ANTS protocol runs for an indefinite amount of time, are the associated 
> resources taken away or not ?? (that's what I mean by preemption, but in 
> NodeOS terminology it probably means that the domain gets terminated)

Currently, the protocol hangs around forever...  Which kinda sucks...

> > Well, you can always just use the jnodeos directly.  ANTS might be
> > overkill for what you want to do and we've actually run jnodeos on a bare
> > machine quite a bit.
> 
> Well what I like about the ANTS concept is that the Application knows what 
> type of capeule it is sending, so I am thinking about something that may be 
> classified as an application-based DiffServ, where the applications sends 
> Capsules with different "codepoints" and the forwarding behaviour at the 
> nodes depends on this codepoint value.
> Some capsules may get dropped (let's say in case of congestion)
> and some get forwarded, and I would like to guarantee  that the forwarded 
> ones always get through, so that's the fuzz about resource reservation :)

Just to be clear though, you can do this on the jnodeos too, the main
difference is that the protocol code isn't automatically distributed for
you.  Which isn't a problem if you're just experimenting and have control
over every router that will be forwarding packets.

> thanks 
> 
> Andreas... 
> 
> P.S. Balazs is my "supervisor" in what I am doing, so the mail asking  for 
> the binaries was related to my work...  
> One step at the time I think I might be able to get the whole thing running,
> but it was good to hear that in case of an emergency a binary would
> also be available :)

okie

tim stack