bandwidth (was Re: AG reservation)

[Rick Stevens]

Tom:  This doesn't completely answer your questions but gives some
insight as to where we are heading with session management.

When we release prototype AG-Metro (One of the tools for management of
multiple simultaneous sessions) Session management and configuration will 
be done via
a Web server interface to a shared AG-Metro database that will set up and 
maintain
a set of virtual sessions that get instantiated when one or more AG nodes
attaches. (The specific choice of multicast addresses etc., will be
done in context of the global Mbone usage and the AG users will not
need to know about these levels of details, and so these can change
without altering the use of the AG). AG-Metro will support multiple
simultaneous sessions mapped to virtual rooms, aggregated into a virtual
meeting place (like a virtual conferencing center will multiple meeting 
rooms)..
You (and your AG-Node) can move from one virtual room to another and the
stream and session management will be handled automatically by the system.
AG-Metro will require a consistent AG node configuration at all the sites
to enable the automated session management features to work (hence our
push for common node configurations).  We have demonstrated a early version
of this last week at our DOE site visit and are working a more robust
version for early deployment to AG sites.. One key requirement for
this is that the sites be running the standard 2 Linux/NT/Win98 configuration
for the AG-Metro interfaces need the Linux/NT resources to be identical at
each of the sites to work (at least in the current prototype version)..  I
hope we can do some of these multiple virtual venue experiments before
SC99 and then again at SC99 :-)

Currently for manual session creation, one needs to use a tool like sd or
manually choose known mcast addresses to create the session and set
parameters for the session, keeping in mind that the Mbone is shared
community resource. Bridges if any have to be manually configured. etc.
We are planning an AG service called the "network flow engine" that would help
manage both mcast and unicast streams and interfaces to quality of service
mechanisms...

More about this at the AG-Tech R&D meeting..


At 02:27 PM 10/1/99 , Tom Coffin wrote:
>great dialogue...
>what about sessions?
>
>How do we set up a accessgrid session
>that others can vic and rat into?
>
>Are there specific hardware requirements
>for the machines hosting these sessions?
>if so, what are they?
>
>_______________________________________________
>At 11:28 AM 10/1/99 -0500, Ivan R. Judson wrote:
>    >
>    >A small historical note:
>    >
>    >People have been making this technology require less and less of the
>bridge
>    >type technology; from the beginning with DVMRP, and mrouted's (where
>mrouted
>    >was tunneled to other mrouted's) and the first Mbone where the whole 
> thing
>    >was tunnels, people have known this was the wrong solution.
>    >
>    >This is why native multicast has been developed...
>    >
>    >--Ivan
>    >
>    >> -----Original Message-----
>    >> From: owner-ag-tech at mcs.anl.gov [mailto:owner-ag-tech at mcs.anl.gov]On
>    >> Behalf Of Tony Rimovsky
>    >> Sent: Friday, October 01, 1999 11:22 AM
>    >> To: Tom Coffin
>    >> Cc: ag-tech at mcs.anl.gov
>    >> Subject: Re: bandwidth (was Re: AG reservation)
>    >>
>    >>
>    >> In my opinion, the 'magic' of bridges should be limited to the greatest
>    >> extent possible.  If a site can use native multicast, it should.
>    >> If a site
>    >> doesn't have native multicast, it should be working on it.  As Bill
>    >> mentioned on the moo a couple of weeks ago, for about $3000 in
>    >> hardware, any
>    >> site can be fully multicast capable.
>    >>
>    >> Don't get me wrong.  The bridges have saved the bacon of many
>    >> demos and will
>    >> be of great utility until the wide-area performance/stability 
> issues get
>    >> resolved .  However, there are scaling and design issues involved
>with any
>    >> sort of widespread deployment of the bridging, and I do not thing
>    >> one of the
>    >> access-grid goals should be to create a variation of the MBone 
> tunneling
>    >> mess.
>    >>
>    >> /tsr
>    >>
>    >> On Fri, 01 Oct 1999, Tom Coffin wrote:
>    >>
>    >> > when do we learn the magic of creating our own bridges and sessions?
>    >>
>    >
>    >
>
>
>___________________________________________________________
>Tom Coffin .......................... tcoffin at ncsa.uiuc.edu