From Whence A 100 Second Delay

Question

I have searched this forum, for which Alex seems to give very prompt, very thoroughly-researched replies, and have Googled the 'rest of the world'  So, probably, I should be able to resolve this like the lawyers when they become perturbed over redundancy:  "Asked and Answered"!  But sorry, if this has been asked and answered, it has been in a different context, or with different enough particulars that none of the information, I have garnered so far has allowed me to find and fix, or at least find and understand exactyl what is going on.
Here's the infrastructure pieces:  Flex SDK 3.3, AIR 1.5, Blaze 3.2, FMS 3.5 -- it ought to make Adobe weep with joy -- I think everything is at a latest stable release level.
Let's start with what works:  When the Flex sub-applications are running inside my AIR application, RemoteObject calls with a streaming-amf channel behave beautifully.  I am not clear how you define -- as with the most recent posting here -- "Multi-Calls".  A behavior that the application certainly relies upon is this:  A.) Some remote object call is made to get a list of objects [requiring IExternalizable handling on the ActionsScript side and Externalizable handling on the Java side].  B.) Some PopUp dialog allows the user to modify that list -- usually to add a new object.  C.)  To prevent duplicates, that PopUp code will make a new remote object call to check that or some related list of objects.  D.)  If no duplicate is found, a remote object call to write [commit] the new object will be made, but if there is a duplicate a subordinate PopUp dialog [not Alert] will be presented, and if overwrite is confirmed by the user, the remote object write will occur at that level.  I don't know if that describes "Multi-Calls", but it does describe a "conversation" over the connection that is something other than a simple request-response.  Bravo and thanks to the Adobe team for a suite of APIs that make this all possible.
What else works:  One of the most popular Flex-only sub-applications that lives inside the AIR  client, can be accessed from the same servers -- Tomcat 6 and FMS 3.5 -- via their browser.  At start up that Flex-only sub application has this:
var channelSet:ChannelSet = new ChannelSet();
channelSet.addChannel (new StreamingAMFChannel ("my-streaming-amf",
        "http://our.site.org:8886/music/messagebroker/streamingamf"));
roPerf = new RemoteObject ("performanceMAC");
roPerf.channelSet = channelSet;
roPerf.addEventListener (ResultEvent.RESULT, remoteResultHandler);
roPerf.addEventListener (FaultEvent.FAULT, remoteFaultHandler);
roProg = new RemoteObject ("programMAC");
roProg.channelSet = channelSet;
roLI = new RemoteObject ("leadInMAC");
roLI.channelSet = channelSet;
roLI.addEventListener (ResultEvent.RESULT, remoteResultHandler);
roLI.addEventListener (FaultEvent.FAULT, remoteFaultHandler);
establishFMSConnection();
When that application is accessed via Firefox 3, it behaves just as it does when it is being 'played' by the AIR runtime.  When I launch that appliction with IE 8, it hangs for 100 seconds, and then plays perfectly.  I can attach two log file snapshots run with this setting in services.xml:
<target class="flex.messaging.log.ServletLogTarget" level="Debug">
but, since it is quite verbose and exposes content that I would probably have to edit out, I have just reduced it to what I think matters -- if you know what to look for, which I certainly do not:
Firefox Success:
Apr 29, 2009 9:39:20 AM org.apache.catalina.core.ApplicationContext log
INFO: [MUSIC]04/29/2009 09:39:20.093 [INFO] [Endpoint.FlexSession] Number of streaming clients for FlexSession with id 'E58902CEDE668A919A8344C917AA8F07' is 1.
Apr 29, 2009 9:39:20 AM org.apache.catalina.core.ApplicationContext log
INFO: [MUSIC]04/29/2009 09:39:20.093 [DEBUG] [Endpoint.StreamingAMF] Number of streaming clients for endpoint with id 'my-streaming-amf' is 1.
Apr 29, 2009 9:39:20 AM org.apache.catalina.core.ApplicationContext log
INFO: [MUSIC]04/29/2009 09:39:20.156 [INFO] [Endpoint.General] Channel endpoint my-streaming-amf received request.
Apr 29, 2009 9:39:20 AM org.apache.catalina.core.ApplicationContext log
INFO: [MUSIC]04/29/2009 09:39:20.156 [DEBUG] [Endpoint.StreamingAMF] Deserializing AMF/HTTP request
IE Failure:
Apr 29, 2009 9:41:36 AM org.apache.catalina.core.ApplicationContext log
INFO: [MUSIC]04/29/2009 09:41:36.812 [INFO] [Endpoint.FlexSession] Number of streaming clients for FlexSession with id '50433B0AE26A35EF30CCE5B58B39E7A9' is 1.
Apr 29, 2009 9:41:36 AM org.apache.catalina.core.ApplicationContext log
INFO: [MUSIC]04/29/2009 09:41:36.812 [DEBUG] [Endpoint.StreamingAMF] Number of streaming clients for endpoint with id 'my-streaming-amf' is 1.
Apr 29, 2009 9:43:16 AM org.apache.catalina.core.ApplicationContext log
INFO: [MUSIC]04/29/2009 09:43:16.828 [INFO] [Endpoint.General] Channel endpoint my-streaming-amf received request.
Apr 29, 2009 9:43:16 AM org.apache.catalina.core.ApplicationContext log
INFO: [MUSIC]04/29/2009 09:43:16.828 [DEBUG] [Endpoint.StreamingAMF] Deserializing AMF/HTTP request
I had assumed, from most of the closely-related postings, that the problem related to 'too many connections', but the INFO message seems to indicate that in both cases, the User Agent is only presenting a single connection request to the server.  And, from what I had read, I thought I would find some log messages indicating that the streaming server was reverting to polling.
Instead, all I can discern is that somewhere there was a 100 second timeout, after which the client and server had arrived at the same state that was true with Firefox after just a couple of milliseconds.  During the 100 second wait, I should add, I did not 'touch' the browser window, did not 'touch' any controls in the applcation, did not even 'touch' anything else except Windows Task Manager hoping to see something happening.

Answer 1

Thanks for a prompt, well-organized, well-thought-out response.  Yes, it is
a bit off topic and I would be happy to just take it off line, but I have
found so many useful pieces of information by being able to read through
closely-related materials, that I will go ahead and post a brief response
for whatever benefit it may provide the community as a whole.  [If it turns
out my analysis and subsequent design are faulty, describing those problems
will benefit me and perhaps others.]
01.  I originally implemented the RemoteObjects using the configuration
definitions referenced as "my-amf".  I think I had understood the tradeoffs
and the application performed acceptably in that manner.  I implicitly knew
I wanted to avoid whatever overhead is involved in 'polling', and didn't
think I needed to flirt with 'streaming'.  But when I began to understand
the benefits of a 'persistent connection', I rather began to like what the
tradeoffs appeared to be.
02.  To follow that line of reasoning you would have to know some other
characteristics of the application, the user community and the funding.  In
reality, almost every user session is going to consume a connection to the
Flash Media Server.  In the world of 'friends-and-family' with no money,
that mostly defines the limits of what I can deliver.  One day I expect to
review an FMS log telling me that someone could not get in because that
would have been the 11th concurrent session!
03.  So, with the heart of the social networking application constrained to
a persistent streamed connection to FMS, I decided to do the same thing for
the rest of the network traffic.  The next limitation, or probably the first
limitation, is trying to deliver all of this 'out of a bedroom' anyway.  I
am too far from my central office to get them to re-provision my ADSL for
greater uplink bandwidth.  How many JukeBoxes can successfully play .mp3s
with 384KB as the setting for the uplink in any case?  So, far, it works
surprisingly well for at least a handful.
04.  So, back to the channel you are the master of.  Actually, after a while
folks tire of just listening to good music in this YouTube world in which we
live, so this application allows drag-n-drop creation of multi-media
Performances.  Now not only do those audio bits have to get down the wire in
good time, a typical 3-minute Elvis song might be accompanied by a dozen or
more JPEGs that are supposed to show up 'on the downbeat' -- so add another
couple of megabytes to the four or five megabytes of the audio coming from
the streaming server -- it would never work.
05.  What does work, is your superb persistent, streaming connection!  In
addition to the sort of business-transactions you have properly guessed as
being involved with those RemoteObjects, there is a Chat Room.  In the Chat
Room, a Producer can 'throw' any JPEGs he would like at the rest of the
members.  The Consumers grab the JPEG images, look at them, and store them
in AIR's application directory.  Now, the next time Elvis is played, and the
CuePoints come popping out indicating that it is time to transition from an
image of a 1958 Cadillac with fins to Fat Elvis in Vegas, at least those
megabytes are not coming, realtime over either FMS's streaming connection or
Blaze's streaming connection.  Indeed a 'Broadcaster" in the Chat Room will
often be speaking into his mic and sending that traffic as 'live' FMS
while typing a reply that is sent over another 'topic'.  So, the bottom line
is that, from my point of view, the "my-streaming-amf" connection is a
session-persistent communication channel that needs to multiplex as much
traffic as possible with the leanest of resources available.  It just seemed
like adding the extra work of opening and closing sockets would get in the
way somewhere in all that 'black-box' code upon which I am relying.