Hi,



We have a FCS 1.5 application (not FMS). It has 2-way audio and a
whiteboard. The server is placed in Dallas, TX. One user is in CA and
another is in India. Here are some observations:

1. When the user in India is connected directly to a DSL modem, the latency
is about 400ms. Although would love to reduce it, but it is acceptable.

2. When the user is on an internal LAN connected through hubs and switches,
the delay is unpredictable. It varies from a few hundred milliseconds to 15
seconds (very troubling). Another observation is that Skype works just fine
on (1-2 second delay) the same network. Is there a special network
configuration that should be used to avoid such delays?

3. Here is another scenario. I installed the application on one computer (HP
notebook; 2.4GHz; 256MB RAM). We had one user on one computer and another
user on another. All connected through a wireless 54Mbps router (Linksys).
All three computers were in the 20 feet radius. The application worked but
as for the delay, it varied from 2ms to 7 secs. Isn't it weird to experience
such delays since everything is working locally?



Skype doesn't seem to have any such problem. Is it because it has a
peer-to-peer architecture?



Also, any pointers to resources to improve this will be helpful.



Thanks,

Harman


=-----------------------------------------------------------
Supported by Fig Leaf Software - http://www.figleaf.com
=-----------------------------------------------------------

To change your subscription options or search the archive:
http://chattyfig.figleaf.com/mailman/listinfo/flashcomm

[ Post a follow-up to this message ]

Harman Singh wrote:
> 1. When the user in India is connected directly to a DSL modem, the latenc
y
> is about 400ms. Although would love to reduce it, but it is acceptable.
>
> Skype doesn't seem to have any such problem. Is it because it has a
> peer-to-peer architecture?
>
Good... at least we are not the only ones finding this weird behaviour.
We did a lot of tests and were never been able to find the source of the
problem.

It happens that users from east (India, China) have an HUGE delay.
It would look like it's due to psysical connection delay, but it's not!
Since skype (sometomes  msn too) don't have this problem at all!
The only thing we could guess is that skype uses https over port 443 so
it's not going to be inspected by government firewalls.
We could not try flash traffic over 443 because we can't afford to buy
FMS due to stupid licensing, so we can't know if this is the solution to
this problem.

Current situation is that: users from particular regions of the eastern
part of the world do have high pings and very variable data rate
(exactly what happens when you have  some kind of sw which is inspecting
your traffic).


Let me know if you can find something more...

Dario De Agostini

=-----------------------------------------------------------
Supported by Fig Leaf Software - http://www.figleaf.com
=-----------------------------------------------------------

To change your subscription options or search the archive:
http://chattyfig.figleaf.com/mailman/listinfo/flashcomm

[ Post a follow-up to this message ]

Hi Harman,
I didn't see a response to your note so I thought I'd try to come up
with something useful. However, there are so many unknowns that you
should not assume that anything I write is the reason for what you are
experiencing.

Skype should have at least two significant advantages over FCS 1.5.
First, I gather Skype does more than make a point-to-pint connection to
deliver audio from one point to another. It keeps multiple connection
paths open between points (or nodes) and dynamically chooses a path that
minimizes latency. So it behaves -  if I understand this correctly -
more like a virtual network layer on top of the Internet. Second, Skype
uses UDP whenever it can. When UDP packets traverse a a congested
network some packets will be dropped. For audio this may not matter.
Skype may be able to construct good enough audio quality without having
to have the missing packets resent. FCS and FMS use TCP. TCP packets
will also be dropped by a congested network but they will always be
automatically resent which can dramatically increase latency. Skype may
have other advantages I'm not aware of and there are likley some limits
to how well it can exploit its virtual network and UDP.

RTMP in FCS 1.5 does monitor network congestion and will drop audio
messages to compensate. But that is not as effective as streaming over
UDP. I thought I read somewhere that FMS 2 improved on FCS 1.5 but I
can't find the reference.  The marketing materials just says it tracks
"the playback experience on the client and correct unforeseen playback
problems due to network congestion in real-time."

So if you were just talking about the situation in points 1 and 2, I
would write off the problem to network congestion. But the problem with
point three might or might not be a network problem. You really should
test in a wired LAN and not with wireless. If your FCS server is being
overwhelmed with work - or even if you are overwhelming one application
instance - you can experience the kind of thing you are describing. For
example if you queue up a massive amount of ActionScript data the
performance on that instance can really slow down until the server has
dealt with it. So it may have nothing to do with network issues or it
may be a combination of a problem application and network problems. If
the server is dedicated to FCS and has lots of memory and CPU to spare,
excellent connectivity, and your app is only sending and receiving two
audio streams, then you might ask if someone can look into network
congestion at the remote site.

Yours truly,
-Brian


Harman Singh wrote:

>Hi,
>
>
>
>We have a FCS 1.5 application (not FMS). It has 2-way audio and a
>whiteboard. The server is placed in Dallas, TX. One user is in CA and
>another is in India. Here are some observations:
>
>1. When the user in India is connected directly to a DSL modem, the latency
>is about 400ms. Although would love to reduce it, but it is acceptable.
>
>2. When the user is on an internal LAN connected through hubs and switches,
>the delay is unpredictable. It varies from a few hundred milliseconds to 15
>seconds (very troubling). Another observation is that Skype works just fine
>on (1-2 second delay) the same network. Is there a special network
>configuration that should be used to avoid such delays?
>
>3. Here is another scenario. I installed the application on one computer (H
P
>notebook; 2.4GHz; 256MB RAM). We had one user on one computer and another
>user on another. All connected through a wireless 54Mbps router (Linksys).
>All three computers were in the 20 feet radius. The application worked but
>as for the delay, it varied from 2ms to 7 secs. Isn't it weird to experienc
e
>such delays since everything is working locally?
>
>
>
>Skype doesn't seem to have any such problem. Is it because it has a
>peer-to-peer architecture?
>
>
>
>Also, any pointers to resources to improve this will be helpful.
>
>
>
>Thanks,
>
>Harman
>
>
>=-----------------------------------------------------------
>Supported by Fig Leaf Software - http://www.figleaf.com
>=-----------------------------------------------------------
>
>To change your subscription options or search the archive:
>http://chattyfig.figleaf.com/mailman/listinfo/flashcomm
>
>


--
 ________________________________________
______________________________
Brian Lesser
Assistant Director, Teaching and Technology Support
Computing and Communications Services
Ryerson University
350 Victoria St.
Toronto, Ontario                   Phone: (416) 979-5000 ext. 6835
M5B 2K3                            Fax: (416) 979-5220
Office: AB48D                      E-mail: blesser-6s6ziW1YCwCw5LPnMra/2Q@pu
blic.gmane.org
(Enter through LB66)               Web: http://www.ryerson.ca/~blesser
 ________________________________________
______________________________


=-----------------------------------------------------------
Supported by Fig Leaf Software - http://www.figleaf.com
=-----------------------------------------------------------

To change your subscription options or search the archive:
http://chattyfig.figleaf.com/mailman/listinfo/flashcomm

[ Post a follow-up to this message ]

I forgot to mention that turning up the silence level can help with
audio latency. For example:
active_mic = Microphone.get();
active_mic.setSilenceLevel(25);
-B

Brian Lesser wrote:

> Hi Harman,
> I didn't see a response to your note so I thought I'd try to come up
> with something useful. However, there are so many unknowns that you
> should not assume that anything I write is the reason for what you are
> experiencing.
>
> Skype should have at least two significant advantages over FCS 1.5.
> First, I gather Skype does more than make a point-to-pint connection
> to deliver audio from one point to another. It keeps multiple
> connection paths open between points (or nodes) and dynamically
> chooses a path that minimizes latency. So it behaves -  if I
> understand this correctly - more like a virtual network layer on top
> of the Internet. Second, Skype uses UDP whenever it can. When UDP
> packets traverse a a congested network some packets will be dropped.
> For audio this may not matter. Skype may be able to construct good
> enough audio quality without having to have the missing packets
> resent. FCS and FMS use TCP. TCP packets will also be dropped by a
> congested network but they will always be automatically resent which
> can dramatically increase latency. Skype may have other advantages I'm
> not aware of and there are likley some limits to how well it can
> exploit its virtual network and UDP.
>
> RTMP in FCS 1.5 does monitor network congestion and will drop audio
> messages to compensate. But that is not as effective as streaming over
> UDP. I thought I read somewhere that FMS 2 improved on FCS 1.5 but I
> can't find the reference.  The marketing materials just says it tracks
> "the playback experience on the client and correct unforeseen playback
> problems due to network congestion in real-time."
> So if you were just talking about the situation in points 1 and 2, I
> would write off the problem to network congestion. But the problem
> with point three might or might not be a network problem. You really
> should test in a wired LAN and not with wireless. If your FCS server
> is being overwhelmed with work - or even if you are overwhelming one
> application instance - you can experience the kind of thing you are
> describing. For example if you queue up a massive amount of
> ActionScript data the performance on that instance can really slow
> down until the server has dealt with it. So it may have nothing to do
> with network issues or it may be a combination of a problem
> application and network problems. If the server is dedicated to FCS
> and has lots of memory and CPU to spare,  excellent connectivity, and
> your app is only sending and receiving two audio streams, then you
> might ask if someone can look into network congestion at the remote site.
>
> Yours truly,
> -Brian
>
>
> Harman Singh wrote:
> 
>
>


--
 ________________________________________
______________________________
Brian Lesser
Assistant Director, Teaching and Technology Support
Computing and Communications Services
Ryerson University
350 Victoria St.
Toronto, Ontario                   Phone: (416) 979-5000 ext. 6835
M5B 2K3                            Fax: (416) 979-5220
Office: AB48D                      E-mail: blesser-6s6ziW1YCwCw5LPnMra/2Q@pu
blic.gmane.org
(Enter through LB66)               Web: http://www.ryerson.ca/~blesser
 ________________________________________
______________________________


=-----------------------------------------------------------
Supported by Fig Leaf Software - http://www.figleaf.com
=-----------------------------------------------------------

To change your subscription options or search the archive:
http://chattyfig.figleaf.com/mailman/listinfo/flashcomm

[ Post a follow-up to this message ]

Brian Lesser wrote:
> I forgot to mention that turning up the silence level can help with
> audio latency. For example:
> active_mic = Microphone.get();
> active_mic.setSilenceLevel(25);

but you have to be warned about the fact you could see weird "lags" when
your audio switches from "not-sent" to "sent" state.
there should be some post about this in this list.

Anyway, the whole tcp vs udp analisys is perfect. But there is a problem
we found out:
if you ping the fcs server from ,say, China you will get , say, 200/300ms
When you connect with fcs you will see a delay which goes up and down
from about 1sec up to 15/20sec

this is strange... the behaviour you see is very similar to the one you
get when there is some kind of antivirus which does traffic inspection...


We have to try fms2 rtmps. Anyone had some fun with it yet ?

Dario De Agostini


=-----------------------------------------------------------
Supported by Fig Leaf Software - http://www.figleaf.com
=-----------------------------------------------------------

To change your subscription options or search the archive:
http://chattyfig.figleaf.com/mailman/listinfo/flashcomm

[ Post a follow-up to this message ]

Thanks for your response guys. We are trying different techniques in FMS and
will keep you posted on the results.

Thanks,
Harman

-----Original Message-----
From: flashcomm-bounces-1Ss2GqJETD3yZ38Mhd3e/9ZfFG6BLHNm@public.gmane.org
[mailto:flashcomm-bounces-1Ss2GqJETD3yZ38Mhd3e/9ZfFG6BLHNm@public.gmane.
org] On Behalf Of Dario De
Agostini
Sent: Tuesday, January 17, 2006 5:45 AM
To: FlashComm Mailing List
Subject: Re: [FlashComm] Audio delay (latency) in FCS 1.5 application

Brian Lesser wrote:
> I forgot to mention that turning up the silence level can help with
> audio latency. For example:
> active_mic = Microphone.get();
> active_mic.setSilenceLevel(25);

but you have to be warned about the fact you could see weird "lags" when
your audio switches from "not-sent" to "sent" state.
there should be some post about this in this list.

Anyway, the whole tcp vs udp analisys is perfect. But there is a problem
we found out:
if you ping the fcs server from ,say, China you will get , say, 200/300ms
When you connect with fcs you will see a delay which goes up and down
from about 1sec up to 15/20sec

this is strange... the behaviour you see is very similar to the one you
get when there is some kind of antivirus which does traffic inspection...


We have to try fms2 rtmps. Anyone had some fun with it yet ?

Dario De Agostini


=-----------------------------------------------------------
Supported by Fig Leaf Software - http://www.figleaf.com
=-----------------------------------------------------------

To change your subscription options or search the archive:
http://chattyfig.figleaf.com/mailman/listinfo/flashcomm



=-----------------------------------------------------------
Supported by Fig Leaf Software - http://www.figleaf.com
=-----------------------------------------------------------

To change your subscription options or search the archive:
http://chattyfig.figleaf.com/mailman/listinfo/flashcomm

[ Post a follow-up to this message ]