Re: Lost /duplicated messages

[Kevin Hawkins <lists@xxxxxxxxxxxxxxxxx>]

barrygordon wrote:

>I have an application where I need to  transfer large blocks of Data,
>lets say 2500-5000 bytes.  Knowing the UDP size limits I broke this
>down into multiple messages each message indicating its sequence number
>for reassembly at the other end so they are order independent.
>
>
Just getting my head around this - multiple xAP message blocks ?  What
are you using to segment and build the packets - your own application
(?) built in xFx ?

>I see the messages going out (all of them) in the correct order (using
>a sniffer to monitor).
>
When you say a sniffer you mean with something like xAP Viewer - ie you
are seeing the packets at the xAP layer ?

> On the receiveing side i almost never get all
>the message blocks.  Most of the time I get n copies of the last block
>and none of the others.
>
Again is this in something like Viewer (on a  different PC or the same
PC) ?

>Only once did I see al the blocks arrive.  I
>have tried delays between blocks on the transmit side, etc.
>
The key indicator that something is wrong here is the repeated packets
as they were just never sent so I think focussing on that rather than
the apparently lost ones might reveal the problem.  Hence delays may not
change things.  The Netiom can send large numbers of xAP UDP packets
(particulalrly if you enable all the I/O and query it using a wildcard
*)  - but these all appear correctly in Viewer for me - the traffic rate
is many thousand packets per minute, well above anything a  typical xAp
network would see.

>I am ready
>to pull  my hair (what little is left) out.  I have played with varying
>block sizes and varying delays between transmissions. The two PC's are
>2.4 GHZ machines and the network is a single subnet at 100 mbps.
>
>Can Anyone help me.
>
>I am running on Win XP pro, the .Net framework is installed and
>the .net hub is running as a service on each system (two for testing).
>i do not understand why I even need the hubs as it is a single xAP
>application on each machine.
>
>
If you only have one xAP 'listener' application on a machine then you
don't need the hub.  Any number of xAP senders can reside on one machine
without a hub but only 1 xAP application can hear on a given machine
without a hub. Later applications will appear deaf.  So for 2+ xAP
receivers a hub is required.  You could remove the hub in your case.

Have you a bit more info on the 'app' and what it;'s written in ?

Just thinking - if this is a lot of frequent data,  is it required that
the data is sent to all nodes 'broadcast/udp'  or would a tcp connection
be preferrable ?  xAP will handle a lot of data but being based on UDP
it is not sequenced, paced or acknowledged. However 5K bytes is almost
nothing data wise, just 4 or 5 packets and it will cope admirably with
this . You just dont want to transfer streams eg audio or something via
xAP.

We have had talks around exactly the task you are tackling, dividing
larger blocks of data across packets and adding sequencing and
acknowledge information, this may well appear as a later official
extension to xAP achieved as a policy layer with special parameters
inserted into the header or blocks.  There are experminental parameters
available for exactly this purpose for use in the header Test00  to
Test99 if you wish to use/abuse the header  -    These should pass
through hubs OK but Stuart will perhaps confirm this.    As ever there
is a golden rule of xAP that unexpected parameters encountered in
headers and blocks should be gracefully ignored which facilitates this
expansion.  However xFX based apps  are particular vigilant at
encouraging strict conformance to the xAP spec so may beat you up a bit
on this.

>Any help would be really appreciated.
>
>
>
This should all work very nicely so there must be a code issue somewhere
I am thinking...

Kevin