[Spread-users] packet loss? (win2000)

Fernando Marotta F.Marotta at itsoftware.it
Mon Jul 22 09:09:10 EDT 2002


I've started spread, the receiver (spflooder ... -ro) and then the sender
(...-wo); what I saw is that the messages are sent very fastly, but the
first X of them are received, and other are lost; it seems that the sender
sends messages at a fast rate, and some of them are dropped. 
In fact, if I reduce the send rate (eg: putting a Sleep(1) after each send);
everything works fine (apart from the performance, of course).

I thought it could be possible that spread uses a (too little) buffer for
the incoming messages and after, like, 700 msgs of 100 bytes, the buffer
gets full and all subsequent messages sent are accepted, but are dropped.

I couldn't figure out how to control spread buffer (or other params); and/or
how to know if some packets are lost (or dropped).

I don't know how to give you other ideas about what's wrong. 

I would like to know if other spread users are using Win2000 Pro (or other
Win32 arch.) and had some kind of problem/high packet loss rate.

I hope somebody can help me in understanding this issue.



> -----Original Message-----
> From: Yair Amir [mailto:yairamir at cnds.jhu.edu]
> Sent: lunedì 22 luglio 2002 15.02
> To: spread-users-admin at lists.spread.org; Fernando Marotta
> Cc: spread-users at lists.spread.org
> Subject: Re[4]: [Spread-users] packet loss? (win2000)
> Hi,
> I just want to double check that when you run your sender, your
> receiver is already running. Otherwise, of course the receiver will
> not receive the messages that were sent before the receiver joined the
> group.
> Also, I think you really should go and read the user guide in terms of
> setting broadcast or IP multicast etc. And you should fix your config
> file so that you understand every line in it and every line works in
> it. Although I don't think that your config file is the cause of the
> problem.
> If you did start your receiver well before you started your sender -
> just give the receiver a little more time to get the messages. If both
> of these don't help - then I have no idea. I never heard of anyone
> experiencing a percentage loss of messages. In my opinion, it is just
> not possible, so it has to be something else.
>     Cheers,
>     :) Yair.
> Fernando> Hi Yair,
> Fernando> I tried the out-of-the-box spflooder and had the 
> same results:
> Fernando> I sent messages using the following command:
> Fernando>                 spflooder -u sender -m 1000 -b 100 -wo
> Fernando> and received that using:
> Fernando>                 spflooder -u recv -m 1000 -b 100 -ro
> Fernando> spread is in execution with this configuration file:
> Fernando> 
> -------------------------------------------------------------
> Fernando> Spread_Segment {
> Fernando>         localhost     
> Fernando> }
> Fernando> Spread_Segment {
> Fernando>         itwks2053     
> Fernando> }
> Fernando> # Spread options
> Fernando> DebugFlags = { PRINT EXIT }
> Fernando> DangerousMonitor = false
> Fernando> 
> ----------------------------------------------------------------
> Fernando> (I had to insert my machine twice: having only the 
> loopback interface
> Fernando> 'localhost' won't work...)
> Fernando> Then, to complete the "message reception" I had to 
> run the send 3 times (so
> Fernando> more or less the same packet loss). 
> Fernando> I also saw that if I change the spread process 
> priority to "above normal" in
> Fernando> Win2000 (while all the spflooder have "normal" 
> priority) all the messages
> Fernando> are received, but the packet rate is very little 
> (<100 msg/sec).
> Fernando> Maybe I'm making something wrong, and/or you can 
> give me some idea. (I
> Fernando> really hope so).
> Fernando> BTW, I also had another question: I read that 
> spread may uses broadcast as
> Fernando> performance optimization; there is a way of 
> disabling this behavior? (I
> Fernando> would like to try spread in an environment where 
> other sw are running using
> Fernando> broadcast, and would prefer avoiding the broadcast 
> storm effect).
> Fernando> Thanks.
> Fernando> Cheers
> Fernando> Fernando

More information about the Spread-users mailing list