Project Home
Project Home		 Wiki
Wiki		 Discussion Forums
Discussions		 Project Information
Project Info
Networking/Discussion/General/Lost packet?/List of Posts
Forum Topic - Lost packet?: (9 Items)
   
Update
Mario Charest
04/10/2008 3:27 PM
post6646
Lost packet?   
Using 6.3.2.  I'm also using the dreaded PCI-E Intel Pro card that is giving us lots of grief ( I'm told the driver is 
being worked on ).

Our system is talking to a PLC and once in a while communication times out.  I managed to grap this occurance in 
WireShark ( Ethereal successor).  I'm don't know enough about TCP/IP, but it does look to me as if a packet got lost.  

10.122.206.3 is the QNX computer and 10.122.206.99 is the PLC.  In a nut shell the program does one send() and then 2 
recv() and that consists of one command.

It's happening on line 5980. I'm guessing a packet got lost.

File is attached, and compresses with 7zip.
Attachment: Text ok.7z 106.55 KB
Sean Boudreau(deleted)
04/10/2008 4:08 PM
post6649
Re: Lost packet?   
On Thu, Apr 10, 2008 at 03:27:08PM -0400, Mario Charest wrote:
> 
> Using 6.3.2.  I'm also using the dreaded PCI-E Intel Pro card that is giving us lots of grief ( I'm told the driver is
 being worked on ).
> 
> Our system is talking to a PLC and once in a while communication times out.  I managed to grap this occurance in 
WireShark ( Ethereal successor).  I'm don't know enough about TCP/IP, but it does look to me as if a packet got lost.  
> 
> 10.122.206.3 is the QNX computer and 10.122.206.99 is the PLC.  In a nut shell the program does one send() and then 2 
recv() and that consists of one command.
> 
> It's happening on line 5980. I'm guessing a packet got lost.
> 
> File is attached, and compresses with 7zip.

We never ack the packet the PLC sends at line 5979 so it
looks like we didn't get it or didn't like something about
it (although the trace doesn't seem to say anything is wrong
with it...).  However the PLC should resend it either
through a timeout or by being poked with our duplicate acks.

-seanb
Mario Charest
04/10/2008 5:56 PM
post6652
Re: Lost packet?   
> 
> We never ack the packet the PLC sends at line 5979 so it
> looks like we didn't get it or didn't like something about
> it (although the trace doesn't seem to say anything is wrong
> with it...).  However the PLC should resend it either
> through a timeout or by being poked with our duplicate acks.
> 
> -seanb

Wow, thanks Sean I appreciate this, I wasn't expecting an answer that fast ;-)

So it looks like the ACK never got transmitted or the packet 5979 was never received.  Gut feeling tells me the packet 
5979 was lost in the driver, which is why the ACK was not sent in the first place.

I'll see if I can move the PLC to a separate network card and one different then Intel.  

I'm not surprise the PLC doesn't handle this properly ;-(
Robert Craig
04/10/2008 6:03 PM
post6653
RE: Lost packet?   
Do you happen to know if this was after a high rate burst of packets?  One
possibility is that either the tx or rx side ran out of buffers temporarily
with the result being a dropped packet.  You can bump up the number of rx /
tx descriptors to get around that problem.

	Robert.

-----Original Message-----
From: Mario Charest [mailto:mcharest@zinformatic.com] 
Sent: Thursday, April 10, 2008 5:56 PM
To: general-networking
Subject: Re: Lost packet?

> 
> We never ack the packet the PLC sends at line 5979 so it
> looks like we didn't get it or didn't like something about
> it (although the trace doesn't seem to say anything is wrong
> with it...).  However the PLC should resend it either
> through a timeout or by being poked with our duplicate acks.
> 
> -seanb

Wow, thanks Sean I appreciate this, I wasn't expecting an answer that fast
;-)

So it looks like the ACK never got transmitted or the packet 5979 was never
received.  Gut feeling tells me the packet 5979 was lost in the driver,
which is why the ACK was not sent in the first place.

I'll see if I can move the PLC to a separate network card and one different
then Intel.  

I'm not surprise the PLC doesn't handle this properly ;-(






_______________________________________________
General
http://community.qnx.com/sf/go/post6652
Mario Charest
04/10/2008 6:43 PM
post6654
Re: RE: Lost packet?   
The transmist and receive descriptor are set at 1024.  That should be enough shouldn't it?  I'll dig around maybe the 
driver thread in  io-net gets bumped by a process of higher priority.

I just saws the thing is in half duplex, what???

> Do you happen to know if this was after a high rate burst of packets?  One
> possibility is that either the tx or rx side ran out of buffers temporarily
> with the result being a dropped packet.  You can bump up the number of rx /
> tx descriptors to get around that problem.
> 
> 	Robert.
> 
> -----Original Message-----
> From: Mario Charest [mailto:mcharest@zinformatic.com] 
> Sent: Thursday, April 10, 2008 5:56 PM
> To: general-networking
> Subject: Re: Lost packet?
> 
> > 
> > We never ack the packet the PLC sends at line 5979 so it
> > looks like we didn't get it or didn't like something about
> > it (although the trace doesn't seem to say anything is wrong
> > with it...).  However the PLC should resend it either
> > through a timeout or by being poked with our duplicate acks.
> > 
> > -seanb
> 
> Wow, thanks Sean I appreciate this, I wasn't expecting an answer that fast
> ;-)
> 
> So it looks like the ACK never got transmitted or the packet 5979 was never
> received.  Gut feeling tells me the packet 5979 was lost in the driver,
> which is why the ACK was not sent in the first place.
> 
> I'll see if I can move the PLC to a separate network card and one different
> then Intel.  
> 
> I'm not surprise the PLC doesn't handle this properly ;-(
> 
> 
> 
> 
> 
> 
> _______________________________________________
> General
> http://community.qnx.com/sf/go/post6652
Mario Charest
04/11/2008 9:47 AM
post6670
Re: RE: Lost packet?   
A bit thanks to everyone that commented on this post. I have been at this ( and several other problems ) for 3 weeks, 
all by myself 5000km from home.  Your comment was the push I needed to get my exhausted brain into the right direction ;
-)

Basically I ended up bumping rx to 2048 and tx to 1024 and also raised the priority of the driver thread.

I still have to figure out why it ends up in half-duplex though.  A reboot doesn't fix it but if I unplug and plug the 
cable, it then goes to full-duplex.  If I have to, I'll force full-duplex on both end ( driver and switch ).

- Mario "Grateful" Charest
Robert Craig
04/11/2008 11:04 AM
post6679
RE: RE: Lost packet?   
Glad to hear that things worked out.  Are you using the experimental version
of the i82544 driver that Hugh provided?  I'm wondering if the half-duplex
negotiation thing is a problem that we may have already fixed.

Hopefully that's "5000km" in the direction of sun and warmth.  It's hovering
around 0C with snow / freezing rain / rain in the mix today in Ottawa.  The
joys of spring...

    Robert.

-----Original Message-----
From: Mario Charest [mailto:mcharest@zinformatic.com] 
Sent: Friday, April 11, 2008 9:48 AM
To: general-networking
Subject: Re: RE: Lost packet?


A bit thanks to everyone that commented on this post. I have been at this (
and several other problems ) for 3 weeks, all by myself 5000km from home.
Your comment was the push I needed to get my exhausted brain into the right
direction ;-)

Basically I ended up bumping rx to 2048 and tx to 1024 and also raised the
priority of the driver thread.

I still have to figure out why it ends up in half-duplex though.  A reboot
doesn't fix it but if I unplug and plug the cable, it then goes to
full-duplex.  If I have to, I'll force full-duplex on both end ( driver and
switch ).

- Mario "Grateful" Charest

_______________________________________________
General
http://community.qnx.com/sf/go/post6670
Mario Charest
04/11/2008 2:34 PM
post6685
Re: RE: RE: Lost packet?   
> Glad to hear that things worked out.  Are you using the experimental version
> of the i82544 driver that Hugh provided? 

Yes, but not the latest one.  It's the one given to me a few weeks back.  I'll try it tomorrow.  What I find strange is 
that another computer beside it with the same configuration and connected to the same switch, detects full-duplex just 
fine.  Hopefully the network cable just got him by one of them cosmic particle ( there is a lot of those in the part of 
the world I'm in....)

> I'm wondering if the half-duplex negotiation thing is a problem that we may have already fixed.
> 
> Hopefully that's "5000km" in the direction of sun and warmth.  It's hovering
> around 0C with snow / freezing rain / rain in the mix today in Ottawa.  The
> joys of spring...

Nah north west.
Sean Boudreau(deleted)
04/11/2008 10:22 AM
post6674
Re: Lost packet?   
On Thu, Apr 10, 2008 at 05:56:12PM -0400, Mario Charest wrote:
> > 
> > We never ack the packet the PLC sends at line 5979 so it
> > looks like we didn't get it or didn't like something about
> > it (although the trace doesn't seem to say anything is wrong
> > with it...).  However the PLC should resend it either
> > through a timeout or by being poked with our duplicate acks.
> > 
> > -seanb
> 
> Wow, thanks Sean I appreciate this, I wasn't expecting an answer that fast ;-)
> 
> So it looks like the ACK never got transmitted or the packet 5979 was never received.  Gut feeling tells me the packet
 5979 was lost in the driver, which is why the ACK was not sent in the first place.
> 
> I'll see if I can move the PLC to a separate network card and one different then Intel.  
> 
> I'm not surprise the PLC doesn't handle this properly ;-(

We ack after packet 5979 but we ack the last bytes we received
which is previous to packet 5979. ie. we didn't get or didn't like
packet 5979.  When the plc sees we're acking earlier bytes it
should re-send starting at where we're acking.

Regards,

-seanb

Return


  A subsidiary of BlackBerry             All content ©2004-2013, QNX Software Systems