[nos-bbs] Interested in REPEAT packet statistics?

Kenneth C Coe Jr kd8ipn at coehome.com
Sun Apr 11 14:42:35 EDT 2010


George,

There is no such thing as "un-necessary packet duplication." Duplicate
TCP data is ALWAYS sent for a specific reason. there are NO exceptions
to this rule.

Data is not just thrown into a hole called TCP and scooped up by a
shovel. There is a process for negotiating a connection, and for
organizing the data being sent. There is also a process for
acknowledging receipt of the TCP data that was sent to you.

This information has no value whatsoever. In order to know why data is
being re-transmitted, I need to know what happened *AFTER* it was
transmitted each time. You are not randomly storing and re-transmitting
packets of data. Something told the stack to do it. Was it Collision,
Timeout, Re-trans Req, or the Evil Bit? was it the same thing each time,
or different?

If I gave you a recording of my end of a phone call and told you it
might be a phone problem, could you tell me what the phone problem was?
Lets give that a try. Please diagnose the problem.

"hello?"
"hello?"
"Yes"
"no"
"no"
"I'm Sorry?"
"no"
"no"
"goodbye."

Why was I repeating myself? Was there a connection problem, Bad Audio, a
line problem, or a bad phone? Did I have an stroke? Let's try this
another way:

"hello?"
<static>
"hello?"
<is this the coe residence?>
"Yes"
<Do you have new whizz bang product?>
"no"
<Have you ever tried new whizz banf product?>
"no"
<Would you like us to (mumble, mumble)?>
"I'm Sorry?"
<Would you like us to ship you one COD?>
"no"
<Can we call you again in a week?> 
"no"
<Are you sure that you wouldn't like to buy two whizz bangs?>
"goodbye."

Oh, now we see! There is no phone probelm at all, and I'm just fine! the
problem is an $%^#$$% on the other end.

Please throw away this script and submit a TCP dump of the same traffic.
I can see a traffic being transmitted all day, but it means nothing if I
don't see the whole conversation. Give us real data, and we may be able
to give you real results...

Ken

On Sun, 2010-04-11 at 13:23 -0400, George [ham] VerDuin wrote:
> Greeting all.
> This is an FYI sorta note -- trivia about jnos & AMPRnet.
> 
> On my jnos node I see some un-necessary packet duplication.
> There are a few reasons -- some I can help, others require re-programming.
> The subject of this note is only "how many" duplicates show up at my 
> station.
> 
> What follows is a summary of TCP packets received from others at my node.
> It is broken into "datagram" and "vc" because I support both modes.
> What I see is that VC causes a lot more duplication of TCP traffic,
>    -- and that does include AX.25 Level II duplication.
> 
> 
>  >>>> The trace file is named: v1
> The period ends:
>    Sun Apr 11 10:44:35 2010 - vhf recv:
> The period begins:
>    Mon Mar 22 12:27:09 2010 - vhf recv:
> That means the data period is a couple hours short of 20 days.
> 
> The total UI packets = 2083
> The DUPLICATE UI packet counts by repeat number.
>       81       2 TCP: 1029->25 Seq xe927e1da Ack xec3e408c ACK PSH FIN 
> Wnd 2048
>        8       3 TCP: 1051->25 Seq xbe0b00b Ack xdd7d030 ACK PSH Wnd 2048
> TOTAL: 97 of 2083 ==> 4.5% of traffic is duplicate
> 
> The total I packets = 511
> The DUPLICATE I packet counts by repeat number.
>       38       2 TCP: 1036->25 Seq xc3aa000 SYN Wnd 2048 MSS 512
>       14       3 TCP: 23->1034 Seq x971d8000 Ack xc7211001 ACK PSH SYN 
> Wnd 5840
>       12       4 TCP: 1046->25 Seq x57050000 SYN Wnd 2048 MSS 512
>        5       5 TCP: 1041->25 Seq x31a05000 SYN Wnd 2048 MSS 512
>        3       6 TCP: 25->1027 Seq x781b9092 Ack xc16484a0 ACK Wnd 5840
>        2       8 TCP: 25->1027 Seq x781b9092 Ack xc1647a80 ACK Wnd 5840
> TOTAL: 151 of 511 ==> 29.5% of traffic is duplicate
> 
> 
> How to read the above?
> The line just above "TOTAL: 151..." says:
>     "on 2 seperate occurrences in the past 20 days, one packet was 
> repeated 8 times."
>     "A sample TCP header record is included"
> 
> How to repeat the above?
> A Linux script is attached for your amusement.
> If you have jnos trace turned ON for a RF interface, you can compute 
> your own statistics.
> 
> Now What?
> HHHMMMMM...
> 
> 73
> Skip
> _______________________________________________
> nos-bbs mailing list
> nos-bbs at tapr.org
> https://www.tapr.org/cgi-bin/mailman/listinfo/nos-bbs






More information about the nos-bbs mailing list