[nos-bbs] Interested in REPEAT packet statistics?

[George [ham] VerDuin]

HI Barry, you're not Kenneth who tells me nothing is wrong...:-)

On 04/13/2010 03:09 PM, Barry wrote:
>
>> Now let's say after the third then you receive 3 duplicate
>> ACKS.
>>      
> I love math problems!
>    
There is some math involved, it centers on the values of SRTT and MDEV.
These buggers are built into jnos -- you can see them via the status 
commands.

>> What does the fact you receive dup ACK tell you about your
>> own node?
>>      
> That your own node has a T1/Retry timer set to expire at least
> three times faster than it should be?
>    
Not necessarily three times, but significantly non the less.
Now if the packest are the SYN packets you are spot on.
A large part of the issue is that you as sysop only have IRTT to adjust.

Backoff strategy, begins to play,
maxtime can be really bollixed up,
blimit might be set too low,
retrys might be a big number,
but mostly these things play big time only after 3 or 4 repeats.

But quite frankly damn the bullets.
The detail is not what my example is about.
Facts seem to bear out that re-transmission is
      out of sight -- out of mind for many of us.
Repeats are one of the most annoying aspects of opening a connection.
And because slow initial response annoys us, we tend to shorten up on 
the variables listed above in the belief we are better off when the 
opposite is the truth.

Back to math.
Wait time in jnos2.0h is based on the formula:
     Wait = srtt + 4 * mdev
There are conditions when this formula servers us less well.
If you are a student of classical statistics, you might enjoy looking.
Also there is a timer-race between AX and TCP or AX and NET/ROM that 
adds to un-necessary repeats.

You have however captured the essence of my question:
Seeing 3 ACKS means you sent 2 too many packets.
For whatever reason.
Perhaps under your control.
Perhaps not.
Two packest of three are not required, and no rebuttal is available.
It's also likely the two extras slowed the first response making it worse.

What to do about repeats is a long story.
When to do it is my focus at this moment.
Without looking, then "never" is the answer.
The script I attached is one way of looking out of many ways.

Thanks Barry
Skip