[Dxspider-support] newbie sysop

[charlie carroll]

Jack:
How long before the users were disconnected?  I've been connected about
an hour without a problem.  Well, I stayed connected until you
rebooted.  You did reboot, didn't you?

Having said that, a few other words... DIRK, pse read.

I have long suspected that there is a problem with Windows based
installations. However, I do not believe it's a DXSpider problem. 
Rather, my theory is that the problem resides with the Windows tcp/ip
stack or its settings.
My K1XX node in New Hampshire has 30-40 node connections at any one
time.  Over time I have noticed two things.  First, Windows-based nodes
tend to disconnect significantly more often than Linux-based nodes. 
Second, user telnet connections tend to stay connected longer than
either type of node.
One day I spent some time with an ethernet monitor watching packets
while trying to understand the disconnects.  I found that the
disconnects from Windows nodes were being caused at the tcp/ip level.  
It has been several months since I made these observations, so my
terminology may not be totally correct.  Anyway, at the packet level, I
was seeing a tcp/ip disconnect flag being issued.  This flag was
apparently the result of packets being sent, but not being acknowledged
in a timely fashion.  In this case, the timing was at the seconds level
as opposed to the disconnects caused by DXSpider not receiving pings
which occur at the minutes level.  What I also noticed was what appeared
to be asymmetric transmission times depending upon which direction the
packet was going.  
Because there is a high degree of handshaking between nodes, short
duration outages or slowdowns in the internet tend to cause the
disconnects at the tcp/ip level.  I suspect that the default parameters
relating to packet retries (or even time between retries) in the Linux
and Windows tcp/ip stacks are different.  I was able to find the means
to change the Windows registry for tcp/ip retries.  However, I was not
able to find the default values for either the Windows or Linux stack.
Why do user telnet connects stay connected longer?  There is minimal
handshaking between user connections and node-to-node connections. 
Hence, short duration outages are not noticed.

That's it for now.  I'd appreciate any other comments on these thoughts.

73 charlie, k1xx/4