Re: CAP TODO: disconnect

[George Babics <georgeb@xxxxxxxxxxx>]

  By doing CAP's transport layer with BEEP, disconnect
will be out of the scope of CAP.

George

Doug Royer wrote:
> 
> > - DISCONNECT: should the client now wait for a response
> >  from the server before disconnecting?
> 
> YES!!
> 
> For two reasons:
> 
>         1) To make sure the CS got the request.
> 
>         2) To keep the CS from hanging open that connection for
>            a while in TCP-WAIT state when:
> 
>                 CUA:    sends - goodbye
>                 CS:     sends - goodbye reply
>                 CUA:    physical hang-up before it gets the CS reply.
>                 CS:     TCP waiting for the ill-mannered CUA's
>                         TCP stack to say 'got message'
> 
> This was discussed in a SMTP IETF meeting. It can cause
> the server side to hang for a while in TCP-WAIT state while
> the TCP stack determines if the connection died or is just slow.
> 
> SPANNERS loved the fact you could fire off junk to an smtp-server
> and disconnect without waiting for the spam target system to reply,
> sometimes resulting in possibly a HUGE NUMBER of sessions hung for
> a couple of minutes while the TCP-stack determining if they hung up
> or are slow. I think that on most Unix systems the default is a 2 minute
> wait per connection.
> 
> -Doug