[Netconf] server-model #43: keep-alive, linger, reconnect interval defaults OK?

[t.petch <[e-mail]>]

 ----- Original Message -----
 From: "Kent Watsen" <[e-mail]>
 To: "Phil Shafer" <[e-mail]>
 Cc: <[e-mail]>
 Sent: Friday, May 15, 2015 12:06 AM
 >
 > On 3/30/15, 1:03 AM, "Phil Shafer" <[e-mail]> wrote:
 > >Kent Watsen writes:
 > >>Note, this is a
 > >>keep-alive message being sent every 15 to 300 seconds, with a
 default
 > >>count-max of '3', results in a new connection after 45 to 900
 seconds -
 > >>that's 3/4 of a minute to 15 minutes to proactively bring up a
 failed
 > >>session.
 > >
 > >Having a connection to your NMS die after 45 seconds during a
network
 > >issue, when the network _is_ having trouble, seems unwise.
 >
 > Hi Phil,
 >
 > This timer is to test the aliveness of the connection.  Sure, if the
 NMS
 > doesn't respond in time, the device tears down that connection in
hope
 to
 > find a connection the works better.
 >
 > 45 seconds seems too short to you, and 15 minutes seems too long to
 me.
 > How about 5 minutes for the default?  Anyone else have an opinion?

 Yes; it depends on the application.

 In voice (not where this will be used), the target MTTR is 50mS so
 failure detection is set at 30mS so the keepalives are sent every 10mS
 and this is seen as the norm.

 In network operation, I see operation centres wanting to know about
 failures before the phone starts ringing, in which case I see 30S used,
 which is on the long side, but a shorter interval may generate too much
 traffic when the network has over 100,000 elements.

 For a non-operational system, minutes or hours might be appropriate,
 even once a day for a backup system.

 I would rather we did not put in a default but something along these
 lines, that it depends how quickly a failure needs to be detected to
 meet the required service.  The count of three attempts before
declaring
 failure is, by contrast, almost universal so I have no qualms there.
 This implies failures are detected in three to four times the interval.

 If forced to put in a default time, I would go for 30S with a caveat
 that this should be reviewed in the light of how much traffic this will
 generate - it is very dependent of network size.

 Tom Petch

> >
> > Thanks,
> > Kent