Andrey,

OK. I see that my discussion of delayed notificaions didn't quite address your question - you're more interested in the polling algorithm. Bob Merrill's quote from the User Guide gives the details, but maybe not quite in the terms that you're asking...

In fact, I believe that the current InterMapper behavior already does what you describe - it always retries many times at a short interval (shorter than the poll interval) before reporting an outage.

There are two separate cases:

1) In TCP probes (SMTP, HTTP, etc.) InterMapper reports an outage because it has failed to establish a connection within 60 (the default) seconds. InterMapper relies on the underlying OS facility to retransmit the SYN/SYN/ACK packets until it works or it has failed multiple times.

Consequently, if InterMapper reports an outage of a TCP-based probe, it has definitively failed. There have been several attempts to start the connection (using the TCP retry algorithm, with exponential backoff, etc.) and it's fair to say that at least one device (InterMapper) has already failed to connect. There's no need to retry again.

2) For UDP probes (Ping, SNMP, etc.) InterMapper sends N requests (default is three attempts) spaced M seconds apart (default is three - both parameters are settable) before declaring the device to be down.

In this case, too, if InterMapper reports an outage, the device has failed to respond to a number of short-interval requests, and there's no additional need to retry.

Having said that... :-)

The subject line of this note is "False Positives protection", and I imagine that you're seeing alerts from things that aren't really a problem.

Could you tell me a little more about what kinds of devices have the problem, what the probe is, the poll interval, timeout, and the retry count/retry interval if it's a UDP probe? Thanks.

Rich
____________________________________________________________________
List archives: http://www.mail-archive.com/intermapper-talk%40list.dartware.com/
To unsubscribe: send email to: [email protected]

Reply via email to