[e2e] method to detect liveness of remote other than ping
spencer_dawkins at yahoo.com
Mon Oct 13 04:35:28 PDT 2003
Dear Jing Shen,
At some point, sites that
- filter ICMP echo,
- filter traceroute, and
- respond negatively to random TCP segments
aren't really participating in an end-to-end Internet.
In some (many?) cases, these sites may also be doing NATP, so
you really can't know how mnay hosts behind a NATP edge you're
monitoring liveness for, anyway.
Imagine a NATP presenting 1000 ports that represent 400 hosts,
30 of which aren't "alive" - how would you know this, without a
lot of site-specific information?
As a slight variation, some sites may be doing content
switching, so it would be interesting to know that the front-end
switch is operational but three of the seven servers behind it
are not, and some content may not be available, even though the
content switch is "alive" - again, how would you know this,
without site-specific information about their content switching
For better or worse, we probably don't have an Internet where
you can do this type of liveness detection without cooperation
from the monitored networks any more, at least not in the
And the monitored networks may choose not to cooperate, choosing
instead to (try to) give the appearance of 100-percent
availability based on their own internal liveness monitoring and
The thinking is something like "if I provide N+3 redundancy for
content switch backends, why would you care if one of my content
switch backends dies and I perform a successful switchover?"
Of course, your users may be affected by UNsuccessful switchover
--- Jing Shen <jshen_cad at yahoo.com.cn> wrote:
> thanks a lot.
> Our situation is that: we need to set up network
> availablity montoring across the backbone network. But
> some sites filtered out both traceroute and ping at
> their edge router. We could see in/out traffic in
> access switch, but sampling by SNMP every 5min is too
> long to realize a real-time monitoring.
> jing shen
More information about the end2end-interest