> Hi, > > On 31-05-16 19:34, [email protected] wrote: > >> I would suggent a double-check on first drop mechanism: resend probing > >> packet to no responding servers after serveral minutes. > >> a. If the server is running fine, only overloading, it may be able to > > The monitoring is not (only) about checking if a server keeps good time, it is > about checking if a server provides good time _to remote clients_ . > > >> reply to probing packet after some time. We have double-check to > >> confirm this, then go back to 20 mins checking cycle. No score > >> modification (addition/deduction) is performed in this scenario. > >> Further no-responding checks will still be double-checked. > > > > Maybe the NTP probing could be combined with some pings - just to check the > > amount of packet loss. > > This would imply that the server needs to respond to ICMP echo requests but > > would help the monitoring > > system to detect if a server is overloaded (maybe by comparing RTT times), > > totally offline or just a bit lossy. > > What does this detection matter? In the end, the server will not reliably > respond > to normal clients, so it should be removed from "the" pool. Remember, we only > have > one big pool with valid servers. From that pool, some servers are selected to > be > included in the answer to specific DNS queries. > > Note that you can request to exclude your server from the global or regional > pools. > Just reply to the automatic e-mail that the monitor sends you in case the > score > drops deep. > > > > Besides that, I have no packet loss to China or Taiwan from Germany (tested > > with the pool servers for CN and TW). > > I assume this will be the same for most other european countries, so maybe > > a monitoring system located in europe > > might help resolving this issue AND could make monitoring more reliable > > since you have two fully (carrier) independet > > monitoring locations. And hosting in Europe is really cheap. > > With two monitors, they may not agree with each other. What do you do in that > case? > For which DNS queries do you in- or exclude the system? > > As far as I can see, a (much more complex) multi monitoring system might > work, if > we define a pool for each individual DNS entry on country, region and global > level. > In that situation, each pool can have its own local monitoring system for > inclusion > of the server in that specific pool. However, you need long term committed > volunteers to run each monitoring system. Given that we have many countries > with > only a few servers, I guess this is not feasible. >
Being very well spread all over the world, couldn't RIPE Atlas probes be used instead or in addition ? > Once we have many servers (and better chance to find volunteers for > monitoring) in > a country, we don't have problems with overloading in the first place. So, in > that > case our current simple, single monitoring system will do. > > Just my € 0,02 > > Arnold > _______________________________________________ > pool mailing list > [email protected] > http://lists.ntp.org/listinfo/pool -- Hrant Dadivanyan (aka Ran d'Adi) hrant(at)dadivanyan.net /* "Feci quod potui, faciant meliora potentes." */ ran(at)psg.com _______________________________________________ pool mailing list [email protected] http://lists.ntp.org/listinfo/pool
