Oh nice, so that kidn of explains the results we see, like on the experiment above
If that is what you want to believe, up to you, obviously. To me, it seems more plausible that as mentioned, some people have requested to have the BE zone added to their server manually, and in some cases, those requests have been granted.
I was talking about how the system works today. And from that perspective, as per my interpretation of Askās latest post (leaving zones in place, just backfilling them with servers from other zones), the implementation of the GeoDNS service does not need to be changed, just how it is parametrized. And that is done elsewhere in the system, around the same place where also the biased handling of the IP versions is encoded currently. I.e., the logic of how the zones are filled needs to be changed, but not the GeoDNS service, which is flexibly serving what it is being handed.
Sorry this isnāt the case, as soon as you lower the speed other zones dissapear.
At 512Kbit itās Country + Continent
At 12Mbit itās Country + Continent + Global
Try yourself. At least the manage-page shows this. Did you set the servers low enough in your evaluation?
Back to the original topic: Here are the rates Iāve seen in the past week. Note that the yaxis scale is now requests/second.
There is a noticeable diurnal pattern. There may be a ~15% increase on 12-15/12-16. I donāt see a significant change such as Profile - MagicNTP - NTP Pool Project showed in his first graph.
There are several abusive clients, but these peak at ~hundreds of requests/second.
There were ~5 1N14 requests / second (see arduino NTP library and this)
I didnāt see anything unusual.
1 Like
Steve,
Can you do the same measurement for 512Kbit rate?
As for a big server these arenāt a problem.
But when you set it low, it spikes from time to time very high.
I realy want to know what happens then.
These abusive clientsā¦are in my opinion the problem, that caused my routers to slowdown my network.
BTW, the MicroTik still didnāt slowdown.
I still believe those peaks are made by GSM operators with CGNAT.
So, can you test again with 512kbit for a week?
@stevesommars, @avij, thanks for your respective analyses! Not sure what your take from this, then, but based on those, I think it might have indeed just a glitch of some DNS server getting stuck with the IP address of my server, as @avij described. But I guess at the end of the day, weāll probably never know for sure what happened.
@stevesommars, the slight increase on 12-15/12-16 could have been because I took my server out of the pool for a few days. I have now re-added it, but still with lower netspeed than I had it on before I noticed the event that was the main subject of this thread. I might be increasing the netspeed a slight bit further beyond where it is now, but probably not go full throttle again. Iāll think about how to possibly set up some proactive monitoring to perhaps detect, and record, this type of event going forward. Right now, I am only recording short-term peaks, but not those gradual increases. Though it is unclear how likely such an event is to occur againā¦
Thanks again for your support in investigating this, much appreciated!!
I reduced the rate for my BE server from 500Mbps to 512kbps as of 2025-12-17 20:00
I plan to keep this configuration for at least 24 hours.
The server is in the BE zone. It is not in the EU or global zone.
After the low speed test completes, I plan to configure my BE server to the same rate that Profile - MagicNTP - NTP Pool Project is running. We should get similar rates even though weāre using different counting techniques
Nice, thanks, curious what happens to your server as well as the load it will show. 
@stevesommars, I just increased the netspeed by another notch to 1.5GBit now. But I think that is as far as Iāll go on this one going forward. Had two scares from other providers in the last week, donāt want anyone to get any ideas on this one. And the CPU isnāt the fastest, so I also would like to keep itās usage in check.
My servers are mostly limited by traffic volume, CPU (hard or fair use), one by bitrate, and only three small ones by packet rate (and there, I donāt even know the actual rate they see/can cope with, I just fiddled with the netspeed until things worked well). That is why I usually measure the load in multiples of bit/s, rather than packets/s.
Anyway, instantaneous packet rate right now is slightly below 3500 packets/s. I think that rate on average would be ok for me in the long run.
Showing my BE server rates for 24 hours after reducing speed to 512Kbps
I plan to continue for at least another 24 hours, to see if/how the drop-off continues. I donāt see unusual patterns.
Looking at netspeed & rates a bit. Before 500Mbps gave about 500 requests/second. Compared to Profile - MagicNTP - NTP Pool Project 1.5Gbps gave 3500 packets/second. Is that number requests/sec or (requests+responses)/sec ?
Itās incoming only, which now sometimes scratches at, or even briefly exceeds, 4000 requests per second (actually, it is all incoming traffic on the VPS, but NTP is essentially the only thing the instance is doing, and at this rate, NTP is absolutely dominating the traffic mix).