Okay. What methodology are you using to show the correlation, so I can try to replicate your findings?
Run tcpdump a couple of hours before and after adding a server to the pool. Itād be interesting to see what you find.
RU zone resurrected. I opened the valve at home internet and it survives load correctly.
People in comments blame that Yandexā smart speaker is the cause of NTP flood. Yesterday evening server load drastically reduced. What was the reason? Increase of server power? Fixing of some firmware bug of a widespread device? Maybe both.
Welcome back, ru.pool.ntp.org. I am proud to be part of it again.
10 Likes
Welcome back 
and thanks for the news.
Maybe the āmonitoring onlyā server will come back and re-join the pool
really many NATās use the whole range of ports. For example (as far as I remember) Mikrotik can use ports starting from 1 unless to-ports range not configured. So I think that most of requests with port <1024 can be legitimate. They just come because of NAT. Maybe home soho routers (CPE) behave the same way.
See Figure 8 in this NIST paper for a discussion of NTP client source ports.
1 Like
As someone who is just setting up to serve the same country zone (yes, I responded to the ādistress callāā¦) ā mind if I ask, which NTP upstream(s) and which strata are you using?
The VNIIFTRI cluster is down, and while Iām lucky to have direct access to MSK-IX (and thus to the MSK-IXās stratum 1 server), Iād prefer if I could have more āreliableā upstreams than just one.
Yes, that was Yandex, who did this: ŠŠ± ŠøŠ½ŃŠøŠ“ŠµŠ½ŃŠµ Ń NTP-ŃŠµŃŠ²ŠµŃŠ°Š¼Šø / Š„Š°Š±Ń
One should google it. Maybe vniiftri was not ready for a habraeffect.
vniiftri have servers in another zones, see on site. People say that they were removed from DNS, but now some have already started working
You can try CDN servers
server time.google.com noselect
server time.facebook.com
server time.cloudflare.com
Google Public NTP serves leap-smeared time. We use this technology to smoothly handle leap seconds with no disruptive events.
it is not standard so it is not recommended
There are a few comments about Vninftri in telegram nag_public. Iās ānear telecom/ISPā talks/flood
Usually, many countries provide legal time services, typically provided by their navies, observatories or measurement authorities. Among such services, NTP is typically one of them, often public NTP servers. Doesnāt Russia?
Those were exactly the VNIIFTRI servers I mentioned (VNIIFTRI is essentially the Russian āNIST Time and Frequency Divisionā). As of yesterday, the entire cluster was offline.
As a matter of fact, they have just restored that service (half of that cluster is reachable from where my server sits in Moscow) ā so the immediate problem is resolved and I have just enough upstreams to abide by the ntppool guidelines.
1 Like
For the record: Š£ŃŠ»ŃŠ³Šø | Š¤ŠŠ£Š ŠŠŠŠŠ¤Š¢Š Š
OT:
NTP.org provides a wiki with Straum 1 and Stratum 2 servers.
https://support.ntp.org/Servers/StratumOneTimeServers
https://support.ntp.org/Servers/StratumTwoTimeServers
1 Like
NTP.org provides a wiki with Straum 1 and Stratum 2 servers
Yeah, this wiki is linked from the NTPPool guidelines, so that was the first place I looked.
Unfortunately, most of (RU) S1 servers in that list are inactive/unreachable. The ones that still work are personal hosts, and all of them are pretty far away from Moscow (judging by ping times), so not an ideal choice.
As Chinese and former employee of Tencent, I must said this post has too many misunderstand/speculation of Chinese ISPs/NTP status.
The main reason why Chinese NTP server canāt operate well is ISP bill for business is way more expensive than residential usage (100MBps monthly for example, ~2600 CNY vs 10 CNY).
Also thanks to IANA, only 350 million IPv4 address are assigned to China, meanwhile we have more than 1 billion netizen, so residential ISP canāt have static IPv4 address and 99.9% are behind NAT due to lack of IPv4 addresses.
Thus NTP servers are mostly operated by big companies or some government based institutions. Hobbyist NTP must be rich enough to bought private own business line (rich, I mean China is a developing country with median income 33036 CNY yearly, 2023)
Another issue is connectivity to āthe Internetā.
First of all, please bear in mind geography and physic, monitors mostly operated in US/EU, thatās 10000 KM alway from China, itās will takes 100ms+ by Einstein āSpecial relativityā, another 60ms by fiber losses, router hop delay etc. RTT should be more than 350ms, for NTP, thatās not good.
Secondly, there is just too many netizen, period.
If you had server in China, during night time, package loss rate will be increase drastic by huge amount of packages, UDP based protocol are first in line drop target of QoS for most of ISPs policy that I known of.
Thirdly, ISPs competition.
Yes. China has three major ISPs and lots of small ISPs, not āONLY one evil government controlled/censoredā Internet provider.
The way ISPs keep their customer including but not limited to: deteriorate other ISP package loss, little gifts/discount if you sign long term contract, setup higher fee for interconnect with ISP, cut off some other ISP fiber āby accidentā.
Oversea Internet package often came from other ISP instead of the ISP that end-user in, so drop others package is quite common if ISP hits high bandwidth usage.
3 Likes
Nearly twice server in ru zone before the downfall!
some of these servers are rented virtual machines, they will be turned off as soon as people get tired of paying for them. The other part are home servers. Letās see how long they will work
By the way, is it okay to use virtual machines for a time server? I thought that a good server (even no stratum-1) requires real hardware.
2 Likes
Ahh yeah, youmention the 30 VPS earlier. But itās looks good anyway.
And regarding VPS - if you need highly accurate time you wonāt use the pool 
You would run in your own network your own NTP or better PTP
You can have straum 6 server which is more accurate as stratum 1 with piggy hardware
Sometimes the host have a high frequency drift which canāt be compensatet anymore by ntp (>= 500ppm) you need / can try chrony wich will work.
Most of my servers a vps are running fine and they arenāt worse than a rented hardware server. But both of them will fail in comparsion to a ntp server with āhardware accelearionā like ntp in FPGA.
And as bas often said - you can build your own accurate ntp server for a couplde of bucks with a ex. a Pi and any good GNNS receiver (Garmin, U-Blox or Trimble or ā¦)
It depends on a lot of unstated details, so itās not really possible to say simply yes or no.
For a pool server, accuracy in the single-digit milliseconds is more than enough to stay in the pool. For a pool monitor, on the other hand, Iād want appreciably less than 1 msec offsets seen locally, and as symmetric path as possible to handoff(s) to other ISPs (AS numbers).
Thereās a previous thread here that discusses the age-old ābare metal only for NTPā which has become less mandatory depending on details like type of hypervisor, CPU virtualization support, etc: Cringe at thought of NTP server on a VM?
Would you say itās fair to describe the price difference between business and residential internet pricing in China as ābusiness internet subsidizes home (residential) internetā?
I presume when you say nearly all home internet is behind NAT, youāre talking about an ISPās NAT, so that numerous homes share a single public IPv4 address? For clarity, itās best to refer to that type of NAT as āCGNATā (carrier-grade NAT). CGNAT implies no ability to forward NTPās UDP port 123 to a specific device on the home network, so no possibility to host a public NTP server or any other service requiring a specific port or ports.
Regarding most NTP pool monitors being outside China, Iām pretty confident monitors inside China would be welcome. If you know people with systems in Chinese data centers interested in helping out, the IP traffic, CPU load and disk space involved are all very small. [1] You might get them to bring up a pool server first and introduce themselves on this forum. Note I do not represent the pool, Iām just an enthusiastic participant.
I think I recall but canāt find right now reading here a message where @Ask mentioned the possibility of donating a bare metal or VM system with good timekeeping (such as a GNSS-connected NTP server or appliance in the same datacenter) running Linux/FreeBSD which he would configure as a monitor. I donāt know if he really wants one more server to manage, though. If it were me Iād much prefer someone willing to manage it themselves. The basic requirements are Linux or FreeBSD with 24x7 solid symmetric connectivity and established ability to run Chrony or ntpd showing stable performance. Given advances in x86/x64 hardware virtualization assistance and type 1 hypervisors that use them, while bare metal is always preferable it is possible do manage well with a VM.