Anybody running a Cesium Frequency Standard or Rubidium Frequency Standard as a source for 10MHz or PPS?
Not at home
Normaly you would need such a thing only as holdover while your receiver isn’t synced.
Hmm, I am operating a bit at the other end of the scale from those kinds of units. One of my servers is a small OpenWrt device running chronyd and getting its time from a clique of local stratum one Raspberry Pis. Those run ntpd classic and in turn get their frequency/time via GPIO-PPS plus a customized version of gpsd, or the NMEA driver (without the PPS option) from various GNSS receivers, mostly different u-blox models, but also one MTK GPS-only receiver on an Adafruit board among them. I’ve been pondering to update and upgrade the OpenWrt device, but it sits behind a consumer-grade home router that doesn’t handle too many small packets very well, so the server already contributes less to the pool than it itself would be capable of, thus I’ve refrained from the update/upgrade effort so far. The stratum one servers are registered in the pool as well, but for monitoring purposes only as they have nominally dynamic IP addresses only.
Another device is a Raspberry Pi 2, this one getting its frequency and time from the PPS output of an older TP2700 PTP device (from before Symmetricom was acquired by Microsemi), plus some local private as well as static Internet NTP servers. I.e., it is at stratum one itself, and as it sits behind a more professional-grade router and a symmetric uplink, I am currently exploring the limits of possible bandwidth settings.
Nothing wrong with low end setups. I’m currently running my NTP servers exclusively on VPSs, all with chrony and all synchronized to official stratum 1 NTP servers from the national government of wherever the VPS is located.
Even a cheap 1-core VPS is able to handle several thousand requests per second, and only one of the VPSs I booked had trouble with keeping accurate time. It lapsed about 200ms every couple of hours, not enough to get dropped by the pool monitoring before it recovered, but still annoying so I removed it.
With many small VPSs I can contribute to a lot of different countries at small cost. Sure, the time is not as accurate as a GPS-synchronized dedicated setup, but it’s still good and I don’t have to open up my home network.
That said, my own stratum 1 server is definitely on my to-do list!
Sound good. I’m just wondering if having a RPi2 being fed by a PTP server doesnt defeat the purpose.
I’m looking at upgrading my router aswell to be able to handle more traffic and being prepared for 10G internet, which will come in the near future.
Added a UBlox GPS server with a LEA-5T chip to my setup to have a second source of pps signals as input for my Lantime M300MRS.
Works like a charm. Had to apply a fixed correction of 120ns compared to 10MHz in to roughly align both sources, however. Strange…
A relatively “corporate IT” answer here: the University of Cambridge Pool servers are shared with the University’s internal NTP service. They are three identical Dell PowerEdge R640 rack servers running Ubuntu 20.04 LTS and the stock Ubuntu chrony package. They’re hosted in three different University buildings across the city.
They would be vastly overspecified if NTP was their only role, but NTP at the University has always been something we run in a small corner of a server that primarily does something else. chrony on each server uses about 15% of a single CPU thread to process 6,000 packets per second. Almost all of that traffic is from the Pool.
For timekeeping we use various external NTP services (Jisc, NPL, and Cloudflare) along with a stratum-1 server run by our Department of Computer Science and Technology.
bjh21, Thanks for posting this. Very interesting to learn about NTP servers in the UK.
I’ve added one to my servers to sync against.
What always has puzzled me is that it appears not a lot of stratum1 zervers are exposed to the internet in the UK, comapred to other countries. The lowest appears to be stratum2, generally.
Any reason for this?
I’m late to this thread (as usual), but here’s a post describing my infrastructure:
And some measurements showing their stats in this post:
@bjohns Nice, low-cost setup!
(This picture is while they were still being put together, testing. I didn’t get one completed before it went in the rack)
Currently only one of these in the pool, handling around 4000qps utilising 10% CPU. The plan will be that all will be going in at some stage. Later this year I’ll be moving location and getting a 10x upgrade in bandwidth, mostly just because the price difference is negligible and most of it will be unnecessary, hence I started looking at things I could use it for.
There are 2 x Pi 3 Bs, 1 x Pi 3B+ and 1 x Pi 2 that I had around. Coupled with Lea6T chipped boards that I believe have been pulled out of Huawei devices of some sort. 3D printed modules that hold the Pi + GPS board and slide in to the metal carrier from a 19" brush panel.
These all serve internal NTP requests as well as various external servers that are in the pool. The Pi 3s sit at an average offset of 78/78ns while the Pi 2 doubles that. Ideally all would be 3 B+s due to the better networking, maybe I’ll upgrade them later. For now good enough for my use and good enough for the pool