Observation on $5 a month NTP servers


#1

For the past 2 months I’ve been running two linode instances as pool severs. tick.no-such-agency.net and tock.no-such-agency.net.

They are the smallest instances linode offers 1cpu, 1TB transfer, 1G ram and cost $5 a month each, running ntpsec. I’ve got both of them set to 1G for both IP4 and IP6. So far I’m seeing about 550GB/month of traffic on each node. Pretty good stability and accuracy (unlike the Digital Ocean nodes I tried which had +/- 100ms excursions).

I’ve been very pleased with the results, for $120 a year I’m adding a lot of capacity to the pool (plus I’m using them for monitoring some work stuff so really the pool use isn’t costing anything I wasn’t already spending).

A few interesting observations. IPv4 traffic is fairly steady at 1.5 Mb/sec with a minimal time of day component to the traffic volume. IPv6 traffic varies on a 24hr cycle, dropping by 50% at around 08:00 UTC and peaking between 22:00 and 02:00 UTC. However IPv6 volume is a fraction of IPv4, averaging only 200Kb/sec.

CPU utilisation on the small instance at the current traffic level is ~20%.

No significant difference between the east and west coast servers in terms of traffic.

Overall traffic grew almost immediately with speed changes and then climbed ~10% over the first month and less than 5% in the second.


#2

Digital Ocean works fine for me.


#3

DigitalOcean used to work fine for me. (As close to perfect as a typical VPS can be, anyway.) Then… it didn’t.

I didn’t do anything, as far as i know, but one day the clock started swinging up to about 120 ms.

(Current version of NTP reference implementation.)

I pegged the poll interval at 8 seconds – not without permission, of course – and it’s “fine” now.


#4

Where do/did you have your droplet/s?

I got mine in Netherlands.


#5

California and and US east coast. Both showed +/- 100 or more swings. Linode has been far better for me.