[time] GPS Time Server Survey
Fri Apr 25 18:28:49 UTC 2008
> So an wild estimate would be 300 packets a second.
> How many NTP users would that translate to realistically?
At poll interval 64, 19200. At poll interval 1024, 307200. (And, if
sufficiently badly behaved, one. :-)
Absent severe misbehaviour, I'd be very surprised if it topped out at
less than 1000 clients, and somewhat surprised if the practical ceiling
were under 10000.
My pool host (220.127.116.11) is a SPARCstation IPX, which means (according
to the jwbirdsall reference) 28.5 MIPS. I don't know how many clients
it's serving, but I've made no particular attempt to limit it, and it
seems to be not only handling them, but handling them without impairing
its other functions (it's also my house router, handling my uplink, my
house routed subnet, and my house nonrouted subnet).
I just now did a 1000-packet tcpdump for port 123 on its uplink. It
took 42 seconds to capture 1000 packets, so there's at least one pool
host which is handling only some 23-24 pps even though no particular
attempt has been made to limit it. Another 1000-packet capture
involved 328 distinct peer addresses; I tried a 10000-packet capture,
which gave me 1593 distinct peers - more hosts than the entire pool.
/~\ The ASCII der Mouse
\ / Ribbon Campaign
X Against HTML mouse at rodents.montreal.qc.ca
/ \ Email! 7D C8 61 52 5D E7 2D 39 4E F1 31 3E E8 B3 27 4B
More information about the pool