How many clients does your master support?
The short answer: It depends…
A slightly longer answer: The M50 and M60 based designs can handle approx. 350 outgoing PTP messages per second. In Multicast this means that
s + d * n < 350 or n < (350 - s) / d
Where s is the number of sync messages (one-step) per second, d is the number of delay request messages per second and n is the number of slaves. We also have the announce messages, but these are usually kept at a very low rate, so it does not really matter. Assuming the rates are the same for sync and delay request this means:
- 1/sec: 349, but we say 250 to be safe. It depends a lot on how well the slaves distribute their delay requests over time.
- 2/sec: 174
- 32/sec: 9 (tested in our lab)
- 64/sec: 4 (tested in our lab)
In Unicast the formula is
(s + d + a) * n < 350 or n < 350 / (s + d + a)
Where s is the number of sync messages (one-step) per second, d is the number of delay request messages per second, a is number of announce messages per second and n is the number of slaves. Assuming the rates are the same for sync and delay request and the default announce rate of 1 every two seconds):
- 1/sec: 140
- 2/sec: 77
- 16/sec: 10 (tested in our lab)
- 32/sec: 5 (tested in our lab)
- 64/sec: 2 (tested in our lab)
Now, this is theoretical calculations (although practically verified to some extent) so keep that in mind and use a sound margin to any numbers.
For the M88, the number of slaves handled in unicast is defined by the product, e.g. M88-32 handles 32 slaves independent on the message rate. In multicast, the number of slaves handled is defined by wire speed, at 1 Gb/s this theoretically means ~100000 slaves at 1 delay request/sec.
Another thing of great importance, is that our concept of Edge Grandmasters and Gateway clocks significantly reduces the need of high message rates, due to the fact that these devices will be much closer to the clients, with less complex networks in between. 16 msg/s or less should be sufficient in most cases.