On Wed, Jul 12, 2023 at 01:39:41PM +0300, Michael Schapira wrote: > Specifically, if you look at Figure 13, this is the *empirical* load for > *netspeed=1/1000* servers *for the strawman solution* where Khronos > watchdog queries are performed *at the same time granularity as NTPv4*. So, > a multiplicative factor of F should be shaved off this number. As you can > see, the actual load will be orders of magnitude lower than your > calculation. I don't see the orders of magnitude. The Figure 13 shows an average increase of 200-300x and my estimate was 87x for F=10 or 870x for F=1. It's not clear to me how many servers did you assume in your calculation the client was using normally and by Khronos, but there seems to be an agreement. With the 10x interval suggested in the Khronos draft the increase for the slowest servers would be still 20-30x by your estimates. I don't think that is acceptable. The draft needs to be clear that for pool.ntp.org the interval needs to be much longer and enforced across restarts. For example, if the Khronos poll happened after each suspend+resume cycle on all laptops, the impact on the load would be huge again. -- Miroslav Lichvar -- last-call mailing list last-call@xxxxxxxx https://www.ietf.org/mailman/listinfo/last-call
