|
At 09:36 AM 4/26/00 +0800, Wensong Zhang wrote:
>For UDP datagrams, we still create entries for state with the timeout of
>IP_MASQ_S_UDP (5*60*HZ) by default, then all UDP datagram from the same
>source to the same destination will be sent to the same server. Therefore,
>in some documents, I write a small note that we call data communication
>between a client's socket and a server's socket {\bf connection}, no
>matter it talks TCP or UDP protocol. :)
>
>Wayne, you probably need to know the communication behavior of ntpdate,
>then it will be easier to know why the ntp service doesn't work well with
>least connection scheduling method.
>
>Thanks,
>
>Wensong
Thanks, Wensong, for taking time answering this question.
I do appreciate it very much. The algorithm has been
changed to round-robin now, it does work better, but it
still has problem on the micro scale. I will take a look
ntpdate source code to figure out if there is anything could
be the cause. That utility is a big animal itself.
If the director CPU was under power, would that cause
this kind of problem?
|
