[Spread-users] Membership timeout calculation?

[gulekim at samsung.co.kr]

Hi, all.

In Spread user guide, in a small local area configuration of daemons,

"Basically, the smallest timeout should not be less than twice the standard kernel scheduling delay (often 10~20ms)
plus twice the packet latency of the network. Alos, the smallest timeout should increase as the number of daemons increases.
The Token, Gather, and Form timeouts should be sufficient to allow a full rotation of the token (so each machine gets it
and has time to do some work while holding it) plus some slack for an occasional retransmission or delay."

Now questions:
1) Why the smallest timeout should be less than above (twice scheduling delay + twice packet latency)?
2) In a LAN configuration, the packet latency is negligible. am I right?
3) The number of daemons increase smallest timeout. how much?
4) How can I calculate full rotation ot the token?
5) What do you mean by slack?

Finally,
If you design new configuration membership timeout and have to make configuration as fast as detecting token loss or network fail,
Just fill out timeout value using this configuration.

--> configuration:
1) Local Area Network (no router, 100Mbit/s), only on segment configuration
2) 10 machines, 10 daemons, 200 clients (1 daemon - 20 clients)
3) Windows 2000 or Windows XP or Solaris

--> timeout:
Token_timeout:
Hurry_timeout:
Alive_timeout:
Join_timeout:
Rep_timeout:
Seg_timeout:
Gather_timeout:
Form_timeout:
Lookup_timeout:


Can you do that? Please help me.

Thank you.

Bye.