[Spread-users] Jitter when process connects to spread?

Soila Pertet spertet at andrew.cmu.edu
Tue Mar 16 16:55:00 EST 2004


I am using spread for group communication in a real-time, fault-tolerant 
distributed system which trys to reduce the jitter experienced by the 
client application in the event of a server crash.

In my current experiment, I have a client that requests the time of day 
from a group of warm-passively replicated servers in a LAN. The normal 
round trip times for client requests is about 2ms. If I kill the primary 
server process, this event is quickly detected by Spread and I failover to 
another server with little effect on the client --- round trip time of 2.5ms

However, when I try to reintegrate a server back into the group, the 
SP_connect event causes a large spike in my round trip times, anywhere 
between 35-100ms. This happens even if I connect my server to a different 
group but within the same spread segment. Please let me know why the 
SP_connect event causes such a delays in message delivery on other 
processes running spread in the same segment. Is there anyway I can 
circumvent this?



More information about the Spread-users mailing list