I configured an ADS to connect on 10 different dacs site:
*ads*dacs*doMultiConnect : True
*ads*dacs*domainList : A, B, C, (...)
*ads*dacs*A*port : 8211*ads*dacs*B*port : 8212*ads*dacs*C*port : 8213(...)
My 10 sites are located all around the world: Hong Kong, London, New York, etc.
The ADS is running on RHLE 6.5.
Since this change, the ads hangs randomly (~once per week):
- the ads process is still running, but it's not doing anything
- Netstat shows all the rssl connections as "CLOSE_WAIT" instead of "ESTABLISHED"
- "adsmon" gives "header is old"
- "kill" fails to terminate the process. We have to use "kill -9"
According to the dacs manual chapter 5.9:
Though no technical limit exists to the number of DACS daemons you can run on a single machine, the practical limit on a single machine is under 12.
According to the dacs manual chapter 7.6.11:
In some cases this feature [DACS Sink Daemon Usage Control] degrades performance for the application, because sending usage data to the DACS Sink Daemon adds a finite amount of CPU usage. For this reason, do not use this feature with clients such as an ADS or any other application sensitive to performance degradation.
An exception to this rule is when the ADS is configured for Multi-connect. In this case, you must enable the DACS_SNKD_P2P_USAGE / DACS_SNKD_P2P_USAGE_SIZE, thus ADS performance deteriorates linearly as the number of DACS daemons (to which it connects) increases.
I have "only" 10 dacs.snkd running, but I guess it's already too much.
- How could I troubleshoot my problem?
- Is there some specific OS parameters I should tune?
