On Thu, Nov 28, 2013 at 6:16 PM, Daan Hoogland <daan.hoogl...@gmail.com>wrote:
> I don't think we should try to go beyond the hypervisor extensions on > the vms (unless it's the systemvms) is this what you mean? > > on the systemvm it seems we could indeed. Is that what you mean? > > on the others we should try to collect through the hypevisor and not > beyond that, would you agree? > @Daan, Apologize for slow response. Yeah, actually we have two approach methods: - using SNMP: I'm going to use pre-installed SNMP agent into VMs. Also, SNMP manager will be external pluggable devices. Problem now is what parameters we decide to monitor? - trying to explore hypervisor's ability in collecting performance data from VMs resided in. Problem is the hypervisor's limited in monitoring VMs. I prefer method 1. @Chiradeep, Okie I'm done with SNMP agent installed into VMs. What I'm thinking is: We use external SNMP managers opensource software, implemented as an external device, same architecture with other plugins like NetScaler of Nicira NVP. The advantage is scalable and pluggable. CloudStack send AS configuration to SNMP devices using their own API (similar to NetScaler's NitroAPI). AutoScale engine periodically request performance information to SNMP device and make AS actions (scaleup/scaledown) to CloudStack orchestration based on response data. So, problem now is: + what SNMP software we use? I found Net-SNMP but not sure it's okie. Need to try and review it. + what metric we decide to monitor for AS action? @guys: Any idea? Cheers, --Tuna