Hi Tuna, I boldly diagrammed out what we talked about here:
https://cwiki.apache.org/confluence/x/M6YTAg The idea is to keep the monitoring part separate from the autoscale decision. So, the monitoring can be SNMP/RRD/whatever. Scale-up using reconfiguration then becomes a mere matter of modifying the autoscale service. On 11/25/13 8:57 AM, "tuna" <ng.t...@gmail.com> wrote: >Hi guys, > >At CCCEU13 I talked about the AutoScale without NetScaler feature working >with XenServer & XCP. For anyone don¹t know about this feature, take a >look into my slide here: >http://www.slideshare.net/tuna20073882/autoscale-without-netscalerccceu13. > >Chiradeep and I had a short talk after the presentation about how to >improve the AutoScale feature in future. We agreed that: > >+ Need to remove Load Balancing feature from AutoScaling. That¹s very >simple to do. >+ Need to use SNMP for monitoring not only instance-level but also >application-level. >+ Also, supporting well KVM hypervisor > >So, I blow up this thread for all of you guys to discuss the way we >design that feature, such as: >+ technical side, how to integrate effectively SNMP into CLoudStack. >Where do we put SNMP monitor components into infrastructure? etc >+ user experience, how user configure that feature with SNMP monitoring. >I image that user can figure out they need AutoScale for which of >following items: application, protocol (tcp, udp), port, bandwidth, disk, >cpu and memory also, etc >+ How about autoscale action, not just only deploy or destroy VM, we need >maybe dynamically increase-decrease memory/cpu, nic bandwidth, disk,Š > >Personally, we should think about a completely autoscaling feature. > >Cheers, > >‹Tuna
