To reduce intra/inter data center traffic, having knowledge of inter-VM communication traffic can be very useful. Using inter-VM traffic information, we can effectively place related VMs on the same server/cluster/rack to reduce cloud traffic.
Question is: Is it possible to predict inter-VMs traffic of a task that its VMs should communicate to each other (e.g., HPC) or not? If yes, I am looking for a reference. Because some papers assumed having an inter-VM traffic matrix without mentioning how they get this matrix. I would like to stress this point that I talk about processing-intensive applications, not web services. Thanks.
I see following are the parameters that needs to be looked into while designing for Inter VM communication
- Functionality achieved by Inter VM (is it Hello Processing/Keep Alive or data base Sync, Protocol Sync- meaning state machine sync, Load Balancer needs etc)
- Based on the above , volume of data to be sync will vary
- Also the freq of Sync(periodical, @ what granularity -ms, Sec etc, On demand or event driven, !)
- Deployment scenario - Is this any life critical system , implying the impact of any miss in InterVM comm
- Any need of HA for such Commn channel
From my 2 cents once the reqs are done then we can design using
- IPC available in any Processor host
- Using OVS in various formats
- Proprietary implementations if any
Check this. It may be insightful.
Look Who’s Talking: Discovering Dependencies between Virtual Machines
Using CPU Utilization
https://www.usenix.org/legacy/event/hotcloud10/tech/full_papers/Apte.pdf
- Badges
- Science method