I am working on cloud computing performance improvement model during live virtual machine migration. Due to the newly migrated virtual machines, ongoing virtual machine at destination physical machine faces interference which leads to performance degradation.
Does anyone have an idea how to determine and predict network I/O usage if VM is in public cloud?
Check some linux commands that measure network.
Then, perform some simple experiments to understand how they react to different workloads.
Then use techniques of prediction as SVR (Support Vector Regression) and ARIMA
(AutoRegressive Integrated Moving Average). There are several techniques to predict, so choose the one that best suits your reality.
Network Usage during migration? Or network usage after migration is completed?
I assume you mean the former and I will continue based on that assumption:
It depends on how you migrate, (in-band or out-band). In-band migration means that you are using a virtualized channel for each VM, and for each VM the requests and migration will served through a single channel. In this case, you will not see any impact on the other VMs performances regarding network, but you will notice a performance Impact on the performance of VM that is migrated during migration time.
If you are using the out-band migration, it means that a shared bandwidth will be used for migration among all the VMs and that is where the network contentions and interference can happen.
If you wanna calculate the network consumption during live migration it is the amount of data transfer and can be calculated by a function introduced in page 193, of this paper, "Remedy: Network-aware steady state VM management for data centers".
Network consumption for a cold migration is basically the amount of memory transfer.
- Badges
- Science method