Interdomain I/O Optimization in Virtualized Sensor Networks.

Area of Science:

• Computer Science
• Networking
• Virtualization

Background:

• Virtualized sensor networks consolidate sensing services on shared hardware for efficiency.
• Inter-VM data transfers are crucial for analytics and collaboration but face performance bottlenecks.
• Traditional TCP/IP communication via virtual network interfaces leads to high I/O overhead.

Purpose of the Study:

• To propose an optimized interdomain communication approach for virtual machines (VMs) within the same hardware.
• To enhance inter-VM communication performance by minimizing kernel interactions and I/O latency.

Main Methods:

• Implemented a shared memory approach managed by the hypervisor, bypassing virtual network interfaces.
• Utilized a customized kernel module for shared memory creation and bidirectional event channels.
• Employed state flags in a circular buffer to optimize data transfer and reduce system calls.

Main Results:

• The proposed shared memory approach achieved five times higher throughput compared to traditional TCP/IP communication.
• Demonstrated a 2.5 times reduction in latency for inter-VM data transfers.
• Minimized system calls and kernel context switches, improving overall VM I/O performance.

Conclusions:

• Shared memory offers a superior alternative to TCP/IP for inter-VM communication in virtualized sensor networks.
• The optimized approach significantly improves data transfer efficiency and reduces overhead.
• This method enhances the performance and scalability of virtual sensor networks.