International Journal of Network Management Компьютерные сети и коммуникации
Toward eBPF/XDP‐Based In‐Kernel Service Filtering for Efficient Multicast‐Based Service Discovery
Аннотация
ABSTRACT Multicast‐based service discovery protocols such as mDNS/DNS‐SD are widely used in IoT environments. However, in heterogeneous IoT deployments, a large fraction of incoming discovery queries do not correspond to any services provided by a given device, resulting in substantial packet‐processing overhead despite producing no useful outcomes. This inefficiency is problematic for resource‐constrained IoT devices. Based on the observation that the distinction between query‐hit and query‐miss packets can be determined using only a compact, query‐oriented view of service state, we propose an in‐kernel early filtering approach that projects service state into kernel space and performs early service matching in kernel. The proposed approach is realized using a Service State Projector and an Early Filtering Module implemented on eBPF and XDP. Experimental results show that, at a 95% query miss rate, throughput increases to over three times that of the baseline, while sustaining nearly 100% service responsiveness and maintains low order‐matched response lag. It also reduces total CPU utilization from over 20% to approximately 7% under a 500‐QPS workload, and reduces the number of system calls by more than 62.5%. Even when no filtering benefit can be realized, the additional processing cost remains limited, with only a 1.27% increase in total CPU utilization under saturated load and less than a 1% reduction in throughput.