COMPARATIVE ANALYSIS OF SBE AND PROTOBUF METHODS IN HIGH-LOAD IOT GATEWAYS

Abstract

The rapid proliferation of Internet of Things devices places increasingly stringent demands on gateway infrastructure, which must continuously collect, encode, and transmit large volumes of telemetry data in real time. As the number of connected endpoints reaches millions, the choice of binary data serialization method becomes a critically important architectural decision that directly affects throughput, latency, CPU utilization, and memory consumption under high-load conditions. Despite the widespread use of serialization methods in distributed systems, their rigorous empirical comparison in the context of high-load IoT gateways remains insufficiently explored in the current literature. This paper presents the results of a comparative analysis of two binary data serialization methods — Simple Binary Encoding (SBE) and Protocol Buffers (Protobuf) — in a high-load IoT gateway environment. The study was conducted on a Raspberry Pi 5 mini-computer equipped with a 64-bit quad-core Arm Cortex-A76 processor clocked at 2.4 GHz and 4 GB of RAM, serving as the IoT gateway. The software, implemented in the Go programming language, received telemetry messages via a socket connection and performed their deserialization. Testing covered two message types with maximum payload sizes of 100 and 1000 bytes, transmitted in batches of 100, 1,000, 10,000, and 100,000 messages. The key measured metrics were total message processing time and CPU utilization level. The results indicate that neither of the evaluated methods holds an unconditional performance advantage. At small message sizes with high intensity — typical of dense sensor telemetry workloads — SBE demonstrates significantly lower processing time owing to its zero-copy in-memory architecture and statically compiled schemas. However, at large message sizes under extreme load, SBE yields to Protobuf both in processing time and CPU resource consumption: at 100,000 messages, SBE's CPU utilization reaches 95.7% compared to 67.1% for Protobuf. The findings confirm that the optimal choice of serialization method should be determined through careful evaluation of the load profile, typical message size, and available computational resources of the specific IoT deployment environment.