Large Language Models (LLMs) have shown remarkable capabilities as autonomous agents; yet existing benchmarks either focus on single-agent tasks or are confined to narrow domains, failing to capture the dynamics of multi-agent coordination and competition. In this paper, we introduce MultiAgentBench, a comprehensive benchmark designed to evaluate LLM-based multi-agent systems across diverse, interactive scenarios. Our framework measures not only task completion but also the quality of collaboration and competition using novel, milestone-based key performance indicators. Moreover, we evaluate various coordination protocols (including star, chain, tree, and graph topologies) and innovative strategies such as group discussion and cognitive planning. Notably, gpt-4o-mini reaches the average highest task score, graph structure performs the best among coordination protocols in the research scenario,and cognitive planning improves milestone achievement rates by 3%. Code and datasets are publicavailable at https://github.com/ulab-uiuc/MARBLE.

Encryption of network traffic has become an inevitable trend. As an important link to Internet encrypted traffic analysis, encrypted traffic service classification can provide support for the coarse-grained network service traffic management and security supervision. But traditional DPI method cannot be effectively applied in an encrypted traffic environment, and the existing methods based on machine learning have two problems in feature selection. One is the complex feature classification over costing problem, the other is the TLS-1.2 suited method is no longer applicable to TLS-1.3 handshake encryption. To solve these problems, in this paper, we consider the differences among encryption network protocol stacks and propose a method of encrypted traffic service classification combining with capsule neural network in a multi-protocol environment by using multi-PDU lengths as the features, making full use of Markov property between PDU length sequences and being suitable to TLS1.3 environment. The feature makes our method much faster than others in feature extraction. Our control experiments on ISCX VPN-nonVPN dataset show that our method achieves a satisfactory performance (0.9860 Pr, 0.9856 Rc, 0.9855 F1), which is superior to the state-of-the-art methods.