The growing adoption of AI-Generated Content (AIGC) has made large-scale processing of multiple Generative AI (GAI) training jobs a key strategy for improving cost-efficiency in computing clusters. However, the distributed nature of GAI models, together with inherent network bottlenecks, imposes significant challenges on system performance. Moreover, differences in training purposes, variations in model sizes, and asynchronous lifecycles create a dynamic environment. As a result, the coexistence of multiple GAI training jobs in a computing cluster exacerbates problems such as resource misallocation, fragmentation, and network contention, leading to low resource utilization and inefficient training performance. These motivate us to explore an efficient resource scheduling approach for completing multiple GAI training jobs. Accordingly, we introduce an intrinsic topology-aware scheduling framework designed to ensure flexible scheduling and efficient distributed training of GAI models. To address the trade-off between the number of concurrent jobs and the communication contention they generate, we formulate a multi-objective optimization problem with two objectives: maximizing the utility of GAI jobs and minimizing communication bandwidth. We then propose the Diffusion Model-based AI-Generated Resources Scheduling (DARS) algorithm, designed to capture dynamic, high-dimensional environments and generate optimal resource scheduling decisions. DARS employs a denoising diffusion process to iteratively refine noisy resource allocations into optimized scheduling decisions. Subsequently, we replace the policy network of Deep Reinforcement Learning (DRL) with DARS to address environmental uncertainty and enhance efficiency. Finally, the simulation results confirm that the proposed algorithm outperforms existing approaches.

With the rise of AI-Generated Content (AIGC), multi-tenant training in large computing clusters has become a prominent trend. However, when multiple tasks from different tenants of varying scales coexist in a cluster, they pose significant co-resident eavesdropping risks. Unfortunately, distributed training tasks are particularly prone to leaks of model parameters and data when they receive uniform security protection. Research on co-resident eavesdropping and hierarchical security defenses for distributed training remains sparse. To address this, we propose a novel Grouping Task Migration Mechanism (GTMM) that integrates server grouping with task migration. First, servers are clustered according to their security levels, enabling tailored protection. We then cast task migration as a multi-objective optimization problem to mitigate co-resident threats. Next, a DDQN-based task-migration algorithm derives optimal migration decisions. Lastly, our experiments demonstrate that GTMM consistently outperforms baseline methods.

With the prosperity of AI-Generated Content (AIGC), efficiently scheduling multiple Generative AI (GAI) distributed training jobs in a computing cluster has become crucial for pursuing higher cost-effectiveness. However, the resource-intensive nature and frequent communication demands of distributed training exacerbate resource fragmentation and network contention, resulting in low utilization and high latency. To this end, we propose an intelligent and dynamic resource scheduling method. Firstly, we propose an innovative scheduling analytical model that describes heterogeneous computing resources, communication contention, and the parameter synchronization architecture. We then formulate it as a multi-objective optimization problem. Next, we propose a Diffusion Model-based AI-generated Resource Scheduling (DARS) algorithm, to capture dynamic and high-dimensional environment and generate the optimal scheduling decisions. Finally, the policy network of deep reinforcement learning (DRL) is replaced with the proposed DARS to address the environmental uncertainty and enhance efficiency. Simulation results demonstrate that our proposed algorithm outperforms associated algorithms.

With the prosperity of AI-generated content (AIGC), multitenant training in AI task clusters has become prevalent. To improve resource utilization, multiple tenants will coexist on the same server, while malicious tenants may exploit side-channel to pose significant co-resident eavesdropping risks. Due to the extensive attack surface, distributed training tasks are particularly vulnerable to model parameters and data leakage when subjected to the same level of security protection as inference tasks. Moreover, traditional security mechanisms, reliant on static encryption or isolation, suffer from high overhead, passive defense and poor scalability, failing to address the dynamic resilience requirements of AI clusters. However, the research on proactive resilience enhancement in AI clusters is almost blank. To fill this gap, we devise a grouping task migration mechanism (GTMM), which jointly considers adaptive server grouping and proactive task migration. Specifically, we first employ an adaptive server grouping algorithm to classify servers, offering customized protection based on tenants’ security requirements. Then, we formulate the task scheduling process as a multiobjective optimization problem for making a tradeoff between security, power consumption, and load balance. Next, we propose a deep reinforcement learning-based task migration algorithm to separate tenants that have completed co-residency for mitigating co-resident threats. Lastly, the simulation experiments demonstrate that GTMM’s security outperforms the baselines with only an affordable performance degradation.

Identifying Decentralized Applications (DApps) from encrypted network traffic plays an important role in areas such as network management and threat detection. However, DApps deployed on the same platform use the same encryption settings, resulting in DApps generating encrypted traffic with great similarity. In addition, existing flow-based methods only consider each flow as an isolated individual and feed it sequentially into the neural network for feature extraction, ignoring other rich information introduced between flows, and therefore the relationship between different flows is not effectively utilized. In this study, we propose a novel encrypted traffic classification model IIT to heterogeneously mine the potential features of intra- and inter-flows, which contain two types of encoders based on the multi-head self-attention mechanism. By combining the complementary intra- and inter-flow perspectives, the entire process of information flow can be more completely understood and described. IIT provides a more complete perspective on network flows, with the intra-flow perspective focusing on information transfer between different packets within a flow, and the inter-flow perspective placing more emphasis on information interaction between different flows. We captured 44 classes of DApps in the real world and evaluated the IIT model on two datasets, including DApps and malicious traffic classification tasks. The results demonstrate that the IIT model achieves a classification accuracy of greater than 97% on the real-world dataset of 44 DApps, outperforming other state-of-the-art methods. In addition, the IIT model exhibits good generalization in the malicious traffic classification task.

Cache attacks pose a serious security threat to cryptographic implementations in processor architectures. In this paper, we first propose cache attacks against Blowfish, which can break the protection of key-dependent S-box. This attack targets at the subkey calculation of Blowfish, and fully exploits features of the subkey calculation to construct a leakage equation group about the key. Without any knowledge of plaintext and ciphertext, the attacker only needs to obtain the cache leakage once to recover a variable-length key in minute-level time. More than that, we establish a leakage model for cache attack situations to evaluate the exhausting space of the intermediate value of block ciphers, and estimate the time complexity of cache attacks. In our experiments, we perform Flush + Reload and Prime + Probe attacks and recover the random key of Blowfish in OpenSSL 1.1.1h in 4 minutes. Furthermore, we have applied our attacks to existing systems, such as JavaScript-blowfish and Bcrypt. Our attack on JavaScript-blowfish can recover any plaintext input by the user. As for Bcrypt, our attack can recover the hash values stored in the database, thereby allowing attackers to impersonate the user’s identity.