Mixture-of-Experts (MoE) showcases tremendous potential to scale large language models (LLMs) with enhanced performance and reduced computational complexity. However, its sparsely activated architecture shifts feed-forward networks (FFNs) from being compute-intensive to memory-intensive during inference, leading to substantially lower GPU utilization and increased operational costs.

Multi-interest sequential recommendations leverage users’ historical behavior to provide recommendations that match multiple interests. Most of these methods have not fully extracted higher-order information hidden in users’ interactions and have overlooked the multiple features of items. To this end, this paper proposes a multi-interest model called “multi-interest sequential recommendation with simplified graph convolution and item multi-features (SGCMF)”. Firstly, a simplified graph convolution module is designed based on bipartite graphs, which utilizes mean pooling to aggregate neighboring information and employs a feedforward neural network (FNN) for nonlinear transformations and combinations. This method reduces redundant information and captures higher-order relationships, thereby simplifying the complexity of modeling high-order interactions and improving prediction accuracy. Secondly, an item multi-feature extraction module is proposed, which represents item features with multiple vectors, and analyzes each feature from multiple perspectives while preserving important relationships between features. The model correlates multiple features of the item with user interests, thereby achieving a fine-grained analysis of user interests. Extensive experiments are conducted on five real-world scenarios, and the results are compared with state-of-the-art methods. The experimental results show that SGCMF outperforms other baselines.

Remote Procedure Call (RPC) is widely used in distributed systems and it usually needs to serialize data before transmission. Serialization accounts for a large proportion of the overhead in RPC and becomes a bottleneck for RPC communications. Because the size of the output serialized message cannot be predicted in advance, there could be multiple memory reallocations and copies in typical serialization libraries (e.g., FlatBuffers), which dominates the overhead. We propose the novel serialization library, zFlatBuffers, to eliminate these avoidable copies during the serialization process and realize zero copy during communication. Unlike the typical serialization library, FlatBuffers, the message generated by zFlatBuffers consists of multiple non-contiguous buffers due to its zero-copy nature. Moreover, we integrate zFlatBuffers with RDMA-based RPC systems. For RDMA Unreliable Datagram, we modify the message buffer of eRPC to enable it to transmit messages composed of multiple buffers. We also build the zRPC system based on RDMA Reliable Connection, which transmits the zFlatBuffers message by the scatter/gather function. Compared to the original FlatBuffers, zFlatBuffers improves the throughput of eRPC and zRPC by 11.2%-33.7% and 5.8%-53.6%, respectively.

Mapping-based cross-domain recommendation (CDR) can effectively tackle the cold-start problem in traditional recommender systems.However, existing mapping-based CDR methods ignore datasparse users in the source domain, which may impact the transfer efficiency of their preferences.To this end, this paper proposes a novel method named Neighbor Interaction-based Personalized Transfer for Cross-Domain Recommendation (NIPT-CDR).This proposed method mainly contains two modules: (i) an intra-domain item supplementing module and (ii) a personalised feature transfer module.The first module introduces neighbour interactions to supplement the potential missing preferences for each source domain user, particularly for those with limited observed interactions.This approach comprehensively captures the preferences of all users.The second module develops an attention mechanism to guide the knowledge transfer process selectively.Moreover, a meta-network based on users' transferable features is trained to construct personalised mapping functions for each user.The experimental results on two real-world datasets show that the proposed NIPT-CDR method achieves significant performance improvements compared to seven baseline models.The proposed model can provide more accurate and personalised recommendation services for cold-start users.

Compared with hardware implementation, the software packet scheduler uses the packet queuing data structure and a ranking function according to different dimensions to flexibly determine the packet dequeue order, which can significantly shorten the renewal cycles and increase the function deployment flexibility. The key data structure in prior work either bounds the number of rank or suffers from high computation overhead. In addition, they only support a single dimension and do not scale well. In this paper, we present Proteus, a software packet scheduling system that supports multi-dimensional and arbitrary numbers of ranks. We design a k-dimension heap data structure and develop “push” and “pop” algorithms to perform “enqueue” and “dequeue” operations. Furthermore, we implement a prototype of Proteus in software switch. Extensive experiments on BESS and numerical simulations show that Proteus can decrease the computation overhead, save the storage space and run much faster than state of the art.

In memory computing and high-end distributed storage demand low latency, high throughput, and zero data loss simultaneously from datacenter networks. Existing reactive congestion control approaches cannot both minimize queuing latency and ensure zero data loss. A token-oriented proactive approach can achieve them together by controlling congestion even before sending data packets. However, state-of-the-art token-oriented approaches only strive to optimize network-level metrics: maximizing throughput while achieving flow-level fairness. This article answers the question of how to support objective-aware traffic scheduling in token-oriented approaches. The novelty of Token-Oriented in-network Prioritization (TOP) is that it prioritizes tokens instead of data packets. We make three contributions. Via simulations over a hypothetical TOP system, our first contribution is demonstrating the potential performance gain that can be brought by TOP. Second, we investigate the applicability of TOP. Although the overhead of enabling necessary TOP features in switches is trivial, we find that mainstream commodity datacenter switches do not support them. We hence propose a readily-deployable remedy to achieve in-network prioritization by pushing both switch and end-host hardware capacity to an extreme end. Lastly, we implement a running TOP system with Linux hosts and commodity switches, and evaluate TOP in testbeds and with large-scale simulations for various scenarios.

Modern data centers are the host for multitude of large-scale distributed applications. These applications generate tremendous amount of network flows to complete their tasks. At this scale, efficient network control manages the network traffic at the level of flow aggregates (or slices ) who need to share the network with respect to operator’s proportionality policy. Existing slice scheduling mechanisms can not meet this goal in multi-path data center networks. Hence, in this paper, we aim to fulfil this goal and satisfy the congestion proportionality policy for network sharing. The policy is applied to the traffic traversing congested links in the network. We propose Uranus, a novel slice scheduler based on a combination of flow-level control mechanisms. The scheduler implements two-tier weight allocation to individual flows. Then, relying on a non-blocking big switch abstraction, slice weights are allocated at the inter-rack level by aggregating the weights of rack-to-rack flows. Finally, Uranus can dynamically divide the rack-level weight to its constituent flows. We also implement Weighted Virtual Congestion Control (WVCC), an end-host shim-layer that enforces weighted bandwidth sharing among competing flows. Trace-driven NS3 simulations demonstrate that Uranus closely approximates the congestion-proportionality and is able to improve the proportional fairness by 31.49% compared to the state-of-the-art mechanisms. The results also prove Uranus’s capability of intra-slice scheduling optimization. Moreover, Uranus’s throughput in Clos fabrics outperforms the state-of-the-art mechanisms by 10%.

Fuzzy support vector machine (FSVM) is a part of machine learning with its good classification effect. So far, there are two most commonly used FSVM models: FSVM on account of class core and fuzzy support vector machine on account of hyperplane that is over class core. Each has its own problems: FSVM on account of class core are dependent on the geometric shape of sample sets. Although FSVM on account of hyperplane that is over class core can solve the above problems to some extent. However, this algorithm has low generalization ability and high time complexity. Therefore, Inspired by these two common models, the paper proposes an improved membership function method. By analyzing and calculating the potential support vector sample points, adjustment factor is obtained, which drives the class core to adjust along the direction away from the outliers. In this way, membership of noise and outliers are reduced and the membership of support vector will also increase to some extent. In this paper, a new experimental comparison method is used, which can make the comparison of classification effect more obvious and convincing. The experimental part compares the proposed FSVM model with the above two FSVM models. It shows that the proposed algorithm improves the stability and classification accuracy to some extent.