Machine learning (ML) has become a cornerstone of key applications at ByteDance. As model complexity and data volumes surge, data management for large-scale ML workloads faces substantial challenges, particularly with recent advances in large recommendation models (LRMs) and large multimodal models (LMMs). Traditional approaches exhibit limitations in storage efficiency, metadata scalability, update mechanisms, and integration with ML frameworks. To address these challenges, we propose Magnus, a holistic data management system built upon Apache Iceberg. Magnus integrates innovative optimizations across resource-efficient storage formats optimized for large wide tables and multimodal data, built-in support for vector and inverted indexes to accelerate data retrieval, scalable metadata planning with Git-like branching and tagging capabilities, and high-performance update/upsert based on lightweight merge-on-read (MOR) strategies. Additionally, Magnus provides native support and specialized enhancement for LRM and LMM training workloads. Experimental results demonstrate significant performance gains in real-world ML scenarios. Magnus has been deployed at ByteDance for over five years, enabling robust and efficient data infrastructure for large-scale ML workloads.

Influence maximization is the problem of finding a small set of seed vertices in a social network, so that the accumulated influence under certain propagation models is maximized. There are two problems about the existing methods: the influence spread of heuristic method is not stable, and the time complexity of greedy method is very high. In this paper, we devote to further reduce the gap between influence spread and running time. Firstly, we assess the comprehensive measurement of vertex, which provide a lightweight evaluation criterion rapidly for every vertex, Next, we construct the candidate vertices set from the communities, which reduce the search scope compared to the whole vertices, Finally, we tackle the efficiency issue of influence maximization from two complementary views. One is to propose the heuristic algorithm based on community candidate vertices (HBCC), attain a more competitive and stable influence spread. The other is to propose greedy algorithm based on community candidate vertices (GBCC), significantly reduce the running time. We carry out experiments on three real-world public datasets, the experiments have verified the effectiveness of the proposed methods compared to the baseline methods.