Daniel P. Ritter

Daniel Lindner, Daniel Ritter, Felix Naumann

Primary key (PK) and foreign key (FK) constraints are widely used for query optimization. Knowledge about additional data dependencies, such as order dependencies, enables further substantial performance improvements. However, such dependencies are not maintained by database systems or are even unknown to the user. Identifying and validating relevant dependencies automatically and efficiently remains an unsolved problem. This paper presents a system that (i) recognizes dependency candidates for optimization, (ii) efficiently validates their applicability, and (iii) optimizes query plans using valid dependencies. First, we demonstrate the performance impact of optimization techniques using data dependencies additional to PKs and FKs. Using rewritten SQL queries, we empirically show that data dependencies improve performance for a wide range of analytical database systems and benchmarks. Second, we present how to integrate data dependencies into a system to use them without (i) manual declaration and maintenance or (ii) SQL rewrites. Our integrated and fully automated system matches the performance of dedicated SQL rewrites: compared to using only PKs and FKs, queries improve with geometric mean speedups of 35 % for TPC-DS and 29 % for JOB. Individual query latencies drop by more than 90 %. The dependency discovery overhead is orders of magnitude lower than the latency improvement of a single workload execution.

Junxiong Wang, Wen-Ding Li, Daniele Paliotta et al.

Effective reasoning is crucial to solving complex mathematical problems. Recent large language models (LLMs) have boosted performance by scaling test-time computation through long chain-of-thought reasoning. However, transformer-based models are inherently limited in extending context length due to their quadratic computational complexity and linear memory requirements. In this paper, we introduce a novel hybrid linear RNN reasoning model, M1, built on the Mamba architecture, which allows memory-efficient inference. Our approach leverages a distillation process from existing reasoning models and is further enhanced through RL training. Experimental results on the AIME and MATH benchmarks show that M1 not only outperforms previous linear RNN models but also matches the performance of state-of-the-art Deepseek R1 distilled reasoning models at a similar scale. We also compare our generation speed with a highly performant general purpose inference engine, vLLM, and observe more than a 3x speedup compared to a same size transformer. With throughput speedup, we are able to achieve higher accuracy compared to DeepSeek R1 distilled transformer reasoning models under a fixed generation time budget using self-consistency voting. Overall, we introduce a hybrid Mamba reasoning model and provide a more effective approach to scaling test-time generation using self-consistency or long chain of thought reasoning.