CS Paper Radar

Fast and Theoretically-Efficient Batch-Parallel Link-Cut Trees, Euler Tour Trees, and Treaps

Quinten De Man, Laxman Dhulipala

cs.DS ·submitted 2026-09-13 19:21 UTC · v1
published

Abstract

Parallel batch-dynamic trees are a fundamental building block in recent theoretical and practical advances in dynamic graph algorithms. However, all existing parallel batch-dynamic tree data structures, including Euler tour trees, UFO trees, topology trees, and rake-compress trees, are all significantly outperformed in the sequential setting by link-cut trees, which have been the sequential state-of-the-art for over 40 years. Despite their excellent performance in the sequential setting, designing efficient batch-parallel link-cut trees has remained a major open problem. In this paper, we close this gap by introducing MOJOS, a unified framework for theoretically- and practically-efficient parallel batch-dynamic trees. We exploit the fact that both Euler tour trees and link-cut trees rely on a common dynamic sequence abstraction that supports splitting and joining. We introduce a new batch-dynamic sequence built using treaps that achieves optimal work and depth, and outperforms existing parallel skip list and treap implementations for batch updates, queries, and memory usage. With MOJOS, we develop a new batch-parallel Euler tour tree algorithm that outperforms prior batch-dynamic tree implementations supporting subtree queries. Unlike prior batch-parallel Euler tour trees which rely on skip list's ability to represent cyclic sequences, MOJOS allows any batch-dynamic sequence data structure to be used as a drop-in replacement. Finally, we develop the first theoretically-efficient batch-parallel link-cut tree, which is also the first batch-dynamic data structure supporting path queries to achieve $O(\log n)$ depth for batch updates in the binary-forking model. Our link-cut tree implementation outperforms all known parallel batch-dynamic tree data structures supporting path queries.

Closest papers

Nearest neighbours by meaning across the whole index.

PaB-tree: Parallel B+-tree with Batch Updates

cs.DS cs.DBcs.DC ·2026-11-06 ·Ziyang Men, Yan Gu, Yihan Sun, Helen Xu
similarity 0.69
Abstract

The B+-tree is a ubiquitous search index due to its provable I/O (cache) efficiency. Yet unlike batch-parallel binary search trees, which are both fast and widely available, no batch-parallel B+-tree offers strong empirical performance together with provable, near-optimal I/O efficiency. Prior B+-tree-based designs are either parallel but purely theoretical, with no high-performance implementation, or concurrent indexes such as the Bw-tree and Masstree, which run an order of magnitude slower on batch updates. We present PaB-tree, the first batch-parallel B+-tree with guarantees on work, span, and cache complexity alongside strong practical performance. Its core challenge -- repairing node overflows and underflows without eroding the tree's I/O advantage -- is resolved by one simple invariant: a repair never touches a sibling and is always pushed to the parent. A Promote primitive reduces batch construction and insertion to repeated node promotion, and batch deletion to reinsertion. Batch updates take O(B log_B n) work and O(log_B n) I/O per element, matching the sequential B+-tree bounds, with polylogarithmic span. Against highly-optimized parallel and concurrent baselines, PaB-tree delivers the best overall performance trade-off: up to 1.4x faster on construction and up to 3.6x faster on batch insertion than the strongest baseline, while preserving B+-tree query efficiency.

0.15

GrAND: GPU-based Dynamic Graph Indexes for Approximate Nearest Neighbour Search

cs.DC cs.DB ·2026-08-21 ·Karthik Venkatasubba, Shivendra Deshpande, Shivram S, Jyothi Vedurada
similarity 0.61
Abstract

Modern Approximate Nearest Neighbour Search (ANNS) applications operate over continuously evolving vector collections and require graph indexes that sustain high-throughput searches while incorporating insertions and deletions with high recall. However, most GPU graph indexes are static or provide limited update support. Updates require neighbour discovery, reverse-edge creation, pruning, and deletion-induced graph repair; executing these operations concurrently introduces redundant distance computations and conflicting accesses to shared adjacency lists. Background-rebuild-based deletion further incurs substantial computation, additional memory consumption, and interference with foreground queries. We present GrAND (GPU-based Dynamic Graph Indexes for Approximate Nearest Neighbour Search), a GPU-native collection of dynamic-update algorithms for two popular graph indexes, Vamana and CAGRA. GrAND consolidates graph repair across a batch, eliminating redundant pruning computations, and employs a lock-free find-and-replace strategy for parallel adjacency-list updates. For reliable in-place deletion, GrAND constructs an on-demand reverse graph on the GPU, accurately identifying incoming edges without permanently duplicating the index. We evaluate GrAND on seven real-world datasets across five streaming workloads, comparing it against SVFusion and FreshDiskANN-GPU (our GPU adaptation of FreshDiskANN). GrAND improves overall workload throughput by 2.2x-8.7x and 6.5x-25.4x, respectively, while maintaining high search throughput and recall over sustained updates.

0.14

Parallel Algorithms for Higher-Order Temporal H-Index Computation

cs.DS ·2026-11-06 ·Lutz Oettershagen
similarity 0.60
Abstract

Ranking vertices by importance is a fundamental task in network mining. The H-index provides such a ranking by asking how many neighbors are themselves well connected; its higher-order temporal variant recursively applies this idea over time-respecting neighborhoods in temporal networks. Scaling its computation is difficult because the existing streaming algorithm intertwines temporal processing with H-index order, incurring maximum-degree-dependent costs and exposing little parallelism. We show that the temporal dependencies can instead be resolved once: each edge's time-dependent lookup is mapped to a fixed successor position, turning every H-index order into a parallel gather followed by a segmented suffix-H computation. Based on this reformulation, we derive a linear-work suffix-H algorithm, a parallel multicore CPU implementation, degree-adaptive GPU kernels, and a multi-GPU scheme that exchanges only distinct remotely queried states. Experiments on synthetic and real-world temporal graphs show that the reformulation alone is substantially faster than the streaming algorithm, and that GPUs add order-round speedups exceeding an order of magnitude over our multicore implementation on synthetic inputs with up to one billion temporal edges. On a 636M-edge Reddit network, eight GPUs compute 128 orders in 3.6 s of round time, whereas the streaming algorithm does not finish within an hour even at order 8.

0.30

Distributed Datalog at Scale with Worst-Case Optimal Joins

cs.DB ·2026-10-08 ·Kunting Qi, Yihao Sun, Kristopher Micinski, Thomas Gilray, Sidharth Kumar
similarity 0.58
Abstract

Scalable logic-based analytics are frequently dominated by cyclic multi-way joins over large relational datasets. State-of-the-art systems evaluate these joins through chains of binary joins, incurring substantial overhead from large intermediate relations that must be materialized and redistributed across the network. Worst-Case Optimal Joins (WCOJ) offer a provably superior alternative by evaluating joins one variable at a time, eliminating intermediate blowup. However, existing WCOJ implementations target either shared-memory machines or one-shot, non-recursive queries on dataflow and big-data platforms, leaving WCOJ-based recursive evaluation on distributed-memory HPC systems an open problem. We present, to our knowledge, the first distributed-memory Datalog engine that uses WCOJ for recursive fixpoint evaluation. Our design builds on HyperCube partitioning, which hash-partitions relations along variable-aligned grid dimensions and replicates them along missing dimensions, and extends it from a single shuffle to iterative execution: in every semi-naive iteration, only newly derived tuples are repartitioned and re-replicated, using MPI all-to-all collectives and per-dimension all-gathers over subcube communicators, while the same layout is reused across iterations so that all WCOJ intersections remain rank-local. We evaluate our system on the Aurora and Polaris leadership-class supercomputers at Argonne National Laboratory using graph mining and program analysis benchmarks, demonstrating strong scalability, outperforming state-of-the-art Datalog engines on most benchmarks, and characterizing how partition dimensionality trades runtime against memory feasibility.

0.31

CQELS-TrieGS Report: Snapshot-Consistent Constant-Delay Enumeration for Streaming Graph Queries

cs.DB ·2026-08-16 ·Danh Le-Phuoc
similarity 0.58
Abstract

Continuous graph-query engines must process edge updates while making current query results available to concurrent consumers. Existing dynamic constant-delay enumeration methods provide strong per-answer delay guarantees, but are commonly formulated as a maintenance-then-enumeration process. Conversely, multicore graph-stream engines emphasize update throughput and match discovery without a query-level snapshot guarantee for concurrent full-result enumeration. We present TrieGS, a shared-memory engine that maintains a query-specific CDE state over a streaming graph. Logically, TrieGS uses the classical free-connex witness-subtree enumerator; dynamically, it maintains multiplicity payloads using exact signed deltas; physically, RDF terms are dictionary-encoded and the required access structures are realized as versioned LFNT relations (Lock-Free Nested Trie). The new systems problem is not relation-level snapshotting alone: an enumeration job must observe one consistent state across all interdependent base relations, projections, views, and indexes. TrieGS therefore publishes an atomic root vector only after an update epoch has fully propagated. Enumeration threads pin one published root vector and traverse immutable versions while later updates continue.

0.04

Join Order Enumeration and Optimization for Double-Star and Helix Join Graphs

cs.DB ·2026-10-10 ·Ansh Taneja, Ryan Marcus
similarity 0.58
Abstract

Join ordering is classic problem in data management. We highlight the connection between structure-specific dynamic programming algorithms and the mathematical enumeration of the possible merge trees from a join graph. We present two case studies, on double-star and helix shaped join graphs, that we believe carry useful techniques for a toolbox of building structure-specific join optimizers. For the double-star case, we show how the enumeration leads directly to a significantly improved search algorithm, whereas for the helix case we show that a algorithm derived from the mathematical enumeration is not necessarily better than structure-agnostic algorithms.

0.24

Space-Efficient Hierholzer for Undirected Graphs

cs.DS cs.DM ·2026-08-19 ·Elena Grigorescu, Ziad Ismaili Alaoui, Tamio-Vesa Nakajima, Shayan Shirazi Mofrad, Sebastian Wild
similarity 0.57
Abstract

We present a simple linear-time algorithm that outputs an Eulerian tour of an undirected multigraph with $n$ vertices and $m$ edges, if one exists, in $O(m)$ time and using $O(n)$ words of working memory. The input is given as read-only adjacency lists, and the output is written to an append-only stream in traversal order. Our algorithm first finds a sparse spanning circuit (a skeleton), then traverses the circuit step-by-step, repeatedly outputting further circuits rooted at the current vertex. This solves a problem left open by Ismaili Alaoui, Plump, and Wild (SOSA 2026): their space-efficient variant of Hierholzer's algorithm handles general directed multigraphs, but it is unclear how to generalize it to general undirected multigraphs. Our result completes the picture in the read-only model for space-efficient output of Eulerian tours.

0.47

Arborist: Algorithm-Hardware Co-Design for Fast and Efficient Motion Planning

cs.AR ·2026-09-11 ·Yaotian Liu, Lingyi Huang, Zishen Wan, Bo Yuan, Cheng Tan +2 more
top venue similarity 0.56
Abstract · Accepted to the 59th IEEE/ACM International Symposium on Microarchitecture (MICRO 2026)

Real-time motion planning must execute under strict latency and energy constraints on resource-limited platforms. This paper presents Arborist, an algorithm-hardware co-design framework that accelerates Fast Marching Tree (FMT*) motion planning through synergistic algorithmic and architectural innovations. At the algorithm level, Arborist introduces safe multi-tree expansion with global competition management and cross-tree optimality checking to expose inter-tree parallelism while preserving path quality. At the architecture level, our proposed ArboristAccel integrates an Arborist Toolbox that provides hardware support for inter-tree parallelism; a KD-tree Subsystem for high-throughput near-neighbor search; a Parallel-Query Retrieval module that balances multi-bank memory accesses to alleviate system-level memory bandwidth bottlenecks; and a Look-ahead Planning Engine that exploits intra-tree parallelism while preserving path quality by in-order commit. Synthesized in 16 nm and operating at 500 MHz, ArboristAccel sustains more than 1000x speedup over CPU across all workloads, peaking at 2800x on 2D-Maze, and delivers geomean improvements of 8.8x speedup, 4.6x area efficiency, and 3.1x power efficiency over an FMT* ASIC baseline. The corresponding factors relative to a state-of-the-art RRT*-based MOPED accelerator are 8.6x, 1.0x, and 1.8x, at comparable path cost. To the best of our knowledge, ArboristAccel is the first dedicated hardware accelerator for FMT*-based motion planning.

0.48