Laxman Dhulipala
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Laxman Dhulipala has a total of 47 co-authors. Here are the top co-authors by number of shared publications:
- Guy E. Blelloch
Carnegie Mellon University
5 shared publications - Julian Shun
Google DeepMind MIT CSAIL MIT CSAIL
5 shared publications - Yan Gu
University of California Riverside
5 shared publications - Jakub Łącki
Google research
4 shared publications - Quinten De Man
University of Maryland College Park USA
3 shared publications - Xiaojun Dong
UC Riverside Riverside CA USA
3 shared publications
Latest Publications
- 11 October 2026
- Published by Springer Nature
- p. 444-452
Fast K-NN Graph Building with Quantized PiPNN and Filtering
- 11 October 2026
- Published by Springer Nature
- p. 444-452
Fast K-NN Graph Building with Quantized PiPNN and Filtering
- Open Access
Faster Parallel Batch-Dynamic Algorithms for Low Out-Degree Orientation
A low out-degree orientation directs each edge of an undirected graph with the goal of minimizing the maximum out-degree of a vertex. In the parallel batch-dynamic setting, one can insert or delete batches of edges, and the goal is to process the entire batch in parallel with work per edge similar to that of a single sequential update and with span (or depth) for the entire batch that is polylogarithmic. In this paper we present faster parallel batch-dynamic algorithms for maintaining a low out-degree orientation of an undirected graph. All results herein achieve polylogarithmic depth, with high probability (whp); the focus of this paper is on minimizing the work, which varies across results. Our first result is the first parallel batch-dynamic algorithm to maintain an asymptotically optimal orientation with asymptotically optimal expected work bounds, in an amortized sense, improving over the prior best work bounds of Liu et al.~[SPAA~'22] by a logarithmic factor. Our second result is a $O(c \log n)$ orientation algorithm with expected worst-case $O(\sqrt{\log n})$ work per edge update, where $c$ is a known upper-bound on the arboricity of the graph. This matches the best-known sequential worst-case $O(c \log n)$ orientation algorithm given by Berglin and Brodal ~[Algorithmica~'18], albeit in expectation. Our final result is a $O(c + \log n)$-orientation algorithm with $O(\log^2 n)$ expected worst-case work per edge update. This algorithm significantly improves upon the recent result of Ghaffari and Koo~[SPAA~'25], which maintains a $O(c)$-orientation with $O(\log^9 n)$ worst-case work per edge whp.Open AccessFaster Parallel Batch-Dynamic Algorithms for Low Out-Degree Orientation
A low out-degree orientation directs each edge of an undirected graph with the goal of minimizing the maximum out-degree of a vertex. In the parallel batch-dynamic setting, one can insert or delete batches of edges, and the goal is to process the entire batch in parallel with work per edge similar to that of a single sequential update and with span (or depth) for the entire batch that is polylogarithmic. In this paper we present faster parallel batch-dynamic algorithms for maintaining a low out-degree orientation of an undirected graph. All results herein achieve polylogarithmic depth, with high probability (whp); the focus of this paper is on minimizing the work, which varies across results. Our first result is the first parallel batch-dynamic algorithm to maintain an asymptotically optimal orientation with asymptotically optimal expected work bounds, in an amortized sense, improving over the prior best work bounds of Liu et al.~[SPAA~'22] by a logarithmic factor. Our second result is a $O(c \log n)$ orientation algorithm with expected worst-case $O(\sqrt{\log n})$ work per edge update, where $c$ is a known upper-bound on the arboricity of the graph. This matches the best-known sequential worst-case $O(c \log n)$ orientation algorithm given by Berglin and Brodal ~[Algorithmica~'18], albeit in expectation. Our final result is a $O(c + \log n)$-orientation algorithm with $O(\log^2 n)$ expected worst-case work per edge update. This algorithm significantly improves upon the recent result of Ghaffari and Koo~[SPAA~'25], which maintains a $O(c)$-orientation with $O(\log^9 n)$ worst-case work per edge whp. - Open Access
UFO Trees: Practical and Provably-Efficient Parallel Batch-Dynamic Trees
The dynamic trees problem is to maintain a tree under edge updates while supporting queries like connectivity queries or path queries. Despite the first data structure for this fundamental problem -- the link-cut tree -- being invented 40 years ago, our experiments reveal that they are still the fastest sequential data structure for the problem. However, link-cut trees cannot support parallel batch-dynamic updates and have limitations on the kinds of queries they support. In this paper, we design a new parallel batch-dynamic trees data structure called UFO trees that simultaneously supports a wide range of query functionality, supports work-efficient parallel batch-dynamic updates, and is competitive with link-cut trees when run sequentially. We prove that a key reason for the strong practical performance of both link-cut trees and UFO trees is that they can perform updates and queries in sub-logarithmic time for low-diameter trees. We perform an experimental study of our optimized C++ implementations of UFO trees with ten other dynamic tree implementations, several of which are new, in a broad benchmark of both synthetic and real-world trees of varying diameter and size. Our results show that, in both sequential and parallel settings, UFO trees are the fastest dynamic tree data structure that supports a wide range of queries. Our new implementation of UFO trees has low space usage and easily scales to billion-size inputs, making it a promising building block for implementing more complex dynamic graph algorithms in practice.
Open AccessUFO Trees: Practical and Provably-Efficient Parallel Batch-Dynamic Trees
The dynamic trees problem is to maintain a tree under edge updates while supporting queries like connectivity queries or path queries. Despite the first data structure for this fundamental problem -- the link-cut tree -- being invented 40 years ago, our experiments reveal that they are still the fastest sequential data structure for the problem. However, link-cut trees cannot support parallel batch-dynamic updates and have limitations on the kinds of queries they support. In this paper, we design a new parallel batch-dynamic trees data structure called UFO trees that simultaneously supports a wide range of query functionality, supports work-efficient parallel batch-dynamic updates, and is competitive with link-cut trees when run sequentially. We prove that a key reason for the strong practical performance of both link-cut trees and UFO trees is that they can perform updates and queries in sub-logarithmic time for low-diameter trees. We perform an experimental study of our optimized C++ implementations of UFO trees with ten other dynamic tree implementations, several of which are new, in a broad benchmark of both synthetic and real-world trees of varying diameter and size. Our results show that, in both sequential and parallel settings, UFO trees are the fastest dynamic tree data structure that supports a wide range of queries. Our new implementation of UFO trees has low space usage and easily scales to billion-size inputs, making it a promising building block for implementing more complex dynamic graph algorithms in practice.
- Open Access
- 24 September 2025
- Published by Springer Nature in The VLDB Journal
- Vol. 34 (6) , 1-25
PIM-tree: A Skew-resistant Index for Processing-in-Memory
The performance of today’s in-memory indexes is bottlenecked by the memory latency/bandwidth wall. Processing-in-memory (PIM) is an emerging approach that potentially mitigates this bottleneck by enabling low-latency memory access whose aggregate memory bandwidth scales with the number of PIM nodes. There is an inherent tension, however, between minimizing inter-node communication and achieving load balance in PIM systems, in the presence of workload skew. This paper presents PIM-tree , an ordered index for PIM systems that achieves both low communication and high load balance, regardless of the degree of skew in data/queries. Our skew-resistant index is based on a novel division of labor between the multi-core host CPU and the PIM nodes, which leverages the strengths of each. We introduce push-pull search , which dynamically decides whether to push queries to a PIM-tree node (CPU $$\rightarrow $$ PIM-node) or pull the node’s keys back to the CPU (PIM-node $$\rightarrow $$ CPU) based on workload skew. Combined with other PIM-friendly optimizations ( shadow subtrees and chunking ), PIM-tree achieves high throughput, (guaranteed) low communication, and (guaranteed) high load balance, for batches of point queries, updates, and range scans. We implement the PIM-tree structure, in addition to prior proposed PIM indexes, on the latest PIM system from UPMEM, with 32 CPU cores and 2048 PIM nodes. On workloads with 500 million keys and batches of 1 million queries, the throughput using PIM-trees is up to $$69.7\times $$ and $$59.1\times $$ higher than the two best prior PIM-based methods. As far as we know these are the first implementations of ordered indexes on real PIM systems.- 24 September 2025
- Published by Springer Nature in The VLDB Journal
- Vol. 34 (6) , 1-25
Open AccessPIM-tree: A Skew-resistant Index for Processing-in-Memory
The performance of today’s in-memory indexes is bottlenecked by the memory latency/bandwidth wall. Processing-in-memory (PIM) is an emerging approach that potentially mitigates this bottleneck by enabling low-latency memory access whose aggregate memory bandwidth scales with the number of PIM nodes. There is an inherent tension, however, between minimizing inter-node communication and achieving load balance in PIM systems, in the presence of workload skew. This paper presents PIM-tree , an ordered index for PIM systems that achieves both low communication and high load balance, regardless of the degree of skew in data/queries. Our skew-resistant index is based on a novel division of labor between the multi-core host CPU and the PIM nodes, which leverages the strengths of each. We introduce push-pull search , which dynamically decides whether to push queries to a PIM-tree node (CPU $$\rightarrow $$ PIM-node) or pull the node’s keys back to the CPU (PIM-node $$\rightarrow $$ CPU) based on workload skew. Combined with other PIM-friendly optimizations ( shadow subtrees and chunking ), PIM-tree achieves high throughput, (guaranteed) low communication, and (guaranteed) high load balance, for batches of point queries, updates, and range scans. We implement the PIM-tree structure, in addition to prior proposed PIM indexes, on the latest PIM system from UPMEM, with 32 CPU cores and 2048 PIM nodes. On workloads with 500 million keys and batches of 1 million queries, the throughput using PIM-trees is up to $$69.7\times $$ and $$59.1\times $$ higher than the two best prior PIM-based methods. As far as we know these are the first implementations of ordered indexes on real PIM systems. - Open Access
Fully-Dynamic Parallel Algorithms for Single-Linkage Clustering
Single-linkage clustering is a popular form of hierarchical agglomerative clustering (HAC) where the distance between two clusters is defined as the minimum distance between any pair of points across the two clusters. In single-linkage HAC, the output is typically the single-linkage dendrogram (SLD), which is the binary tree representing the hierarchy of clusters formed by iteratively contracting the two closest clusters. In the dynamic setting, prior work has only studied maintaining a minimum spanning forest over the data since single-linkage HAC reduces to computing the SLD on the minimum spanning forest of the data. In this paper, we study the problem of maintaining the SLD in the fully-dynamic setting. We assume the input is a dynamic forest $F$ (representing the minimum spanning forest of the data) which receives a sequence of edge insertions and edge deletions. To our knowledge, no prior work has provided algorithms to update an SLD asymptotically faster than recomputing it from scratch. All of our update algorithms are asymptotically faster than the best known static SLD computation algorithm, which takes $O(n \log h)$ time where $h$ is the height of the dendrogram ($h \leq n-1$). Furthermore, our algorithms are much faster in many cases, such as when $h$ is low. Our first set of results are an insertion algorithm in $O(h)$ time and a deletion algorithm in $O(h \log (1+n/h))$ time. Next, we describe parallel and batch-parallel versions of these algorithms which are work-efficient or nearly work-efficient and have poly-logarithmic depth. Finally, we show how to perform insertions near-optimally in $O(c \log(1+n/c))$ time, where $c$ is the number of structural changes in the dendrogram caused by the update, and give a work-efficient parallel version of this algorithm that has polylogarithmic depth.Open AccessFully-Dynamic Parallel Algorithms for Single-Linkage Clustering
Single-linkage clustering is a popular form of hierarchical agglomerative clustering (HAC) where the distance between two clusters is defined as the minimum distance between any pair of points across the two clusters. In single-linkage HAC, the output is typically the single-linkage dendrogram (SLD), which is the binary tree representing the hierarchy of clusters formed by iteratively contracting the two closest clusters. In the dynamic setting, prior work has only studied maintaining a minimum spanning forest over the data since single-linkage HAC reduces to computing the SLD on the minimum spanning forest of the data. In this paper, we study the problem of maintaining the SLD in the fully-dynamic setting. We assume the input is a dynamic forest $F$ (representing the minimum spanning forest of the data) which receives a sequence of edge insertions and edge deletions. To our knowledge, no prior work has provided algorithms to update an SLD asymptotically faster than recomputing it from scratch. All of our update algorithms are asymptotically faster than the best known static SLD computation algorithm, which takes $O(n \log h)$ time where $h$ is the height of the dendrogram ($h \leq n-1$). Furthermore, our algorithms are much faster in many cases, such as when $h$ is low. Our first set of results are an insertion algorithm in $O(h)$ time and a deletion algorithm in $O(h \log (1+n/h))$ time. Next, we describe parallel and batch-parallel versions of these algorithms which are work-efficient or nearly work-efficient and have poly-logarithmic depth. Finally, we show how to perform insertions near-optimally in $O(c \log(1+n/c))$ time, where $c$ is the number of structural changes in the dendrogram caused by the update, and give a work-efficient parallel version of this algorithm that has polylogarithmic depth. - Open Access
Towards Scalable and Practical Batch-Dynamic Connectivity
We study the problem of dynamically maintaining the connected components of an undirected graph subject to edge insertions and deletions. We give the first parallel algorithm for the problem which is work-efficient, supports batches of updates, runs in polylogarithmic depth, and uses only linear total space. The existing algorithms for the problem either use super-linear space, do not come with strong theoretical bounds, or are not parallel. On the empirical side, we provide the first implementation of the cluster forest algorithm, the first linear-space and poly-logarithmic update time algorithm for dynamic connectivity. Experimentally, we find that our algorithm uses up to 19.7x less space and is up to 6.2x faster than the level-set algorithm of HDT, arguably the most widely-implemented dynamic connectivity algorithm with strong theoretical guarantees.
Open AccessTowards Scalable and Practical Batch-Dynamic Connectivity
We study the problem of dynamically maintaining the connected components of an undirected graph subject to edge insertions and deletions. We give the first parallel algorithm for the problem which is work-efficient, supports batches of updates, runs in polylogarithmic depth, and uses only linear total space. The existing algorithms for the problem either use super-linear space, do not come with strong theoretical bounds, or are not parallel. On the empirical side, we provide the first implementation of the cluster forest algorithm, the first linear-space and poly-logarithmic update time algorithm for dynamic connectivity. Experimentally, we find that our algorithm uses up to 19.7x less space and is up to 6.2x faster than the level-set algorithm of HDT, arguably the most widely-implemented dynamic connectivity algorithm with strong theoretical guarantees.
- Open Access
The ParClusterers Benchmark Suite (PCBS): A Fine-Grained Analysis of Scalable Graph Clustering
We introduce the ParClusterers Benchmark Suite (PCBS) -- a collection of highly scalable parallel graph clustering algorithms and benchmarking tools that streamline comparing different graph clustering algorithms and implementations. The benchmark includes clustering algorithms that target a wide range of modern clustering use cases, including community detection, classification, and dense subgraph mining. The benchmark toolkit makes it easy to run and evaluate multiple instances of different clustering algorithms, which can be useful for fine-tuning the performance of clustering on a given task, and for comparing different clustering algorithms based on different metrics of interest, including clustering quality and running time. Using PCBS, we evaluate a broad collection of real-world graph clustering datasets. Somewhat surprisingly, we find that the best quality results are obtained by algorithms that not included in many popular graph clustering toolkits. The PCBS provides a standardized way to evaluate and judge the quality-performance tradeoffs of the active research area of scalable graph clustering algorithms. We believe it will help enable fair, accurate, and nuanced evaluation of graph clustering algorithms in the future.
Open AccessThe ParClusterers Benchmark Suite (PCBS): A Fine-Grained Analysis of Scalable Graph Clustering
We introduce the ParClusterers Benchmark Suite (PCBS) -- a collection of highly scalable parallel graph clustering algorithms and benchmarking tools that streamline comparing different graph clustering algorithms and implementations. The benchmark includes clustering algorithms that target a wide range of modern clustering use cases, including community detection, classification, and dense subgraph mining. The benchmark toolkit makes it easy to run and evaluate multiple instances of different clustering algorithms, which can be useful for fine-tuning the performance of clustering on a given task, and for comparing different clustering algorithms based on different metrics of interest, including clustering quality and running time. Using PCBS, we evaluate a broad collection of real-world graph clustering datasets. Somewhat surprisingly, we find that the best quality results are obtained by algorithms that not included in many popular graph clustering toolkits. The PCBS provides a standardized way to evaluate and judge the quality-performance tradeoffs of the active research area of scalable graph clustering algorithms. We believe it will help enable fair, accurate, and nuanced evaluation of graph clustering algorithms in the future.
- Open Access
BYO: A Unified Framework for Benchmarking Large-Scale Graph Containers
A fundamental building block in any graph algorithm is a graph container - a data structure used to represent the graph. Ideally, a graph container enables efficient access to the underlying graph, has low space usage, and supports updating the graph efficiently. In this paper, we conduct an extensive empirical evaluation of graph containers designed to support running algorithms on large graphs. To our knowledge, this is the first apples-to-apples comparison of graph containers rather than overall systems, which include confounding factors such as differences in algorithm implementations and infrastructure. We measure the running time of 10 highly-optimized algorithms across over 20 different containers and 10 graphs. Somewhat surprisingly, we find that the average algorithm running time does not differ much across containers, especially those that support dynamic updates. Specifically, a simple container based on an off-the-shelf B-tree is only 1.22x slower on average than a highly optimized static one. Moreover, we observe that simplifying a graph-container Application Programming Interface (API) to only a few simple functions incurs a mere 1.16x slowdown compared to a complete API. Finally, we also measure batch-insert throughput in dynamic-graph containers for a full picture of their performance. To perform the benchmarks, we introduce BYO, a unified framework that standardizes evaluations of graph-algorithm performance across different graph containers. BYO extends the Graph Based Benchmark Suite (Dhulipala et al. 18), a state-of-the-art graph algorithm benchmark, to easily plug into different dynamic graph containers and enable fair comparisons between them on a large suite of graph algorithms. While several graph algorithm benchmarks have been developed to date, to the best of our knowledge, BYO is the first system designed to benchmark graph containers
Open AccessBYO: A Unified Framework for Benchmarking Large-Scale Graph Containers
A fundamental building block in any graph algorithm is a graph container - a data structure used to represent the graph. Ideally, a graph container enables efficient access to the underlying graph, has low space usage, and supports updating the graph efficiently. In this paper, we conduct an extensive empirical evaluation of graph containers designed to support running algorithms on large graphs. To our knowledge, this is the first apples-to-apples comparison of graph containers rather than overall systems, which include confounding factors such as differences in algorithm implementations and infrastructure. We measure the running time of 10 highly-optimized algorithms across over 20 different containers and 10 graphs. Somewhat surprisingly, we find that the average algorithm running time does not differ much across containers, especially those that support dynamic updates. Specifically, a simple container based on an off-the-shelf B-tree is only 1.22x slower on average than a highly optimized static one. Moreover, we observe that simplifying a graph-container Application Programming Interface (API) to only a few simple functions incurs a mere 1.16x slowdown compared to a complete API. Finally, we also measure batch-insert throughput in dynamic-graph containers for a full picture of their performance. To perform the benchmarks, we introduce BYO, a unified framework that standardizes evaluations of graph-algorithm performance across different graph containers. BYO extends the Graph Based Benchmark Suite (Dhulipala et al. 18), a state-of-the-art graph algorithm benchmark, to easily plug into different dynamic graph containers and enable fair comparisons between them on a large suite of graph algorithms. While several graph algorithm benchmarks have been developed to date, to the best of our knowledge, BYO is the first system designed to benchmark graph containers
- Open Access
Fine-Grained Privacy Guarantees for Coverage Problems
We introduce a new notion of neighboring databases for coverage problems such as Max Cover and Set Cover under differential privacy. In contrast to the standard privacy notion for these problems, which is analogous to node-privacy in graphs, our new definition gives a more fine-grained privacy guarantee, which is analogous to edge-privacy. We illustrate several scenarios of Set Cover and Max Cover where our privacy notion is desired one for the application. Our main result is an $ε$-edge differentially private algorithm for Max Cover which obtains an $(1-1/e-η,\tilde{O}(k/ε))$-approximation with high probability. Furthermore, we show that this result is nearly tight: we give a lower bound show that an additive error of $Ω(k/ε)$ is necessary under edge-differential privacy. Via group privacy properties, this implies a new algorithm for $ε$-node differentially private Max Cover which obtains an $(1-1/e-η,\tilde{O}(fk/ε))$-approximation, where $f$ is the maximum degree of an element in the set system. When $f\ll k$, this improves over the best known algorithm for Max Cover under pure (node) differential privacy, which obtains an $(1-1/e,\tilde{O}(k^2/ε))$-approximation.Open AccessFine-Grained Privacy Guarantees for Coverage Problems
We introduce a new notion of neighboring databases for coverage problems such as Max Cover and Set Cover under differential privacy. In contrast to the standard privacy notion for these problems, which is analogous to node-privacy in graphs, our new definition gives a more fine-grained privacy guarantee, which is analogous to edge-privacy. We illustrate several scenarios of Set Cover and Max Cover where our privacy notion is desired one for the application. Our main result is an $ε$-edge differentially private algorithm for Max Cover which obtains an $(1-1/e-η,\tilde{O}(k/ε))$-approximation with high probability. Furthermore, we show that this result is nearly tight: we give a lower bound show that an additive error of $Ω(k/ε)$ is necessary under edge-differential privacy. Via group privacy properties, this implies a new algorithm for $ε$-node differentially private Max Cover which obtains an $(1-1/e-η,\tilde{O}(fk/ε))$-approximation, where $f$ is the maximum degree of an element in the set system. When $f\ll k$, this improves over the best known algorithm for Max Cover under pure (node) differential privacy, which obtains an $(1-1/e,\tilde{O}(k^2/ε))$-approximation. - Open Access
ParlayANN: Scalable and Deterministic Parallel Graph-Based Approximate Nearest Neighbor Search Algorithms
Approximate nearest-neighbor search (ANNS) algorithms are a key part of the modern deep learning stack due to enabling efficient similarity search over high-dimensional vector space representations (i.e., embeddings) of data. Among various ANNS algorithms, graph-based algorithms are known to achieve the best throughput-recall tradeoffs. Despite the large scale of modern ANNS datasets, existing parallel graph based implementations suffer from significant challenges to scale to large datasets due to heavy use of locks and other sequential bottlenecks, which 1) prevents them from efficiently scaling to a large number of processors, and 2) results in nondeterminism that is undesirable in certain applications. In this paper, we introduce ParlayANN, a library of deterministic and parallel graph-based approximate nearest neighbor search algorithms, along with a set of useful tools for developing such algorithms. In this library, we develop novel parallel implementations for four state-of-the-art graph-based ANNS algorithms that scale to billion-scale datasets. Our algorithms are deterministic and achieve high scalability across a diverse set of challenging datasets. In addition to the new algorithmic ideas, we also conduct a detailed experimental study of our new algorithms as well as two existing non-graph approaches. Our experimental results both validate the effectiveness of our new techniques, and lead to a comprehensive comparison among ANNS algorithms on large scale datasets with a list of interesting findings.
Open AccessParlayANN: Scalable and Deterministic Parallel Graph-Based Approximate Nearest Neighbor Search Algorithms
Approximate nearest-neighbor search (ANNS) algorithms are a key part of the modern deep learning stack due to enabling efficient similarity search over high-dimensional vector space representations (i.e., embeddings) of data. Among various ANNS algorithms, graph-based algorithms are known to achieve the best throughput-recall tradeoffs. Despite the large scale of modern ANNS datasets, existing parallel graph based implementations suffer from significant challenges to scale to large datasets due to heavy use of locks and other sequential bottlenecks, which 1) prevents them from efficiently scaling to a large number of processors, and 2) results in nondeterminism that is undesirable in certain applications. In this paper, we introduce ParlayANN, a library of deterministic and parallel graph-based approximate nearest neighbor search algorithms, along with a set of useful tools for developing such algorithms. In this library, we develop novel parallel implementations for four state-of-the-art graph-based ANNS algorithms that scale to billion-scale datasets. Our algorithms are deterministic and achieve high scalability across a diverse set of challenging datasets. In addition to the new algorithmic ideas, we also conduct a detailed experimental study of our new algorithms as well as two existing non-graph approaches. Our experimental results both validate the effectiveness of our new techniques, and lead to a comprehensive comparison among ANNS algorithms on large scale datasets with a list of interesting findings.