Search

We present TeraPart, a memory-efficient multilevel graph partitioning method that is designed to scale to extremely large graphs. In balanced graph partitioning, the goal is to divide the vertices into $k$ blocks with balanced size while cutting as few edges as possible. Due to its NP-hard nature, heuristics are prevalent in this field, with the multilevel framework as the state-of-the-art method. Recent work has seen tremendous progress in speeding up partitioning algorithms through parallelism. The current obstacle in scaling to larger graphs is the high memory usage due to auxiliary data structures and storing the graph itself in memory. In this paper, we present and study several optimizations to significantly reduce their memory footprint. We devise parallel label propagation clustering and graph contraction algorithms that use $O(n)$ auxiliary space instead of $O(np)$, where $p$ is the number of processors. Moreover, we employ an existing compressed graph representation that enables iterating over a neighborhood by on-the-fly decoding at speeds close to the uncompressed graph. Combining these optimizations yields up to a 16-fold reduction in peak memory, while retaining the same solution quality and similar speed. This configuration can partition a graph with one trillion edges in under 8 minutes \emph{on a single machine} using around 900\,GiB of RAM. This is the first work to employ the multilevel framework at this scale, which is vital to achieving low edge cuts. Moreover, our distributed memory implementation handles graphs of up to 16 trillion edges on 128 machines with 256\,GiB each in just under 10 minutes. Finally, we present a version of shared-memory parallel FM local search that uses $O(m)$ space instead of $O(nk)$, reducing peak memory by factor 5.8 on medium-sized graphs without affecting running time.

In bi-criteria optimization problems, the goal is typically to compute the set of Pareto-optimal solutions. Many algorithms for these types of problems rely on efficient merging or combining of partial solutions and filtering of dominated solutions in the resulting sets. In this article, we consider the task of computing the Pareto sum of two given Pareto sets $A, B$ of size $n$. The Pareto sum $C$ contains all non-dominated points of the Minkowski sum $M = \{a+b|a \in A, b\in B\}$. Since the Minkowski sum has a size of $n^2$, but the Pareto sum $C$ can be much smaller, the goal is to compute $C$ without having to compute and store all of $M$. We present several new algorithms for efficient Pareto sum computation, including an output-sensitive successive algorithm with a running time of $O(n \log n + nk)$ and a space consumption of $O(n+k)$ for $k=|C|$. If the elements of $C$ are streamed, the space consumption reduces to $O(n)$. For output sizes $k \geq 2n$, we prove a conditional lower bound for Pareto sum computation, which excludes running times in $O(n^{2-\delta})$ for $\delta > 0$ unless the (min,+)-convolution hardness conjecture fails. The successive algorithm matches this lower bound for $k \in \Theta(n)$. However, for $k \in \Theta(n^2)$, the successive algorithm exhibits a cubic running time. But we also present an algorithm with an output-sensitive space consumption and a running time of $O(n^2 \log n)$, which matches the lower bound up to a logarithmic factor even for large $k$. Furthermore, we describe suitable engineering techniques to improve the practical running times of our algorithms. Finally, we provide an extensive comparative experimental study on generated and real-world data. As a showcase application, we consider preprocessing-based bi-criteria route planning in road networks.

Partitioning a graph into blocks of roughly equal weight while cutting only few edges is a fundamental problem in computer science with numerous practical applications. While shared-memory parallel partitioners have recently matured to achieve the same quality as widely used sequential partitioners, there is still a pronounced quality gap between distributed partitioners and their sequential counterparts. In this work, we shrink this gap considerably by describing the engineering of an unconstrained local search algorithm suitable for distributed partitioners. We integrate the proposed algorithm in a distributed multilevel partitioner. Our extensive experiments show that the resulting algorithm scales to thousands of PEs while computing cuts that are, on average, only 3.5% larger than those of a state-of-the-art high-quality shared-memory partitioner. Compared to previous distributed partitioners, we obtain on average 6.8% smaller cuts than the best-performing competitor while being more than 9 times faster.

The NP-hard scheduling problem P||C_max encompasses a set of tasks with known execution time which must be mapped to a set of identical machines such that the overall completion time is minimized. In this work, we improve existing techniques for optimal P||C_max scheduling with a combination of new theoretical insights and careful practical engineering. Most importantly, we derive techniques to prune vast portions of the search space of branch-and-bound (BnB) approaches. We also propose improved upper and lower bounding techniques which can be combined with any approach to P||C_max. Moreover, we present new benchmarks for P||C_max, based on diverse application data, which can shed light on aspects which prior synthetic instances fail to capture. In an extensive evaluation, we observe that our pruning techniques reduce the number of explored nodes by 90$\times$ and running times by 12$\times$. Compared to a state-of-the-art ILP-based approach, our approach is preferable for short running time limits and for instances with large makespans.

A minimal perfect hash function (MPHF) maps a set of n keys to {1, ..., n} without collisions. Such functions find widespread application e.g. in bioinformatics and databases. In this paper we revisit PTHash - a construction technique particularly designed for fast queries. PTHash distributes the input keys into small buckets and, for each bucket, it searches for a hash function seed that places its keys in the output domain without collisions. The collection of all seeds is then stored in a compressed way. Since the first buckets are easier to place, buckets are considered in non-increasing order of size. Additionally, PTHash heuristically produces an imbalanced distribution of bucket sizes by distributing 60% of the keys into 30% of the buckets. Our main contribution is to characterize, up to lower order terms, an optimal distribution of expected bucket sizes. We arrive at a simple, closed form solution which improves construction throughput for space efficient configurations in practice. Our second contribution is a novel encoding scheme for the seeds. We split the keys into partitions. Within each partition, we run the bucket distribution and search step. We then store the seeds in an interleaved way by consecutively placing the seeds for the i-th buckets from all partitions. The seeds for the i-th bucket of each partition follow the same statistical distribution. This allows us to tune a compressor for each bucket. Hence, we call our technique PHOBIC - Perfect Hashing with Optimized Bucket sizes and Interleaved Coding. Compared to PTHash, PHOBIC is 0.17 bits/key more space efficient for same query time and construction throughput. We also contribute a GPU implementation to further accelerate MPHF construction. For a configuration with fast queries, PHOBIC-GPU can construct a perfect hash function at 2.17 bits/key in 28 ns per key, which can be queried in 37 ns on the CPU.

String sorting is an important part of tasks such as building index data structures. Unfortunately, current string sorting algorithms do not scale to massively parallel distributed-memory machines since they either have latency (at least) proportional to the number of processors $p$ or communicate the data a large number of times (at least logarithmic). We present practical and efficient algorithms for distributed-memory string sorting that scale to large $p$. Similar to state-of-the-art sorters for atomic objects, the algorithms have latency of about $p^{1/k}$ when allowing the data to be communicated $k$ times. Experiments indicate good scaling behavior on a wide range of inputs on up to 49152 cores. Overall, we achieve speedups of up to 5 over the current state-of-the-art distributed string sorting algorithms.

The Message-Passing Interface (MPI) and C++ form the backbone of high-performance computing, but MPI only provides C and Fortran bindings. While this offers great language interoperability, high-level programming languages like C++ make software development quicker and less error-prone. We propose novel C++ language bindings that cover all abstraction levels from low-level MPI calls to convenient STL-style bindings, where most parameters are inferred from a small subset of parameters, by bringing named parameters to C++. This enables rapid prototyping and fine-tuning runtime behavior and memory management. A flexible type system and additional safety guarantees help to prevent programming errors. By exploiting C++'s template metaprogramming capabilities, this has (near) zero overhead, as only required code paths are generated at compile time. We demonstrate that our library is a strong foundation for a future distributed standard library using multiple application benchmarks, ranging from text-book sorting algorithms to phylogenetic interference., Comment: To appear at SC24, November 17-22, 2024, Atlanta, Georgia, USA

We introduce an improved algorithm for the dynamic taxi sharing problem, i.e. a dispatcher that schedules a fleet of shared taxis as it is used by services like UberXShare and Lyft Shared. We speed up the basic online algorithm that looks for all possible insertions of a new customer into a set of existing routes, we generalize the objective function, and we efficiently support a large number of possible pick-up and drop-off locations. This lays an algorithmic foundation for taxi sharing systems with higher vehicle occupancy - enabling greatly reduced cost and ecological impact at comparable service quality. We find that our algorithm computes assignments between vehicles and riders several times faster than a previous state-of-the-art approach. Further, we observe that allowing meeting points for vehicles and riders can reduce the operating cost of vehicle fleets by up to 15% while also reducing rider wait and trip times., Comment: 26 pages, 7 figures, 4 tables. To be presented at ALENEX'24. arXiv admin note: substantial text overlap with arXiv:2305.05417

A minimal perfect hash function (MPHF) maps a set S of n keys to the first n integers without collisions. There is a lower bound of n*log(e)=1.44n bits needed to represent an MPHF. This can be reached by a brute-force algorithm that tries e^n hash function seeds in expectation and stores the first seed leading to an MPHF. The most space-efficient previous algorithms for constructing MPHFs all use such a brute-force approach as a basic building block. In this paper, we introduce ShockHash - Small, heavily overloaded cuckoo hash tables for minimal perfect hashing. ShockHash uses two hash functions h_0 and h_1, hoping for the existence of a function f : S->{0, 1} such that x -> h_{f(x)}(x) is an MPHF on S. It then uses a 1-bit retrieval data structure to store f using n + o(n) bits. In graph terminology, ShockHash generates n-edge random graphs until stumbling on a pseudoforest - where each component contains as many edges as nodes. Using cuckoo hashing, ShockHash then derives an MPHF from the pseudoforest in linear time. We show that ShockHash needs to try only about (e/2)^n=1.359^n seeds in expectation. This reduces the space for storing the seed by roughly n bits (maintaining the asymptotically optimal space consumption) and speeds up construction by almost a factor of 2^n compared to brute-force. Bipartite ShockHash reduces the expected construction time again to 1.166^n by maintaining a pool of candidate hash functions and checking all possible pairs. ShockHash as a building block within the RecSplit framework can be constructed up to 3 orders of magnitude faster than competing approaches. It can build an MPHF for 10 million keys with 1.489 bits per key in about half an hour. When instead using ShockHash after an efficient k-perfect hash function, it achieves space usage similar to the best competitors, while being significantly faster to construct and query., Comment: Expands arXiv:2310.14959v1 (Bipartite ShockHash) and contains content from arXiv:2308.09561v2 (Plain ShockHash). An earlier title of this article was "Bipartite ShockHash: Pruning ShockHash Search for Efficient Perfect Hashing". Now the article is a preprint of a journal version combining ShockHash and bipartite ShockHash

A minimal perfect hash function (MPHF) maps a set $S$ of $n$ keys to the first $n$ integers without collisions. There is a lower bound of $n\log_2e-O(\log n)$ bits of space needed to represent an MPHF. A matching upper bound is obtained using the brute-force algorithm that tries random hash functions until stumbling on an MPHF and stores that function's seed. In expectation, $e^n\textrm{poly}(n)$ seeds need to be tested. The most space-efficient previous algorithms for constructing MPHFs all use such a brute-force approach as a basic building block. In this paper, we introduce ShockHash - Small, heavily overloaded cuckoo hash tables. ShockHash uses two hash functions $h_0$ and $h_1$, hoping for the existence of a function $f : S \rightarrow \{0,1\}$ such that $x \mapsto h_{f(x)}(x)$ is an MPHF on $S$. In graph terminology, ShockHash generates $n$-edge random graphs until stumbling on a pseudoforest - a graph where each component contains as many edges as nodes. Using cuckoo hashing, ShockHash then derives an MPHF from the pseudoforest in linear time. It uses a 1-bit retrieval data structure to store $f$ using $n + o(n)$ bits. By carefully analyzing the probability that a random graph is a pseudoforest, we show that ShockHash needs to try only $(e/2)^n\textrm{poly}(n)$ hash function seeds in expectation, reducing the space for storing the seed by roughly $n$ bits. This makes ShockHash almost a factor $2^n$ faster than brute-force, while maintaining the asymptotically optimal space consumption. An implementation within the RecSplit framework yields the currently most space efficient MPHFs, i.e., competing approaches need about two orders of magnitude more work to achieve the same space.

One established metric to classify the significance of a mountain peak is its isolation. It specifies the distance between a peak and the closest point of higher elevation. Peaks with high isolation dominate their surroundings and provide a nice view from the top. With the availability of worldwide Digital Elevation Models (DEMs), the isolation of all mountain peaks can be computed automatically. Previous algorithms run in worst case time that is quadratic in the input size. We present a novel sweep-plane algorithm that runs in time $\mathcal{O}(n\log n+p T_{NN})$ where $n$ is the input size, $p$ the number of considered peaks and $T_{NN}$ the time for a 2D nearest-neighbor query in an appropriate geometric search tree. We refine this to a two-level approach that has high locality and good parallel scalability. Our implementation reduces the time for calculating the isolation of every peak on earth from hours to minutes while improving precision., Comment: Submitted to European Symposium on Algorithms ESA'23

We introduce KaRRi, an improved algorithm for scheduling a fleet of shared vehicles as it is used by services like UberXShare and Lyft Shared. We speed up the basic online algorithm that looks for all possible insertions of a new customer into a set of existing routes, we generalize the objective function, and efficiently support a large number of possible pick-up and drop-off locations. This lays an algorithmic foundation for ridesharing systems with higher vehicle occupancy -- enabling greatly reduced cost and ecological impact at comparable service quality. We find that our algorithm computes assignments between vehicles and riders several times faster than a previous state-of-the-art approach. Further, we observe that allowing meeting points for vehicles and riders can reduce the operating cost of vehicle fleets by up to $15\%$ while also reducing passenger wait and trip times., Comment: 29 pages, 6 figures, 5 tables Submitted to the European Symposium on Algorithms (ESA) 2023

A Monotone Minimal Perfect Hash Function (MMPHF) constructed on a set S of keys is a function that maps each key in S to its rank. On keys not in S, the function returns an arbitrary value. Applications range from databases, search engines, data encryption, to pattern-matching algorithms. In this paper, we describe LeMonHash, a new technique for constructing MMPHFs for integers. The core idea of LeMonHash is surprisingly simple and effective: we learn a monotone mapping from keys to their rank via an error-bounded piecewise linear model (the PGM-index), and then we solve the collisions that might arise among keys mapping to the same rank estimate by associating small integers with them in a retrieval data structure (BuRR). On synthetic random datasets, LeMonHash needs 34% less space than the next larger competitor, while achieving about 16 times faster queries. On real-world datasets, the space usage is very close to or much better than the best competitors, while achieving up to 19 times faster queries than the next larger competitor. As far as the construction of LeMonHash is concerned, we get an improvement by a factor of up to 2, compared to the competitor with the next best space usage. We also investigate the case of keys being variable-length strings, introducing the so-called LeMonHash-VL: it needs space within 13% of the best competitors while achieving up to 3 times faster queries than the next larger competitor.

We present {\bf Sli}ding Blo{\bf ck} Hashing (Slick), a simple hash table data structure that combines high performance with very good space efficiency. This preliminary report outlines avenues for analysis and implementation that we intend to pursue.

Balanced hypergraph partitioning is an NP-hard problem with many applications, e.g., optimizing communication in distributed data placement problems. The goal is to place all nodes across $k$ different blocks of bounded size, such that hyperedges span as few parts as possible. This problem is well-studied in sequential and distributed settings, but not in shared-memory. We close this gap by devising efficient and scalable shared-memory algorithms for all components employed in the best sequential solvers without compromises with regards to solution quality. This work presents the scalable and high-quality hypergraph partitioning framework Mt-KaHyPar. Its most important components are parallel improvement algorithms based on the FM algorithm and maximum flows, as well as a parallel clustering algorithm for coarsening - which are used in a multilevel scheme with $\log(n)$ levels. As additional components, we parallelize the $n$-level partitioning scheme, devise a deterministic version of our algorithm, and present optimizations for plain graphs. We evaluate our solver on more than 800 graphs and hypergraphs, and compare it with 25 different algorithms from the literature. Our fastest configuration outperforms almost all existing hypergraph partitioners with regards to both solution quality and running time. Our highest-quality configuration achieves the same solution quality as the best sequential partitioner KaHyPar, while being an order of magnitude faster with ten threads. Thus, two of our configurations occupy all fronts of the Pareto curve for hypergraph partitioning. Furthermore, our solvers exhibit good speedups, e.g., 29.6x in the geometric mean on 64 cores (deterministic), 22.3x ($\log(n)$-level), and 25.9x ($n$-level).

Yao graphs are geometric spanners that connect each point of a given point set to its nearest neighbor in each of $k$ cones drawn around it. Yao graphs were introduced to construct minimum spanning trees in $d$ dimensional spaces. Moreover, they are used for instance in topology control in wireless networks. An optimal \Onlogn time algorithm to construct Yao graphs for given point set has been proposed in the literature but -- to the best of our knowledge -- never been implemented. Instead, algorithms with a quadratic complexity are used in popular packages to construct these graphs. In this paper we present the first implementation of the optimal Yao graph algorithm. We develop and tune the data structures required to achieve the O(n log n) bound and detail algorithmic adaptions necessary to take the original algorithm from theory to practice. We propose a priority queue data structure that separates static and dynamic events and might be of independent interest for other sweepline algorithms. Additionally, we propose a new Yao graph algorithm based on a uniform grid data structure that performs well for medium-sized inputs. We evaluate our implementations on a wide variety synthetic and real-world datasets and show that our implementation outperforms current publicly available implementations by at least an order of magnitude., Comment: Submitted to 21st Symposium on Experimental Algorithms (SEA'23)

We describe the engineering of the distributed-memory multilevel graph partitioner dKaMinPar. It scales to (at least) 8192 cores while achieving partitioning quality comparable to widely used sequential and shared-memory graph partitioners. In comparison, previous distributed graph partitioners scale only in more restricted scenarios and often induce a considerable quality penalty compared to non-distributed partitioners. When partitioning into a large number of blocks, they even produce infeasible solution that violate the balancing constraint. dKaMinPar achieves its robustness by a scalable distributed implementation of the deep-multilevel scheme for graph partitioning. Crucially, this includes new algorithms for balancing during refinement and coarsening.

We develop and extensively evaluate highly scalable distributed-memory algorithms for computing minimum spanning trees (MSTs). At the heart of our solutions is a scalable variant of Boruvka's algorithm. For partitioned graphs with many local edges, we improve this with an effective form of contracting local parts of the graph during a preprocessing step. We also adapt the filtering concept of the best practical sequential algorithm to develop a massively parallel Filter-Boruvka algorithm that is very useful for graphs with poor locality and high average degree. Our experiments indicate that our algorithms scale well up to at least 65 536 cores and are up to 800 times faster than previous distributed MST algorithms., Comment: 12 pages, 6 figures

Counting triangles in a graph and incident to each vertex is a fundamental and frequently considered task of graph analysis. We consider how to efficiently do this for huge graphs using massively parallel distributed-memory machines. Unsurprisingly, the main issue is to reduce communication between processors. We achieve this by counting locally whenever possible and reducing the amount of information that needs to be sent in order to handle (possible) nonlocal triangles. We also achieve linear memory requirements despite superlinear communication volume by introducing a new asynchronous sparse-all-to-all operation. Furthermore, we dramatically reduce startup overheads by allowing this communication to use indirect routing. Our algorithms scale (at least) up to 32 768 cores and are up to 18 times faster than the previous state of the art., Comment: 11 pages, 8 figures, to be published in 2023 IEEE International Parallel and Distributed Processing Symposium (IPDPS), St. Petersburg, FL, USA, pp. 702-712

A minimal perfect hash function (MPHF) bijectively maps a set S of objects to the first |S| integers. It can be used as a building block in databases and data compression. RecSplit [Esposito et al., ALENEX'20] is currently the most space efficient practical minimal perfect hash function. It heavily relies on trying out hash functions in a brute force way. We introduce rotation fitting, a new technique that makes the search more efficient by drastically reducing the number of tried hash functions. Additionally, we greatly improve the construction time of RecSplit by harnessing parallelism on the level of bits, vectors, cores, and GPUs. In combination, the resulting improvements yield speedups up to 239 on an 8-core CPU and up to 5438 using a GPU. The original single-threaded RecSplit implementation needs 1.5 hours to construct an MPHF for 5 Million objects with 1.56 bits per object. On the GPU, we achieve the same space usage in just 5 seconds. Given that the speedups are larger than the increase in energy consumption, our implementation is more energy efficient than the original implementation.

A Perfect Hash Function (PHF) is a hash function that has no collisions on a given input set. PHFs can be used for space efficient storage of data in an array, or for determining a compact representative of each object in the set. In this paper, we present the PHF construction algorithm SicHash - Small Irregular Cuckoo Tables for Perfect Hashing. At its core, SicHash uses a known technique: It places objects in a cuckoo hash table and then stores the final hash function choice of each object in a retrieval data structure. We combine the idea with irregular cuckoo hashing, where each object has a different number of hash functions. Additionally, we use many small tables that we overload beyond their asymptotic maximum load factor. The most space efficient competitors often use brute force methods to determine the PHFs. SicHash provides a more direct construction algorithm that only rarely needs to recompute parts. Our implementation improves the state of the art in terms of space usage versus construction time for a wide range of configurations. At the same time, it provides very fast queries.

In recent years, significant advances have been made in the design and evaluation of balanced (hyper)graph partitioning algorithms. We survey trends of the last decade in practical algorithms for balanced (hyper)graph partitioning together with future research directions. Our work serves as an update to a previous survey on the topic. In particular, the survey extends the previous survey by also covering hypergraph partitioning and streaming algorithms, and has an additional focus on parallel algorithms.

Previous efforts on making Satisfiability (SAT) solving fit for high performance computing (HPC) have lead to super-linear speedups on particular formulae, but for most inputs cannot make efficient use of a large number of processors. Moreover, long latencies (minutes to days) of job scheduling make large-scale SAT solving on demand impractical for most applications. We address both issues with Mallob, a framework for job scheduling in the context of SAT solving which exploits malleability, i.e., the ability to add or remove processing power from a job during its computation. Mallob includes a massively parallel, distributed, and malleable SAT solving engine based on Hordesat with a more succinct and communication-efficient approach to clause sharing and numerous further improvements over its precursor. For example, Mallob on 640 cores outperforms an updated and improved configuration of Hordesat on 2560 cores. Moreover, Mallob can also solve many formulae in parallel while dynamically adapting the assigned resources, and jobs arriving in the system are usually initiated within a fraction of a second., Comment: This preprint has not undergone peer review or any post-submission improvements or corrections. The Version of Record of this contribution is published in the proceedings of the International Conference on Theory and Applications of Satisfiability Testing 2021 (pp. 518-534), Springer, Cham, and is available online at https://doi.org/10.1007/978-3-030-80223-3_35

Recent works showed that implementations of Quicksort using vector CPU instructions can outperform the non-vectorized algorithms in widespread use. However, these implementations are typically single-threaded, implemented for a particular instruction set, and restricted to a small set of key types. We lift these three restrictions: our proposed 'vqsort' algorithm integrates into the state-of-the-art parallel sorter 'ips4o', with a geometric mean speedup of 1.59. The same implementation works on seven instruction sets (including SVE and RISC-V V) across four platforms. It also supports floating-point and 16-128 bit integer keys. To the best of our knowledge, this is the fastest sort for non-tuple keys on CPUs, up to 20 times as fast as the sorting algorithms implemented in standard libraries. This paper focuses on the practical engineering aspects enabling the speed and portability, which we have not yet seen demonstrated for a Quicksort implementation. Furthermore, we introduce compact and transpose-free sorting networks for in-register sorting of small arrays, and a vector-friendly pivot sampling strategy that is robust against adversarial input., Comment: 21 pages

We introduce PaCHash, a hash table that stores its objects contiguously in an array without intervening space, even if the objects have variable size. In particular, each object can be compressed using standard compression techniques. A small search data structure allows locating the objects in constant expected time. PaCHash is most naturally described as a static external hash table where it needs a constant number of bits of internal memory per block of external memory. Here, in some sense, PaCHash beats a lower bound on the space consumption of k-perfect hashing. An implementation for fast SSDs needs about 5 bits of internal memory per block of external memory, requires only one disk access (of variable length) per search operation, and has small internal search overhead compared to the disk access cost. Our experiments show that it has lower space consumption than all previous approaches even when considering objects of identical size.

Fault-tolerant distributed applications require mechanisms to recover data lost via a process failure. On modern cluster systems it is typically impractical to request replacement resources after such a failure. Therefore, applications have to continue working with the remaining resources. This requires redistributing the workload and that the non-failed processes reload data. We present an algorithmic framework and its C++ library implementation ReStore for MPI programs that enables recovery of data after process failures. By storing all required data in memory via an appropriate data distribution and replication, recovery is substantially faster than with standard checkpointing schemes that rely on a parallel file system. As the application developer can specify which data to load, we also support shrinking recovery instead of recovery using spare compute nodes. We evaluate ReStore in both controlled, isolated environments and real applications. Our experiments show loading times of lost input data in the range of milliseconds on up to 24 576 processors and a substantial speedup of the recovery time for the fault-tolerant version of a widely used bioinformatics application.

We present a shared-memory parallelization of flow-based refinement, which is considered the most powerful iterative improvement technique for hypergraph partitioning at the moment. Flow-based refinement works on bipartitions, so current sequential partitioners schedule it on different block pairs to improve $k$-way partitions. We investigate two different sources of parallelism: a parallel scheduling scheme and a parallel maximum flow algorithm based on the well-known push-relabel algorithm. In addition to thoroughly engineered implementations, we propose several optimizations that substantially accelerate the algorithm in practice, enabling the use on extremely large hypergraphs (up to 1 billion pins). We integrate our approach in the state-of-the-art parallel multilevel framework Mt-KaHyPar and conduct extensive experiments on a benchmark set of more than 500 real-world hypergraphs, to show that the partition quality of our code is on par with the highest quality sequential code (KaHyPar), while being an order of magnitude faster with 10 threads.

A retrieval data structure for a static function $f:S\rightarrow \{0,1\}^r$ supports queries that return $f(x)$ for any $x \in S$. Retrieval data structures can be used to implement a static approximate membership query data structure (AMQ), i.e., a Bloom filter alternative, with false positive rate $2^{-r}$. The information-theoretic lower bound for both tasks is $r|S|$ bits. While succinct theoretical constructions using $(1+o(1))r|S|$ bits were known, these could not achieve very small overheads in practice because they have an unfavorable space--time tradeoff hidden in the asymptotic costs or because small overheads would only be reached for physically impossible input sizes. With bumped ribbon retrieval (BuRR), we present the first practical succinct retrieval data structure. In an extensive experimental evaluation BuRR achieves space overheads well below 1\,\% while being faster than most previously used retrieval data structures (typically with space overheads at least an order of magnitude larger) and faster than classical Bloom filters (with space overhead $\geq 44\,\%$). This efficiency, including favorable constants, stems from a combination of simplicity, word parallelism, and high locality. We additionally describe homogeneous ribbon filter AMQs, which are even simpler and faster at the price of slightly larger space overhead.

Priority queues with parallel access are an attractive data structure for applications like prioritized online scheduling, discrete event simulation, or greedy algorithms. However, a classical priority queue constitutes a severe bottleneck in this context, leading to very small throughput. Hence, there has been significant interest in concurrent priority queues with relaxed semantics. We investigate the complementary quality criteria rank error (how close are deleted elements to the global minimum) and delay (for each element x, how many elements with lower priority are deleted before x). In this paper, we introduce MultiQueues as a natural approach to relaxed priority queues based on multiple sequential priority queues. Their naturally high theoretical scalability is further enhanced by using three orthogonal ways of batching operations on the sequential queues. Experiments indicate that MultiQueues present a very good performance-quality tradeoff and considerably outperform competing approaches in at least one of these aspects. We employ a seemingly paradoxical technique of "wait-free locking" that might be of more general interest to convert sequential data structures to relaxed concurrent data structures.

An alias table is a data structure that allows for efficiently drawing weighted random samples in constant time and can be constructed in linear time. The PSA algorithm by H\"ubschle-Schneider and Sanders is able to construct alias tables in parallel on the CPU. In this report, we transfer the PSA algorithm to the GPU. Our construction algorithm achieves a speedup of 17 on a consumer GPU in comparison to the PSA method on a 16-core high-end desktop CPU. For sampling, we achieve an up to 24 times higher throughput. Both operations also require several times less energy than on the CPU. Adaptations helping to achieve this include changing memory access patterns to do coalesced access. Where this is not possible, we first copy data to the faster shared memory using coalesced access. We also enhance a generalization of binary search enabling to search for a range of items in parallel. Besides naive sampling, we also give improved batched sampling algorithms.

This paper considers the balanced hypergraph partitioning problem, which asks for partitioning the vertices into $k$ disjoint blocks of bounded size while minimizing an objective function over the hyperedges. Here, we consider the most commonly used connectivity metric. We describe our open source hypergraph partitioner KaHyPar which is based on the successful multi-level approach -- driving it to the extreme of one level for (almost) every vertex. Using carefully designed data structures and dynamic update techniques, this approach offers a very good time-quality tradeoff. We present two preprocessing techniques -- pin sparsification using locality sensitive hashing and community detection based on the Louvain algorithm. The community structure is used to guide the coarsening process that incrementally contracts vertices. Portfolio-based partitioning of the contracted hypergraph already achieves good initial solutions. While reversing the contractions, a combination of highly-localized direct $k$-way local search and flow-based techniques that take a more global view, refine the partition to achieve high quality. Optionally, a memetic algorithm evolves a pool of solution candidates to obtain even higher quality. We evaluate KaHyPar on a large set of instances from a wide range of application domains. With respect to quality, KaHyPar outperforms all previously considered systems that can handle large hypergraphs such as hMETIS, PaToH, Mondriaan, or Zoltan. KaHyPar is also faster than most of these systems except for PaToH which represents a different speed-quality tradeoff. The results even extend to the special case of graph partitioning, where specialized systems such as KaHIP should have an advantage.

Partitioning a graph into blocks of "roughly equal" weight while cutting only few edges is a fundamental problem in computer science with a wide range of applications. In particular, the problem is a building block in applications that require parallel processing. While the amount of available cores in parallel architectures has significantly increased in recent years, state-of-the-art graph partitioning algorithms do not work well if the input needs to be partitioned into a large number of blocks. Often currently available algorithms compute highly imbalanced solutions, solutions of low quality, or have excessive running time for this case. This is because most high-quality general-purpose graph partitioners are multilevel algorithms which perform graph coarsening to build a hierarchy of graphs, initial partitioning to compute an initial solution, and local improvement to improve the solution throughout the hierarchy. However, for large number of blocks, the smallest graph in the hierarchy that is used for initial partitioning still has to be large. In this work, we substantially mitigate these problems by introducing deep multilevel graph partitioning and a shared-memory implementation thereof. Our scheme continues the multilevel approach deep into initial partitioning -- integrating it into a framework where recursive bipartitioning and direct k-way partitioning are combined such that they can operate with high performance and quality. Our approach is stronger, more flexible, arguably more elegant, and reduces bottlenecks for parallelization compared to other multilevel approaches. For example, for large number of blocks our algorithm is on average an order of magnitude faster than competing algorithms while computing balanced partitions with comparable solution quality. For small number of blocks, our algorithms are the fastest among competing systems with comparable quality.

We present a shared-memory algorithm to compute high-quality solutions to the balanced $k$-way hypergraph partitioning problem. This problem asks for a partition of the vertex set into $k$ disjoint blocks of bounded size that minimizes the connectivity metric (i.e., the sum of the number of different blocks connected by each hyperedge). High solution quality is achieved by parallelizing the core technique of the currently best sequential partitioner KaHyPar: the most extreme $n$-level version of the widely used multilevel paradigm, where only a single vertex is contracted on each level. This approach is made fast and scalable through intrusive algorithms and data structures that allow precise control of parallelism through atomic operations and fine-grained locking. We perform extensive experiments on more than 500 real-world hypergraphs with up to $140$ million vertices and two billion pins (sum of hyperedge sizes). We find that our algorithm computes solutions that are on par with a comparable configuration of KaHyPar while being an order of magnitude faster on average. Moreover, we show that recent non-multilevel algorithms specifically designed to partition large instances have considerable quality penalties and no clear advantage in running time.

We study the problem of servicing a set of ride requests by dispatching a set of shared vehicles, which is faced by ridesharing companies such as Uber and Lyft. Solving this problem at a large scale might be crucial in the future for effectively using large fleets of autonomous vehicles. Since finding a solution for the entire set of requests that minimizes the total driving time is NP-complete, most practical approaches process the requests one by one. Each request is inserted into any vehicle's route such that the increase in driving time is minimized. Although this variant is solvable in polynomial time, it still takes considerable time in current implementations, even when inexact filtering heuristics are used. In this work, we present a novel algorithm for finding best insertions, based on (customizable) contraction hierarchies with local buckets. Our algorithm finds provably exact solutions, is still 30 times faster than a state-of-the-art algorithm currently used in industry and academia, and scales much better. When used within iterative transport simulations, our algorithm decreases the simulation time for largescale scenarios with many requests from days to hours., Comment: Previous version augmented in several ways

Hypergraph partitioning is an important preprocessing step for optimizing data placement and minimizing communication volumes in high-performance computing applications. To cope with ever growing problem sizes, it has become increasingly important to develop fast parallel partitioning algorithms whose solution quality is competitive with existing sequential algorithms. To this end, we present Mt-KaHyPar, the first shared-memory multilevel hypergraph partitioner with parallel implementations of many techniques used by the sequential, high-quality partitioning systems: a parallel coarsening algorithm that uses parallel community detection as guidance, initial partitioning via parallel recursive bipartitioning with work-stealing, a scalable label propagation refinement algorithm, and the first fully-parallel direct $k$-way formulation of the classical FM algorithm. Experiments performed on a large benchmark set of instances from various application domains demonstrate the scalability and effectiveness of our approach. With 64 cores, we observe self-relative speedups of up to 51 and a harmonic mean speedup of 23.5. In terms of solution quality, we outperform the distributed hypergraph partitioner Zoltan on 95% of the instances while also being a factor of 2.1 faster. With just four cores,Mt-KaHyPar is also slightly faster than the fastest sequential multilevel partitioner PaToH while producing better solutions on 83% of all instances. The sequential high-quality partitioner KaHyPar still finds better solutions than our parallel approach, especially when using max-flow-based refinement. This, however, comes at the cost of considerably longer running times., Comment: Accepted for publication at ALENEX21

We present sorting algorithms that represent the fastest known techniques for a wide range of input sizes, input distributions, data types, and machines. A part of the speed advantage is due to the feature to work in-place. Previously, the in-place feature often implied performance penalties. Our main algorithmic contribution is a blockwise approach to in-place data distribution that is provably cache-efficient. We also parallelize this approach taking dynamic load balancing and memory locality into account. Our comparison-based algorithm, In-place Superscalar Samplesort (IPS$^4$o), combines this technique with branchless decision trees. By taking cases with many equal elements into account and by adapting the distribution degree dynamically, we obtain a highly robust algorithm that outperforms the best in-place parallel comparison-based competitor by almost a factor of three. IPS$^4$o also outperforms the best comparison-based competitors in the in-place or not in-place, parallel or sequential settings. IPS$^4$o even outperforms the best integer sorting algorithms in a wide range of situations. In many of the remaining cases (often involving near-uniform input distributions, small keys, or a sequential setting), our new in-place radix sorter turns out to be the best algorithm. Claims to have the, in some sense, "best" sorting algorithm can be found in many papers which cannot all be true. Therefore, we base our conclusions on extensive experiments involving a large part of the cross product of 21 state-of-the-art sorting codes, 6 data types, 10 input distributions, 4 machines, 4 memory allocation strategies, and input sizes varying over 7 orders of magnitude. This confirms the robust performance of our algorithms while revealing major performance problems in many competitors outside the concrete set of measurements reported in the associated publications.

Random graph models are frequently used as a controllable and versatile data source for experimental campaigns in various research fields. Generating such data-sets at scale is a non-trivial task as it requires design decisions typically spanning multiple areas of expertise. Challenges begin with the identification of relevant domain-specific network features, continue with the question of how to compile such features into a tractable model, and culminate in algorithmic details arising while implementing the pertaining model. In the present survey, we explore crucial aspects of random graph models with known scalable generators. We begin by briefly introducing network features considered by such models, and then discuss random graphs alongside with generation algorithms. Our focus lies on modelling techniques and algorithmic primitives that have proven successful in obtaining massive graphs. We consider concepts and graph models for various domains (such as social network, infrastructure, ecology, and numerical simulations), and discuss generators for different models of computation (including shared-memory parallelism, massive-parallel GPUs, and distributed systems).

We explain how the popular, highly abstract MapReduce model of parallel computation (MRC) can be rooted in reality by explaining how it can be simulated on realistic distributed-memory parallel machine models like BSP. We first refine the model (MRC$^+$) to include parameters for total work $w$, bottleneck work $\hat{w}$, data volume $m$, and maximum object sizes $\hat{m}$. We then show matching upper and lower bounds for executing a MapReduce calculation on the distributed-memory machine -- $\Theta(w/p+\hat{w}+\log p)$ work and $\Theta(m/p+\hat{m}+\log p)$ bottleneck communication volume using $p$ processors.

Sorting a set of items is a task that can be useful by itself or as a building block for more complex operations. That is why a lot of effort has been put into finding sorting algorithms that sort large sets as fast as possible. But the more sophisticated and complex the algorithms become, the less efficient they are for small sets of items due to large constant factors. We aim to determine if there is a faster way than insertion sort to sort small sets of items to provide a more efficient base case sorter. We looked at sorting networks, at how they can improve the speed of sorting few elements, and how to implement them in an efficient manner by using conditional moves. Since sorting networks need to be implemented explicitly for each set size, providing networks for larger sizes becomes less efficient due to increased code sizes. To also enable the sorting of slightly larger base cases, we adapted sample sort to Register Sample Sort, to break down those larger sets into sizes that can in turn be sorted by sorting networks. From our experiments we found that when sorting only small sets, the sorting networks outperform insertion sort by a factor of at least 1.76 for any array size between six and sixteen, and by a factor of 2.72 on average across all machines and array sizes. When integrating sorting networks as a base case sorter into Quicksort, we achieved far less performance improvements, which is probably due to the networks having a larger code size and cluttering the L1 instruction cache. But for x86 machines with a larger L1 instruction cache of 64 KiB or more, we obtained speedups of 12.7% when using sorting networks as a base case sorter in std::sort. In conclusion, the desired improvement in speed could only be achieved under special circumstances, but the results clearly show the potential of using conditional moves in the field of sorting algorithms., Comment: arXiv admin note: substantial text overlap with arXiv:1908.08111

There has been surprisingly little work on algorithms for sorting strings on distributed-memory parallel machines. We develop efficient algorithms for this problem based on the multi-way merging principle. These algorithms inspect only characters that are needed to determine the sorting order. Moreover, communication volume is reduced by also communicating (roughly) only those characters and by communicating repetitions of the same prefixes only once. Experiments on up to 1280 cores reveal that these algorithm are often more than five times faster than previous algorithms., Comment: Full version to appear at IPDPS 2020