arXiv subjects

Sascha Witt

Publications and source records attributed to Sascha Witt.

7 recordsLinked to original sources

arXiv2026

T-REX: Fast and Dynamic Journey Planning for Continental-Scale Public Transit Networks

We present T-REX (Transfer-Ranked EXploration), a new algorithm for journey planning in public transit networks on the country and continental scale. Our algorithm applies the principles of multi-level overlays to Trip-Based Public Transit Routing (TB). Using a multi-level partition of the network, T-REX identifies transfers between trips that are relevant for long-distance travel in a short precomputation phase. This information is then used to prune irrelevant local transfers during a query. Like other state-of-the-art algorithms, T-REX Pareto-optimizes arrival time and the number of used trips. T-REX dramatically outperforms previous overlay-based algorithms for three key reasons: (1) a better partition, (2) reducing the search space by focusing on transfers rather than trips, and (3) a redesigned query algorithm with improved memory efficiency and throughput. As a result, T-REX answers queries in less than 10ms on a network of Europe, including local and long-distance transit. This constitutes a speedup of 20 compared to TB and 80 compared to algorithms without preprocessing. The memory footprint is moderate and the precomputation takes only two minutes, while real-time schedule updates can be incorporated in a few seconds. These properties make T-REX the first public transit journey planning algorithm that fulfills the requirements of interactive real-time applications on the continental scale.

cs.SI↗

arXiv2023

FLASH-TB: Integrating Arc-Flags and Trip-Based Public Transit Routing

We present FLASH-TB, a journey planning algorithm for public transit networks that combines Trip-Based Public Transit Routing (TB) with the Arc-Flags speedup technique. The basic idea is simple: The network is partitioned into a configurable number of cells. For each cell and each possible transfer between two vehicles, the algorithm precomputes a flag that indicates whether the transfer is required to reach the cell. During a query, only flagged transfers are explored. Our algorithm improves upon previous attempts to apply Arc-Flags to public transit networks, which saw limited success due to conflicting rules for pruning the search space. We show that these rules can be reconciled while still producing correct results. Because the number of cells is configurable, FLASH-TB offers a tradeoff between query time and memory consumption. It is significantly more space-efficient than existing techniques with a comparable preprocessing time, which store generalized shortest-path trees: to match their query performance, it requires up to two orders of magnitude less memory. The fastest configuration of FLASH-TB achieves a speedup of more than two orders of magnitude over TB, offering sub-millisecond query times even on large countrywide networks.

cs.DS↗

arXiv2021

Extending the Time Horizon: Efficient Public Transit Routing on Arbitrary-Length Timetables

We study the problem of computing all Pareto-optimal journeys in a public transit network regarding the two criteria of arrival time and number of transfers taken. In recent years, great advances have been made in making public transit network routing more scalable to larger networks. However, most approaches are silent on scalability in another dimension: Time. Experimental evaluations are often done on slices of timetables spanning a couple of days, when in reality, the planning horizon is much longer. We introduce an extension to trip-based public transit routing, proposed in [12], that allows efficient handling of arbitrarily long timetables. Our experimental evaluation shows that the resulting algorithm achieves fast queries on year-spanning timetables, and can incorporate updates such as delays or changed routes quickly even on large networks.

cs.DS↗

arXiv2021

Engineering In-place (Shared-memory) Sorting Algorithms

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.

cs.DC↗

arXiv2017

In-place Parallel Super Scalar Samplesort (IPS$^4$o)

We present a sorting algorithm that works in-place, executes in parallel, is cache-efficient, avoids branch-mispredictions, and performs work O(n log n) for arbitrary inputs with high probability. The main algorithmic contributions are new ways to make distribution-based algorithms in-place: On the practical side, by using coarse-grained block-based permutations, and on the theoretical side, we show how to eliminate the recursion stack. Extensive experiments show that our algorithm IPS$^4$o scales well on a variety of multi-core machines. We outperform our closest in-place competitor by a factor of up to 3. Even as a sequential algorithm, we are up to 1.5 times faster than the closest sequential competitor, BlockQuicksort.

cs.DC↗

arXiv2016

Trip-Based Public Transit Routing Using Condensed Search Trees

We study the problem of planning Pareto-optimal journeys in public transit networks. Most existing algorithms and speed-up techniques work by computing subjourneys to intermediary stops until the destination is reached. In contrast, the trip-based model focuses on trips and transfers between them, constructing journeys as a sequence of trips. In this paper, we develop a speed-up technique for this model inspired by principles behind existing state-of-the-art speed-up techniques, Transfer Pattern and Hub Labelling. The resulting algorithm allows us to compute Pareto-optimal (with respect to arrival time and number of transfers) 24-hour profiles on very large real-world networks in less than half a millisecond. Compared to the current state of the art for bicriteria queries on public transit networks, this is up to two orders of magnitude faster, while increasing preprocessing overhead by at most one order of magnitude.

cs.DS↗

arXiv2015

Trip-Based Public Transit Routing

We study the problem of computing all Pareto-optimal journeys in a public transit network regarding the two criteria of arrival time and number of transfers taken. We take a novel approach, focusing on trips and transfers between them, allowing fine-grained modeling. Our experiments on the metropolitan network of London show that the algorithm computes full 24-hour profiles in 70 ms after a preprocessing phase of 30 s, allowing fast queries in dynamic scenarios.

cs.DS↗