arXiv subjects

Bernhard Haeupler

Publications and source records attributed to Bernhard Haeupler.

At least 91 records · Page 5Linked to original sources

arXiv2017

Optimal Gossip Algorithms for Exact and Approximate Quantile Computations

This paper gives drastically faster gossip algorithms to compute exact and approximate quantiles. Gossip algorithms, which allow each node to contact a uniformly random other node in each round, have been intensely studied and been adopted in many applications due to their fast convergence and their robustness to failures. Kempe et al. [FOCS'03] gave gossip algorithms to compute important aggregate statistics if every node is given a value. In particular, they gave a beautiful $O(\log n + \log \frac{1}ε)$ round algorithm to $ε$-approximate the sum of all values and an $O(\log^2 n)$ round algorithm to compute the exact $ϕ$-quantile, i.e., the the $\lceil ϕn \rceil$ smallest value. We give an quadratically faster and in fact optimal gossip algorithm for the exact $ϕ$-quantile problem which runs in $O(\log n)$ rounds. We furthermore show that one can achieve an exponential speedup if one allows for an $ε$-approximation. We give an $O(\log \log n + \log \frac{1}ε)$ round gossip algorithm which computes a value of rank between $ϕn$ and $(ϕ+ε)n$ at every node.% for any $0 \leq ϕ\leq 1$ and $0 < ε< 1$. Our algorithms are extremely simple and very robust - they can be operated with the same running times even if every transmission fails with a, potentially different, constant probability. We also give a matching $Ω(\log \log n + \log \frac{1}ε)$ lower bound which shows that our algorithm is optimal for all values of $ε$.

cs.DS↗

arXiv2017

Synchronization Strings: Explicit Constructions, Local Decoding, and Applications

This paper gives new results for synchronization strings, a powerful combinatorial object that allows to efficiently deal with insertions and deletions in various communication settings: $\bullet$ We give a deterministic, linear time synchronization string construction, improving over an $O(n^5)$ time randomized construction. Independently of this work, a deterministic $O(n\log^2\log n)$ time construction was just put on arXiv by Cheng, Li, and Wu. We also give a deterministic linear time construction of an infinite synchronization string, which was not known to be computable before. Both constructions are highly explicit, i.e., the $i^{th}$ symbol can be computed in $O(\log i)$ time. $\bullet$ This paper also introduces a generalized notion we call long-distance synchronization strings that allow for local and very fast decoding. In particular, only $O(\log^3 n)$ time and access to logarithmically many symbols is required to decode any index. We give several applications for these results: $\bullet$ For any $δ<1$ and $ε>0$ we provide an insdel correcting code with rate $1-δ-ε$ which can correct any $O(δ)$ fraction of insdel errors in $O(n\log^3n)$ time. This near linear computational efficiency is surprising given that we do not even know how to compute the (edit) distance between the decoding input and output in sub-quadratic time. We show that such codes can not only efficiently recover from $δ$ fraction of insdel errors but, similar to [Schulman, Zuckerman; TransInf'99], also from any $O(δ/\log n)$ fraction of block transpositions and replications. $\bullet$ We show that highly explicitness and local decoding allow for infinite channel simulations with exponentially smaller memory and decoding time requirements. These simulations can be used to give the first near linear time interactive coding scheme for insdel errors.

cs.IT↗

arXiv2017

Parallel algorithms and concentration bounds for the Lovasz Local Lemma via witness DAGs

The Lovász Local Lemma (LLL) is a cornerstone principle in the probabilistic method of combinatorics, and a seminal algorithm of Moser & Tardos (2010) provides an efficient randomized algorithm to implement it. This can be parallelized to give an algorithm that uses polynomially many processors and runs in $O(\log^3 n)$ time on an EREW PRAM, stemming from $O(\log n)$ adaptive computations of a maximal independent set (MIS). Chung et al. (2014) developed faster local and parallel algorithms, potentially running in time $O(\log^2 n)$, but these algorithms require more stringent conditions than the LLL. We give a new parallel algorithm that works under essentially the same conditions as the original algorithm of Moser & Tardos but uses only a single MIS computation, thus running in $O(\log^2 n)$ time on an EREW PRAM. This can be derandomized to give an NC algorithm running in time $O(\log^2 n)$ as well, speeding up a previous NC LLL algorithm of Chandrasekaran et al. (2013). We also provide improved and tighter bounds on the run-times of the sequential and parallel resampling-based algorithms originally developed by Moser & Tardos. These apply to any problem instance in which the tighter Shearer LLL criterion is satisfied.

cs.DM↗

arXiv2017

Broadcasting in Noisy Radio Networks

The widely-studied radio network model [Chlamtac and Kutten, 1985] is a graph-based description that captures the inherent impact of collisions in wireless communication. In this model, the strong assumption is made that node $v$ receives a message from a neighbor if and only if exactly one of its neighbors broadcasts. We relax this assumption by introducing a new noisy radio network model in which random faults occur at senders or receivers. Specifically, for a constant noise parameter $p \in [0,1)$, either every sender has probability $p$ of transmitting noise or every receiver of a single transmission in its neighborhood has probability $p$ of receiving noise. We first study single-message broadcast algorithms in noisy radio networks and show that the Decay algorithm [Bar-Yehuda et al., 1992] remains robust in the noisy model while the diameter-linear algorithm of Gasieniec et al., 2007 does not. We give a modified version of the algorithm of Gasieniec et al., 2007 that is robust to sender and receiver faults, and extend both this modified algorithm and the Decay algorithm to robust multi-message broadcast algorithms. We next investigate the extent to which (network) coding improves throughput in noisy radio networks. We address the previously perplexing result of Alon et al. 2014 that worst case coding throughput is no better than worst case routing throughput up to constants: we show that the worst case throughput performance of coding is, in fact, superior to that of routing -- by a $Θ(\log(n))$ gap -- provided receiver faults are introduced. However, we show that any coding or routing scheme for the noiseless setting can be transformed to be robust to sender faults with only a constant throughput overhead. These transformations imply that the results of Alon et al., 2014 carry over to noisy radio networks with sender faults.

cs.DC↗

arXiv2017

Synchronization Strings: Codes for Insertions and Deletions Approaching the Singleton Bound

We introduce synchronization strings as a novel way of efficiently dealing with synchronization errors, i.e., insertions and deletions. Synchronization errors are strictly more general and much harder to deal with than commonly considered half-errors, i.e., symbol corruptions and erasures. For every $ε>0$, synchronization strings allow to index a sequence with an $ε^{-O(1)}$ size alphabet such that one can efficiently transform $k$ synchronization errors into $(1+ε)k$ half-errors. This powerful new technique has many applications. In this paper, we focus on designing insdel codes, i.e., error correcting block codes (ECCs) for insertion deletion channels. While ECCs for both half-errors and synchronization errors have been intensely studied, the later has largely resisted progress. Indeed, it took until 1999 for the first insdel codes with constant rate, constant distance, and constant alphabet size to be constructed by Schulman and Zuckerman. Insdel codes for asymptotically large or small noise rates were given in 2016 by Guruswami et al. but these codes are still polynomially far from the optimal rate-distance tradeoff. This makes the understanding of insdel codes up to this work equivalent to what was known for regular ECCs after Forney introduced concatenated codes in his doctoral thesis 50 years ago. A direct application of our synchronization strings based indexing method gives a simple black-box construction which transforms any ECC into an equally efficient insdel code with a slightly larger alphabet size. This instantly transfers much of the highly developed understanding for regular ECCs over large constant alphabets into the realm of insdel codes. Most notably, we obtain efficient insdel codes which get arbitrarily close to the optimal rate-distance tradeoff given by the Singleton bound for the complete noise spectrum.

cs.IT↗

arXiv2017

Making Asynchronous Distributed Computations Robust to Noise

We consider the problem of making distributed computations robust to noise, in particular to worst-case (adversarial) corruptions of messages. We give a general distributed interactive coding scheme which simulates any asynchronous distributed protocol while tolerating an optimal corruption of a $Θ(1/n)$ fraction of all messages while incurring a moderate blowup of $O(n\log^2 n)$ in the communication complexity. Our result is the first fully distributed interactive coding scheme in which the topology of the communication network is not known in advance. Prior work required either a coordinating node to be connected to all other nodes in the network or assumed a synchronous network in which all nodes already know the complete topology of the network.

cs.DS↗

arXiv2016

Bridging the Capacity Gap Between Interactive and One-Way Communication

We study the communication rate of coding schemes for interactive communication that transform any two-party interactive protocol into a protocol that is robust to noise. Recently, Haeupler (FOCS '14) showed that if an $ε> 0$ fraction of transmissions are corrupted, adversarially or randomly, then it is possible to achieve a communication rate of $1 - \widetilde{O}(\sqrtε)$. Furthermore, Haeupler conjectured that this rate is optimal for general input protocols. This stands in contrast to the classical setting of one-way communication in which error-correcting codes are known to achieve an optimal communication rate of $1 - Θ(H(ε)) = 1 - \widetildeΘ(ε)$. In this work, we show that the quadratically smaller rate loss of the one-way setting can also be achieved in interactive coding schemes for a very natural class of input protocols. We introduce the notion of average message length, or the average number of bits a party sends before receiving a reply, as a natural parameter for measuring the level of interactivity in a protocol. Moreover, we show that any protocol with average message length $\ell = Ω(\mathrm{poly}(1/ε))$ can be simulated by a protocol with optimal communication rate $1 - Θ(H(ε))$ over an oblivious adversarial channel with error fraction $ε$. Furthermore, under the additional assumption of access to public shared randomness, the optimal communication rate is achieved ratelessly, i.e., the communication rate adapts automatically to the actual error rate $ε$ without having to specify it in advance. This shows that the capacity gap between one-way and interactive communication can be bridged even for very small (constant in $ε$) average message lengths, which are likely to be found in many applications.

cs.IT↗

arXiv2015

Breathe before Speaking: Efficient Information Dissemination Despite Noisy, Limited and Anonymous Communication

Distributed computing models typically assume reliable communication between processors. While such assumptions often hold for engineered networks, e.g., due to underlying error correction protocols, their relevance to biological systems, wherein messages are often distorted before reaching their destination, is quite limited. In this study we take a first step towards reducing this gap by rigorously analyzing a model of communication in large anonymous populations composed of simple agents which interact through short and highly unreliable messages. We focus on the broadcast problem and the majority-consensus problem. Both are fundamental information dissemination problems in distributed computing, in which the goal of agents is to converge to some prescribed desired opinion. We initiate the study of these problems in the presence of communication noise. Our model for communication is extremely weak and follows the push gossip communication paradigm: In each round each agent that wishes to send information delivers a message to a random anonymous agent. This communication is further restricted to contain only one bit (essentially representing an opinion). Lastly, the system is assumed to be so noisy that the bit in each message sent is flipped independently with probability $1/2-ε$, for some small $ε>0$. Even in this severely restricted, stochastic and noisy setting we give natural protocols that solve the noisy broadcast and the noisy majority-consensus problems efficiently. Our protocols run in $O(\log n / ε^2)$ rounds and use $O(n \log n / ε^2)$ messages/bits in total, where $n$ is the number of agents. These bounds are asymptotically optimal and, in fact, are as fast and message efficient as if each agent would have been simultaneously informed directly by an agent that knows the prescribed desired opinion.

cs.DC↗

arXiv2015

Maximal Noise in Interactive Communication over Erasure Channels and Channels with Feedback

We provide tight upper and lower bounds on the noise resilience of interactive communication over noisy channels with feedback. In this setting, we show that the maximal fraction of noise that any robust protocol can resist is 1/3. Additionally, we provide a simple and efficient robust protocol that succeeds as long as the fraction of noise is at most 1/3 - ε. Surprisingly, both bounds hold regardless of whether the parties send bits or symbols from an arbitrarily large alphabet. We also consider interactive communication over erasure channels. We provide a protocol that matches the optimal tolerable erasure rate of 1/2 - εof previous protocols (Franklin et al., CRYPTO '13) but operates in a much simpler and more efficient way. Our protocol works with an alphabet of size 4, in contrast to prior protocols in which the alphabet size grows as epsilon goes to zero. Building on the above algorithm with a fixed alphabet size, we are able to devise a protocol for binary erasure channels that tolerates erasure rates of up to 1/3 - ε.

cs.DS↗

arXiv2014

Interactive Channel Capacity Revisited

We provide the first capacity approaching coding schemes that robustly simulate any interactive protocol over an adversarial channel that corrupts any $ε$ fraction of the transmitted symbols. Our coding schemes achieve a communication rate of $1 - O(\sqrt{ε\log \log 1/ε})$ over any adversarial channel. This can be improved to $1 - O(\sqrtε)$ for random, oblivious, and computationally bounded channels, or if parties have shared randomness unknown to the channel. Surprisingly, these rates exceed the $1 - Ω(\sqrt{H(ε)}) = 1 - Ω(\sqrt{ε\log 1/ε})$ interactive channel capacity bound which [Kol and Raz; STOC'13] recently proved for random errors. We conjecture $1 - Θ(\sqrt{ε\log \log 1/ε})$ and $1 - Θ(\sqrtε)$ to be the optimal rates for their respective settings and therefore to capture the interactive channel capacity for random and adversarial errors. In addition to being very communication efficient, our randomized coding schemes have multiple other advantages. They are computationally efficient, extremely natural, and significantly simpler than prior (non-capacity approaching) schemes. In particular, our protocols do not employ any coding but allow the original protocol to be performed as-is, interspersed only by short exchanges of hash values. When hash values do not match, the parties backtrack. Our approach is, as we feel, by far the simplest and most natural explanation for why and how robust interactive communication in a noisy environment is possible.

cs.DS↗

arXiv2014

Consistent Weighted Sampling Made Fast, Small, and Easy

Document sketching using Jaccard similarity has been a workable effective technique in reducing near-duplicates in Web page and image search results, and has also proven useful in file system synchronization, compression and learning applications. Min-wise sampling can be used to derive an unbiased estimator for Jaccard similarity and taking a few hundred independent consistent samples leads to compact sketches which provide good estimates of pairwise-similarity. Subsequent works extended this technique to weighted sets and show how to produce samples with only a constant number of hash evaluations for any element, independent of its weight. Another improvement by Li et al. shows how to speedup sketch computations by computing many (near-)independent samples in one shot. Unfortunately this latter improvement works only for the unweighted case. In this paper we give a simple, fast and accurate procedure which reduces weighted sets to unweighted sets with small impact on the Jaccard similarity. This leads to compact sketches consisting of many (near-)independent weighted samples which can be computed with just a small constant number of hash function evaluations per weighted element. The size of the produced unweighted set is furthermore a tunable parameter which enables us to run the unweighted scheme of Li et al. in the regime where it is most efficient. Even when the sets involved are unweighted, our approach gives a simple solution to the densification problem that other works attempted to address. Unlike previously known schemes, ours does not result in an unbiased estimator. However, we prove that the bias introduced by our reduction is negligible and that the standard deviation is comparable to the unweighted case. We also empirically evaluate our scheme and show that it gives significant gains in computational efficiency, without any measurable loss in accuracy.

cs.DS↗

arXiv2014

Broadcast Throughput in Radio Networks: Routing vs. Network Coding

The broadcast throughput in a network is defined as the average number of messages that can be transmitted per unit time from a given source to all other nodes when time goes to infinity. Classical broadcast algorithms treat messages as atomic tokens and route them from the source to the receivers by making intermediate nodes store and forward messages. The more recent network coding approach, in contrast, prompts intermediate nodes to mix and code together messages. It has been shown that certain wired networks have an asymptotic network coding gap, that is, they have asymptotically higher broadcast throughput when using network coding compared to routing. Whether such a gap exists for wireless networks has been an open question of great interest. We approach this question by studying the broadcast throughput of the radio network model which has been a standard mathematical model to study wireless communication. We show that there is a family of radio networks with a tight $Θ(\log \log n)$ network coding gap, that is, networks in which the asymptotic throughput achievable via routing messages is a $Θ(\log \log n)$ factor smaller than that of the optimal network coding algorithm. We also provide new tight upper and lower bounds that show that the asymptotic worst-case broadcast throughput over all networks with $n$ nodes is $Θ(1 / \log n)$ messages-per-round for both routing and network coding.

cs.DS↗

arXiv2014

Optimal Error Rates for Interactive Coding II: Efficiency and List Decoding

We study coding schemes for error correction in interactive communications. Such interactive coding schemes simulate any $n$-round interactive protocol using $N$ rounds over an adversarial channel that corrupts up to $ρN$ transmissions. Important performance measures for a coding scheme are its maximum tolerable error rate $ρ$, communication complexity $N$, and computational complexity. We give the first coding scheme for the standard setting which performs optimally in all three measures: Our randomized non-adaptive coding scheme has a near-linear computational complexity and tolerates any error rate $δ< 1/4$ with a linear $N = Θ(n)$ communication complexity. This improves over prior results which each performed well in two of these measures. We also give results for other settings of interest, namely, the first computationally and communication efficient schemes that tolerate $ρ< \frac{2}{7}$ adaptively, $ρ< \frac{1}{3}$ if only one party is required to decode, and $ρ< \frac{1}{2}$ if list decoding is allowed. These are the optimal tolerable error rates for the respective settings. These coding schemes also have near linear computational and communication complexity. These results are obtained via two techniques: We give a general black-box reduction which reduces unique decoding, in various settings, to list decoding. We also show how to boost the computational and communication efficiency of any list decoder to become near linear.

cs.DS↗

arXiv2014

Fast Structuring of Radio Networks for Multi-Message Communications

We introduce collision free layerings as a powerful way to structure radio networks. These layerings can replace hard-to-compute BFS-trees in many contexts while having an efficient randomized distributed construction. We demonstrate their versatility by using them to provide near optimal distributed algorithms for several multi-message communication primitives. Designing efficient communication primitives for radio networks has a rich history that began 25 years ago when Bar-Yehuda et al. introduced fast randomized algorithms for broadcasting and for constructing BFS-trees. Their BFS-tree construction time was $O(D \log^2 n)$ rounds, where $D$ is the network diameter and $n$ is the number of nodes. Since then, the complexity of a broadcast has been resolved to be $T_{BC} = Θ(D \log \frac{n}{D} + \log^2 n)$ rounds. On the other hand, BFS-trees have been used as a crucial building block for many communication primitives and their construction time remained a bottleneck for these primitives. We introduce collision free layerings that can be used in place of BFS-trees and we give a randomized construction of these layerings that runs in nearly broadcast time, that is, w.h.p. in $T_{Lay} = O(D \log \frac{n}{D} + \log^{2+ε} n)$ rounds for any constant $ε>0$. We then use these layerings to obtain: (1) A randomized algorithm for gathering $k$ messages running w.h.p. in $O(T_{Lay} + k)$ rounds. (2) A randomized $k$-message broadcast algorithm running w.h.p. in $O(T_{Lay} + k \log n)$ rounds. These algorithms are optimal up to the small difference in the additive poly-logarithmic term between $T_{BC}$ and $T_{Lay}$. Moreover, they imply the first optimal $O(n \log n)$ round randomized gossip algorithm.

cs.NI↗

arXiv2014

Simple, Fast and Deterministic Gossip and Rumor Spreading

We study gossip algorithms for the rumor spreading problem which asks each node to deliver a rumor to all nodes in an unknown network. Gossip algorithms allow nodes only to call one neighbor per round and have recently attracted attention as message efficient, simple and robust solutions to the rumor spreading problem. Recently, non-uniform random gossip schemes were devised to allow efficient rumor spreading in networks with bottlenecks. In particular, [Censor-Hillel et al., STOC'12] gave an O(log^3 n) algorithm to solve the 1-local broadcast problem in which each node wants to exchange rumors locally with its 1-neighborhood. By repeatedly applying this protocol one can solve the global rumor spreading quickly for all networks with small diameter, independently of the conductance. This and all prior gossip algorithms for the rumor spreading problem have been inherently randomized in their design and analysis. This resulted in a parallel research direction trying to reduce and determine the amount of randomness needed for efficient rumor spreading. This has been done via lower bounds for restricted models and by designing gossip algorithms with a reduced need for randomness. The general intuition and consensus of these results has been that randomization plays a important role in effectively spreading rumors. In this paper we improves over this state of the art in several ways by presenting a deterministic gossip algorithm that solves the the k-local broadcast problem in 2(k+log n)log n rounds. Besides being the first efficient deterministic solution to the rumor spreading problem this algorithm is interesting in many aspects: It is simpler, more natural, more robust and faster than its randomized pendant and guarantees success with certainty instead of with high probability. Its analysis is furthermore simple, self-contained and fundamentally different from prior works.

cs.DS↗

arXiv2014

Near Optimal Leader Election in Multi-Hop Radio Networks

We present distributed randomized leader election protocols for multi-hop radio networks that elect a leader in almost the same time $T_{BC}$ required for broadcasting a message. For the setting without collision detection, our algorithm runs with high probability in $O(D \log \frac{n}{D} + \log^3 n) \min\{\log\log n,\log \frac{n}{D}\}$ rounds on any $n$-node network with diameter $D$. Since $T_{BC} = Θ(D \log \frac{n}{D} + \log^2 n)$ is a lower bound, our upper bound is optimal up to a factor of at most $\log \log n$ and the extra $\log n$ factor on the additive term. This algorithm is furthermore the first $O(n)$ time algorithm for this setting. Our algorithms improve over a 25 year old simulation approach of Bar-Yehuda, Goldreich and Itai with a $O(T_{BC} \log n)$ running time: In 1987 they designed a fast broadcast protocol and subsequently in 1989 they showed how it can be used to simulate one round of a single-hop network that has collision detection in $T_{BC}$ time. The prime application of this simulation was to simulate Willards single-hop leader election protocol, which elects a leader in $O(\log n)$ rounds with high probability and $O(\log \log n)$ rounds in expectation. While it was subsequently shown that Willards bounds are tight, it was unclear whether the simulation approach is optimal. Our results break this barrier and essentially remove the logarithmic slowdown over the broadcast time $T_{BC}$ by going away from the simulation approach. We also give a distributed randomized leader election algorithm for the setting with collision detection that runs in $O(D + \log n \log \log n) \cdot \min\{\log \log n, \log \frac{n}{D}\}$ rounds. This round complexity is optimal up to $O(\log \log n)$ factors and improves over a deterministic algorithm that requires $Θ(n)$ rounds independently of the diameter $D$.

cs.DS↗

arXiv2014

Randomized Broadcast in Radio Networks with Collision Detection

We present a randomized distributed algorithm that in radio networks with collision detection broadcasts a single message in $O(D + \log^6 n)$ rounds, with high probability. This time complexity is most interesting because of its optimal additive dependence on the network diameter $D$. It improves over the currently best known $O(D\log\frac{n}{D}\,+\,\log^2 n)$ algorithms, due to Czumaj and Rytter [FOCS 2003], and Kowalski and Pelc [PODC 2003]. These algorithms where designed for the model without collision detection and are optimal in that model. However, as explicitly stated by Peleg in his 2007 survey on broadcast in radio networks, it had remained an open question whether the bound can be improved with collision detection. We also study distributed algorithms for broadcasting $k$ messages from a single source to all nodes. This problem is a natural and important generalization of the single-message broadcast problem, but is in fact considerably more challenging and less understood. We show the following results: If the network topology is known to all nodes, then a $k$-message broadcast can be performed in $O(D + k\log n + \log^2 n)$ rounds, with high probability. If the topology is not known, but collision detection is available, then a $k$-message broadcast can be performed in $O(D + k\log n + \log^6 n)$ rounds, with high probability. The first bound is optimal and the second is optimal modulo the additive $O(\log^6 n)$ term.

cs.DS↗

arXiv2014

Optimal Gossip with Direct Addressing

Gossip algorithms spread information by having nodes repeatedly forward information to a few random contacts. By their very nature, gossip algorithms tend to be distributed and fault tolerant. If done right, they can also be fast and message-efficient. A common model for gossip communication is the random phone call model, in which in each synchronous round each node can PUSH or PULL information to or from a random other node. For example, Karp et al. [FOCS 2000] gave algorithms in this model that spread a message to all nodes in $Θ(\log n)$ rounds while sending only $O(\log \log n)$ messages per node on average. Recently, Avin and Elsässer [DISC 2013], studied the random phone call model with the natural and commonly used assumption of direct addressing. Direct addressing allows nodes to directly contact nodes whose ID (e.g., IP address) was learned before. They show that in this setting, one can "break the $\log n$ barrier" and achieve a gossip algorithm running in $O(\sqrt{\log n})$ rounds, albeit while using $O(\sqrt{\log n})$ messages per node. We study the same model and give a simple gossip algorithm which spreads a message in only $O(\log \log n)$ rounds. We also prove a matching $Ω(\log \log n)$ lower bound which shows that this running time is best possible. In particular we show that any gossip algorithm takes with high probability at least $0.99 \log \log n$ rounds to terminate. Lastly, our algorithm can be tweaked to send only $O(1)$ messages per node on average with only $O(\log n)$ bits per message. Our algorithm therefore simultaneously achieves the optimal round-, message-, and bit-complexity for this setting. As all prior gossip algorithms, our algorithm is also robust against failures. In particular, if in the beginning an oblivious adversary fails any $F$ nodes our algorithm still, with high probability, informs all but $o(F)$ surviving nodes.

cs.DS↗