Robust random number generation for peer-to-peer systems

We consider the problem of designing an efficient and robust distributed random number generator for peer-to-peer systems that is easy to implement and works even if all communication channels are public. A robust random number generator is crucial for avoiding adversarial join–leave attacks on peer-to-peer overlay networks. We show that our new generator together with a light-weight rule recently proposed in [B. Awerbuch, C. Scheideler, Towards a scalable and robust DHT, in: Proc. of the 18th ACM Symp. on Parallel Algorithms and Architectures, SPAA, 2006. See also http://www14.in.tum.de/personen/scheideler] for keeping peers well distributed can keep various structured overlay networks in a robust state even under a constant fraction of adversarial peers.     

PCIR: Combining DHTs and peer clusters for efficient full-text P2P indexing

Distributed hash tables (DHTs) are very efficient for querying based on key lookups. However, building huge term indexes, as required for IR-style keyword search, poses a scalability challenge for plain DHTs. Due to the large sizes of document term vocabularies, peers joining the network cause huge amounts of key inserts and, consequently, a large number of index maintenance messages. Thus, the key to exploiting DHTs for distributed information retrieval is to reduce index maintenance costs. Various approaches in this direction have been pursued, including the use of hybrid infrastructures, or changing the granularity of the inverted index to peer level. We show that indexing costs can be significantly reduced further by letting peers form groups in a self-organized fashion. Instead of each individual peer submitting index information separately, all peers of a group cooperate to publish the index updates to the DHT in batches. Our evaluation shows that this approach reduces index maintenance cost by an order of magnitude, while still keeping a complete and correct term index for query processing.     

Reducing communication costs in robust peer-to-peer networks

Several recent research results describe how to design Distributed Hash Tables (DHTs) that are robust to adversarial attack via Byzantine faults. Unfortunately, all of these results require a significant blowup in communication costs over standard DHTs. For example, to perform a lookup operation, all such robust DHTs of which we are aware require sending O(log³n) messages while standard DHTs require sending only O(logn), where n is the number of nodes in the network. In this paper, we describe protocols to reduce the communication costs of all such robust DHTs. In particular, we give a protocol to reduce the number of messages sent to perform a lookup operation from O(log³n) to O(log²n) in expectation. Moreover, we also give a protocol for sending a large (i.e. containing Ω(log⁴n) bits) message securely through a robust DHT that requires, in expectation, only a constant blowup in the total number of bits sent compared with performing the same operation in a standard DHT. This is an improvement over the O(log²n) bit blowup that is required to perform such an operation in all current robust DHTs. Both of our protocols are robust against an adaptive adversary.     

分布式深度学习隐私与安全攻击研究进展与挑战

不同于集中式深度学习模式,分布式深度学习摆脱了模型训练过程中数据必须中心化的限制,实现了数据的本地操作,允许各方参与者在不交换数据的情况下进行协作,显著降低了用户隐私泄露风险,从技术层面可以打破数据孤岛,显著提升深度学习的效果,能够广泛应用于智慧医疗、智慧金融、智慧零售和智慧交通等领域.但生成对抗式网络攻击、成员推理攻...     

Montra: A large-scale DHT traffic monitor

This paper presents a new technique, called Montra, for accurately capturing traffic in a widely deployed DHT. The basic idea is to make the traffic monitors minimally visible to participating peers to avoid disruption in the system. We describe how Montra leverages the required redundancy in published content and routing to minimize disruption of the system. Validations of Montra over two widely deployed DHTs, namely Kad and Azureus, show that it can accurately capture more than 90% of traffic destined to monitored peers. Furthermore, the lightweight nature of Montra allows it to monitor a large number of peers with a moderate amount of resources. We use Montra to characterize several aspects of traffic in our two target DHTs.     

Efficient range query processing over DHTs based on the balanced Kautz tree

Distributed Hash Tables (DHTs) are scalable, self‐organizing, and adaptive to underlying topology changes, thus being a promising infrastructure for hosting large‐scale distributed applications. The ever‐wider use of DHT infrastructures has found more and more applications that require support for range queries. Recently, a number of DHT‐based range query schemes have been proposed. However, most of them suffer from high query delay or imbalanced load distribution. To address these problems, in this paper we first present an efficient indexing structure called Balanced Kautz (BK) tree that uniformly maps the m‐dimensional data space onto DHT nodes, and then propose a BK tree‐based range query scheme called ERQ that processes range queries in a parallel fashion and guarantees to return the results in a bounded delay. In a DHT with N nodes, ERQ can answer any range of query in less than rmlog N(2loglog N + 1) hops in a load‐balanced manner, irrespective of the queried range, the whole space size, or the number of queried attributes. The effectiveness of our proposals is demonstrated through experiments. Copyright © 2010 John Wiley & Sons, Ltd.     

基于剪枝技术和鲁棒蒸馏融合的轻量对抗攻击防御方法

对抗训练是一类常用的对抗攻击防御方法,其通过将对抗样本纳入训练过程,从而有效抵御对抗攻击。然而,对抗训练模型的鲁棒性通常依赖于网络容量的提升,即对抗训练所获得的网络为防御对抗攻击而大幅提升网络的模型容量,对其可用性造成较大约束。因此,如何在保证对抗训练模型鲁棒性的同时,降低模型容量,提出轻量对抗攻击防御方法是一大挑战。为解决以上问题,提出一种基于剪枝技术和鲁棒蒸馏融合的轻量对抗攻击防御方法。该方法以对抗鲁棒准确率为优化条件,在对预训练的鲁棒对抗模型进行分层自适应剪枝压缩的基础上,再对剪枝后的网络进行基于数据过滤的鲁棒蒸馏,实现鲁棒对抗训练模型的有效压缩,降低其模型容量。在CIFAR-10和CIFAR-100数据集上对所提出的方法进行性能验证与对比实验,实验结果表明,在相同 TRADES 对抗训练下,所提出的分层自适应剪枝技术相较于现有剪枝技术,其剪枝所得到的网络结构在多种 FLOPs 下均表现出更强的鲁棒性。此外,基于剪枝技术和鲁棒蒸馏融合的轻量对抗攻击防御方法相较于其他鲁棒蒸馏方法表现出更高的对抗鲁棒准确率。因此,实验结果证明所提方法在降低对抗训练模型容量的同时,相较于现有方法具有更强的鲁棒性,提升了对抗训练模型在物联网边缘计算环境的适用性。     

An effective single‐hop distributed hash table with high lookup performance and low traffic overhead

Distributed hash tables (DHTs) have been used in several applications, but most DHTs have opted to solve lookups with multiple hops, to minimize bandwidth costs while sacrificing lookup latency. This paper presents D1HT, an original DHT that has a peer‐to‐peer and self‐organizing architecture and maximizes lookup performance with reasonable maintenance traffic, and a Quarantine mechanism to reduce overheads caused by volatile peers. We implemented both D1HT and a prominent single‐hop DHT, and we performed an extensive and highly representative DHT experimental comparison, followed by complementary analytical studies. In comparison with current single‐hop DHTs, our results showed that D1HT consistently had the lowest bandwidth requirements, with typical reductions of up to one order of magnitude, and that D1HT could be used even in popular Internet applications with millions of users. In addition, we ran the first latency experiments comparing DHTs to directory servers, which revealed that D1HT can achieve latencies equivalent to or better than a directory server, and confirmed its greater scalability properties. Overall, our extensive set of results allowed us to conclude that D1HT can provide a very effective solution for a broad range of environments, from large‐scale corporate data centers to widely deployed Internet applications. Copyright © 2014 John Wiley & Sons, Ltd.     

Efficient Range Query Processing in Peer-to-Peer Systems

With the increasing popularity of the peer-to-peer (P2P) computing paradigm, many general range query schemes for distributed hash table (DHT)-based P2P systems have been proposed in recent years. Although those schemes can provide range query capability without modifying the underlying DHTs, they have the query delay depending on both the scale of the system and the size of the query space or the specific query, and thus cannot guarantee to return the query results in a bounded delay. In this paper, we propose Armada, an efficient range query processing scheme to support delay-bounded single-attribute and multiple-attribute range queries. It is the first delay-bounded general range query scheme on constant-degree DHTs, and can return the results for any range query within 2logN hops in a P2P system with N peers. Results of analysis and simulations show that the average delay in Armada is less than logN, and the average message cost of single-attribute range queries is about logN+2n 2 (n is the number of peers that intersect with the query). These results are very close to the lower bounds on delay and message cost of range queries over constant-degree DHTs.     

虚拟计算环境中的可扩展分布式资源信息服务

基于网络资源的"成长性"、"自治性"和"多样性",近年来,人们提出以通用DHT(distributed Hash table)信息服务的形式实现网络资源信息的发布和查询.然而,现有的资源信息服务在通用性、易用性和自适应性等方面仍存在不足.针对虚拟计算环境iVCE(Internet-based virtual computing environment)的资源聚合需求,提出构建可扩展的分布式资源信息服务SDIRIS(scalable distributed resource information service).首先,提出采用自适应DHT(adaptive FissionE,简称A-FissionE)底层架构,以对上层应用透明的方式适应不同的系统规模和稳定性;其次,基于自适应DHT提出高效的多属性区间搜索算法(multiple-attribute range FissionE,简称MR-FissionE).理论分析和模拟结果表明,SDIRIS能够高效地实现资源信息的发布与查询功能.     

LIGHT: A Query-Efficient Yet Low-Maintenance Indexing Scheme over DHTs

DHT is a widely used building block for scalable P2P systems. However, as uniform hashing employed in DHTs destroys data locality, it is not a trivial task to support complex queries (e.g., range queries and k-nearest-neighbor queries) in DHT-based P2P systems. In order to support efficient processing of such complex queries, a popular solution is to build indexes on top of the DHT. Unfortunately, existing over-DHT indexing schemes suffer from either query inefficiency or high maintenance cost. In this paper, we propose LIGhtweight Hash Tree (LIGHT)—a query-efficient yet low-maintenance indexing scheme. LIGHT employs a novel naming mechanism and a tree summarization strategy for graceful distribution of its index structure. We show through analysis that it can support various complex queries with near-optimal performance. Extensive experimental results also demonstrate that, compared with state of the art over-DHT indexing schemes, LIGHT saves 50-75 percent of index maintenance cost and substantially improves query performance in terms of both response time and bandwidth consumption. In addition, LIGHT is designed over generic DHTs and hence can be easily implemented and deployed in any DHT-based P2P system.     

A Read-Only Distributed Hash Table

A distributed hash table (DHT) is an infrastructure to support resource discovery in large distributed systems. In a DHT, data items such as resources, indexes of resources or resource metadata, are distributed across an overlay network based on a hash function. However, this may not be desirable in commercial applications such as Grid and cloud computing whereby the presence of multiple administrative domains leads to the issues of data ownership and self-economic interests. In this paper, we present R-DHT (Read-only DHT), a DHT-based resource discovery scheme without distributing data items. To map each data item back onto its resource owner, a physical host, we virtualize each host into virtual nodes. Nodes are further organized as a segment-based overlay network which increases node failure resiliency without replicating data items. We demonstrate the feasibility of our proposed scheme by presenting R-Chord, an implementation of R-DHT using Chord as the underlying overlay graph, with lookup and maintenance optimizations. Through analytical and simulation analyses, we evaluate the performance of R-DHT and compare it with traditional DHTs in terms of lookup path length, resiliency to node failures, and maintenance overhead. Overall, we found that R-DHT is effective and efficient for resource indexing and discovery in large distributed systems with a strong commercial requirement.     

Re-Chord: A Self-stabilizing Chord Overlay Network

The Chord peer-to-peer system is considered, together with CAN, Tapestry and Pastry, as one of the pioneering works on peer-to-peer distributed hash tables (DHT) that inspired a large volume of papers and projects on DHTs as well as peer-to-peer systems in general. Chord, in particular, has been studied thoroughly, and many variants of Chord have been presented that optimize various criteria. Also, several implementations of Chord are available on various platforms. Though Chord is known to be very efficient and scalable and it can handle churn quite well, no protocol is known yet that guarantees that Chord is self-stabilizing, i.e., the Chord network can be recovered from any initial state in which the network is still weakly connected. This is not too surprising since it is known that the Chord network is not locally checkable for its current topology. We present a slight extension of the Chord network, called Re-Chord (reactive Chord), that turns out to be locally checkable, and we present a self-stabilizing distributed protocol for it that can recover the Re-Chord network from any initial state, in which the n peers are weakly connected. in O(nlogn) communication rounds. We also show that our protocol allows a new peer to join or an old peer to leave an already stable Re-Chord network so that within O((logn)²) communication rounds the Re-Chord network is stable again.     

内容寻址网络CAN研究

新一代结构化可扩展P2P系统采用支持分布式哈希表(Distributed Hash Table,DHT)的路由算法。CAN（Content-Addressable Network）在d维虚拟坐标空间上利用DHT来实现内容定位,具有较好的可扩展性、容错性和完全自组等特点。介绍了CAN的原理,重点分析了CAN的构建和路由算法,并讨论了CAN算法的几种改进策略。     

Extraction of Distinguished Hyperbolic Trajectories for 2D Time-Dependent Vector Field Topology

This paper does two main contributions to 2D time-dependent vector field topology. First, we present a technique for robust, accurate, and efficient extraction of distinguished hyperbolic trajectories (DHT), the generative structures of 2D time-dependent vector field topology. It is based on refinement of initial candidate curves. In contrast to previous approaches, it is robust because the refinement converges for reasonably close initial candidates, it is accurate due to its adaptive scheme, and it is efficient due to its high convergence speed. Second, we provide a detailed evaluation and discussion of previous approaches for the extraction of DHTs and time-dependent vector field topology in general. We demonstrate the utility of our approach using analytical flows, as well as data from computational fluid dynamics.     

Psu: a novel low-latency constant-degree overlay network

Many structured peer-to-peer (P2P) systems supported by distributed hash table (DHT) schemas have been proposed recently to improve the scalability of distributed virtual application systems. By organizing the peers based on interconnection topologies, existing proposed schemas are purely based on the logical relationship without knowledge of the physical networks. In this paper, we propose a new structured DHT schema, which receives routing information not just from virtual neighbors in P2P overlay network, but also from nearby physical neighbors. The average degree of our model is 5, the diameter is logarithmic. The simulation shows that our model achieves shorter query path length, higher clustering, and better robustness than other overlay networks which have the same level of degree and diameter.     

Psu: a novel low-latency constant-degree overlay network

Many structured peer-to-peer (P2P) systems supported by distributed hash table (DHT) schemas have been proposed recently to improve the scalability of distributed virtual application systems. By organizing the peers based on interconnection topologies, existing proposed schemas are purely based on the logical relationship without knowledge of the physical networks. In this paper, we propose a new structured DHT schema, which receives routing information not just from virtual neighbors in P2P overlay network, but also from nearby physical neighbors. The average degree of our model is 5, the diameter is logarithmic. The simulation shows that our model achieves shorter query path length, higher clustering, and better robustness than other overlay networks which have the same level of degree and diameter.     

基于分布式范围树的结构化P2P多维范围查询

基于分布式哈希表（DHT）的结构化P2P网络具有扩展性好、健壮和自组织等优点,但只支持精确匹配的查询.本文提出一种基于分布式范围树的结构化P2P范围查询方法（DRT-RQ）,该方法将多维索引的分布式范围树分发到已有的结构化DHT覆盖网络中,利用DHT系统提供的数据查找接口,有效实现数据对象的范围查询.实验结果表明,基于分布式范围树的范围查询（DRT-RQ）比基于前缀哈希树的范围查询（PHT-RQ）需要更短的查询延时.     

深度学习模型鲁棒性研究综述

在大数据时代下,深度学习理论和技术取得的突破性进展,为人工智能提供了数据和算法层面的强有力支撑,同时促进了深度学习的规模化和产业化发展.然而,尽管深度学习模型在现实应用中有着出色的表现,但其本身仍然面临着诸多的安全威胁.为了构建安全可靠的深度学习系统,消除深度学习模型在实际部署应用中的潜在安全风险,深度学习模型鲁棒性分...