Citation Context

... bandwidth improves over time, we think that this tradeoff should favor latency rather than bandwidth. In contrast, most DHTs that have been proposed so far solve the lookups with multiple hops (e.g. =-=[14, 18, 21, 24, 25]-=-) in an attempt to minimize the maintenance traffic (network traffic required to maintain the routing tables). However, recent results [10] have shown that in some cases single-hop DHTs may generate l...

Citation Context

... of nodes in previous single-hop DHT [4]. Our results also show that D1HT is able to support vast P2P systems whose dynamics are similar to those of widely deployed P2P applications, such as Gnutella =-=[22]-=- and BitTorrent [2], with reasonable maintenance bandwidth demands. For instance, a huge one-million D1HT system, with dynamics similar to BitTorrent, would require only 3 kbps of duplex maintenance t...

Citation Context

... bandwidth improves over time, we think that this tradeoff should favor latency rather than bandwidth. In contrast, most DHTs that have been proposed so far solve the lookups with multiple hops (e.g. =-=[14, 18, 21, 24, 25]-=-) in an attempt to minimize the maintenance traffic (network traffic required to maintain the routing tables). However, recent results [10] have shown that in some cases single-hop DHTs may generate l...

Citation Context

...nce. Section 6 discuss related work and Section 7 concludes the paper. 2. System Design A D1HT system is composed of a set D of n peers 1 and maps items (or keys) to peers based on consistent hashing =-=[8]-=-, where both peers and keys are hashed to integer identifiers (IDs) in the same ID space [0..N], N ≫ n. Typically a key ID is the cryptographic hash SHA-1 of the key value, a peer ID is based on the S...

Citation Context

...o DHTs in order to reduce the number of hops for popular keys. Fig. 9. D1HT bandwidth requirements with and without Quarantine (Tq=10min). There is a number of systems, including Chord[24] and SkipNet=-=[6]-=-, where each peer uses pointers to nodes (fingers) with 2 i distances (usually 0 ≤ i ≤ log(N)), but those pointers are used only to route the lookups in O(log(n)) hops. In contrast, D1HT uses its 2 l ...

Citation Context

... bandwidth improves over time, we think that this tradeoff should favor latency rather than bandwidth. In contrast, most DHTs that have been proposed so far solve the lookups with multiple hops (e.g. =-=[14, 18, 21, 24, 25]-=-) in an attempt to minimize the maintenance traffic (network traffic required to maintain the routing tables). However, recent results [10] have shown that in some cases single-hop DHTs may generate l...

Citation Context

... dispersed in huge distributed environments. For this reason, DHTs have already been proposed as a base for a variety of distributed and P2P applications, ranging from grid services [23] to databases =-=[7]-=-, showing the large acceptance of DHTs as a useful distributed software. DHT systems implement a hash-table-like lookup facility where the keys (information) are distributed among the participant node...

Citation Context

...hash function is expected to randomly distribute the peers IDs along the ring, which can be accomplished by using a cryptographic hash function such as SHA-1[16]. Then, as in many other studies (e.g. =-=[4, 9, 10, 12, 14, 24]-=-), we will assume that the events are oblivious to the peers IDs, leading to a randomly distributed rate of r events per second in the system, and so the average amounts of incoming and outgoing traff...

Citation Context

...r quickly departs from the system. This problem is aggravated in the case of single hop DHTs as any joining peer should be acknowledged by the whole system. On the other hand, P2P measurement studies =-=[3, 22]-=- have shown that the statistical distributions of peer session lengths are usually heavy tailed, which means that peers that are connected to the system for a long time are likely to remain alive long...

Citation Context

...far most DHT evaluations based on real implementations used a hundred physical nodes at most (e.g. [7, 25]), while the DHT simulations presented are usually restricted to a maximum of 20K nodes (e.g. =-=[4, 5, 7, 9, 10, 11, 15, 17, 24, 25]-=-), and so they are not representative of popular P2P systems, which are able to support up to millions of users [1]. On the other hand, we believe that to be accepted as good estimates of real impleme...

Citation Context

...st one hop, to achieve f it is enough to assure that the hops will fail with probability f at most. Assuming that the lookup targets are randomly spread along the ring (as in many other studies, e.g. =-=[4, 9, 12, 10, 13, 15, 24]-=-), the average fraction of routing failures will be a direct result of the number of stale routing tables’ entries. In that manner, to satisfy a pre-defined average fraction of routing failures f, it ...

Citation Context

...r solve the lookups with multiple hops (e.g. [14, 18, 21, 24, 25]) in an attempt to minimize the maintenance traffic (network traffic required to maintain the routing tables). However, recent results =-=[10]-=- have shown that in some cases single-hop DHTs may generate less traffic than multi-hop ones, even for dynamic systems. Those results corroborate previous work [19], which indicated that low-overhead ...

Citation Context

...hash function is expected to randomly distribute the peers IDs along the ring, which can be accomplished by using a cryptographic hash function such as SHA-1[16]. Then, as in many other studies (e.g. =-=[4, 9, 10, 12, 14, 24]-=-), we will assume that the events are oblivious to the peers IDs, leading to a randomly distributed rate of r events per second in the system, and so the average amounts of incoming and outgoing traff...

Citation Context

...verhead multi-hop DHTs are required only for vast and very dynamic systems. On the other hand, there is only one proposed DHT system that ensures that most lookups are really solved with only one hop =-=[4]-=-, but this system imposes high levels of load imbalance and bandwidth overheads in order to maintain the routing tables. We consider that an effective single-hop DHT must exhibit the following four ma...

Citation Context

...T assures that a high fraction of the lookups takes just one hop. To the best of our knowledge, there is no DHT system proposed so far that uses an event reporting algorithm similar to EDRA. Accordion=-=[11]-=- also addresses the tradeoff between lookup latency and bandwidth requirements, but its approach is quite different from ours. Accordion implements some very clever adaptation techniques that aim to s...

Citation Context

...far most DHT evaluations based on real implementations used a hundred physical nodes at most (e.g. [7, 25]), while the DHT simulations presented are usually restricted to a maximum of 20K nodes (e.g. =-=[4, 5, 7, 9, 10, 11, 15, 17, 24, 25]-=-), and so they are not representative of popular P2P systems, which are able to support up to millions of users [1]. On the other hand, we believe that to be accepted as good estimates of real impleme...

Citation Context

...st one hop, to achieve f it is enough to assure that the hops will fail with probability f at most. Assuming that the lookup targets are randomly spread along the ring (as in many other studies, e.g. =-=[4, 9, 12, 10, 13, 15, 24]-=-), the average fraction of routing failures will be a direct result of the number of stale routing tables’ entries. In that manner, to satisfy a pre-defined average fraction of routing failures f, it ...

Citation Context

...s single-hop DHT [4]. Our results also show that D1HT is able to support vast P2P systems whose dynamics are similar to those of widely deployed P2P applications, such as Gnutella [22] and BitTorrent =-=[2]-=-, with reasonable maintenance bandwidth demands. For instance, a huge one-million D1HT system, with dynamics similar to BitTorrent, would require only 3 kbps of duplex maintenance traffic to assure 1-...

Citation Context

...peer for f =1%. Fig. 7. D1HT peer bandwidth demands in kbits/sec for Savg =2.9 hours. Fig. 8. Duration of the Θ interval in seconds for f =1%and different Savg values. 6. Related Work Rodrigues et al =-=[20]-=- proposed a single hop DHT in a complete different context from ours, as their system was based on dedicated servers arranged on a two level hierarchy, and their main goal was to obtain robustness aga...

Citation Context

... bandwidth improves over time, we think that this tradeoff should favor latency rather than bandwidth. In contrast, most DHTs that have been proposed so far solve the lookups with multiple hops (e.g. =-=[14, 18, 21, 24, 25]-=-) in an attempt to minimize the maintenance traffic (network traffic required to maintain the routing tables). However, recent results [10] have shown that in some cases single-hop DHTs may generate l...

Citation Context

...ng tables). However, recent results [10] have shown that in some cases single-hop DHTs may generate less traffic than multi-hop ones, even for dynamic systems. Those results corroborate previous work =-=[19]-=-, which indicated that low-overhead multi-hop DHTs are required only for vast and very dynamic systems. On the other hand, there is only one proposed DHT system that ensures that most lookups are real...

Citation Context

...information widely dispersed in huge distributed environments. For this reason, DHTs have already been proposed as a base for a variety of distributed and P2P applications, ranging from grid services =-=[23]-=- to databases [7], showing the large acceptance of DHTs as a useful distributed software. DHT systems implement a hash-table-like lookup facility where the keys (information) are distributed among the...