Citation Context

... distribute web sites among servers distributed around the globe relieving hot spots in the Internet [9]. Besides the area of Web Caching they are popular in Peerto-Peer Networks, see CAN [15], Chord =-=[16]-=-, Pastry [7], Tapestry [8], and many more. A further important application field are Storage Area Networks (SAN), to overcome problems induced by huge RAID arrays. Here, the task is to distribute data...

Citation Context

...ntroduced to distribute web sites among servers distributed around the globe relieving hot spots in the Internet [9]. Besides the area of Web Caching they are popular in Peerto-Peer Networks, see CAN =-=[15]-=-, Chord [16], Pastry [7], Tapestry [8], and many more. A further important application field are Storage Area Networks (SAN), to overcome problems induced by huge RAID arrays. Here, the task is to dis...

Citation Context

...mmonly based on consistent hashing [9] scheme or on the work of Plaxton et al. [13]. Usual distributed hash tables includes CAN[15], Chord[16], Distance Halving[12], Koorde[10], Pastry[7] or Tapestry =-=[17]-=- to name some of them. All these approaches have in common that they are for homogeneous purpose only, which means their methods neglects somehow the heterogeneity of nodes. We have deployed a consist...

Citation Context

...two methods for weighted consistent hashing also known as weighted distributed hash tables. The first method, called Linear Method, combines the standard consistent hasing introduced by Karger et al. =-=[9]-=- with a linear weighted distance measure. By using node copies and different partitions of the hash space, the balance of this scheme approximates the fair weight relationship with high probability. T...

Citation Context

...eighted hashing schemes. In the last section we summarize our results. 2. RELATED RESEARCH Distributed hash tables are commonly based on consistent hashing [9] scheme or on the work of Plaxton et al. =-=[13]-=-. Usual distributed hash tables includes CAN[15], Chord[16], Distance Halving[12], Koorde[10], Pastry[7] or Tapestry [17] to name some of them. All these approaches have in common that they are for ho...

Citation Context

...og n) copies of each node. Then by applying Chernoff bounds one can show that the error reduces to a small factor of 1±ɛ for any ɛ > 0. Recently, other approaches have been presented for this problem =-=[11]-=-. It is an open problem how the principle of multiple choice, or the principle of two choices can be applied to this weighted distributed hashing. 4. THE LINEAR METHOD In this section we present the L...

Citation Context

... servers distributed around the globe relieving hot spots in the Internet [9]. Besides the area of Web Caching they are popular in Peerto-Peer Networks, see CAN [15], Chord [16], Pastry [7], Tapestry =-=[8]-=-, and many more. A further important application field are Storage Area Networks (SAN), to overcome problems induced by huge RAID arrays. Here, the task is to distribute data on multiple heterogeneous...

Citation Context

... RESEARCH Distributed hash tables are commonly based on consistent hashing [9] scheme or on the work of Plaxton et al. [13]. Usual distributed hash tables includes CAN[15], Chord[16], Distance Halving=-=[12]-=-, Koorde[10], Pastry[7] or Tapestry [17] to name some of them. All these approaches have in common that they are for homogeneous purpose only, which means their methods neglects somehow the heterogene...

Citation Context

...h range minimizes the distance to the image of the data element. If a sufficient number of copies of the hosts are mapped to the range, one can show that this leads to a fair balance of data elements =-=[14, 6]-=-. Now, if a new host is added to this system, then only data elements need to be reassigned which will be stored on the new host. This feature is called consistency. Such distributed hash tables are u...

Citation Context

...h range minimizes the distance to the image of the data element. If a sufficient number of copies of the hosts are mapped to the range, one can show that this leads to a fair balance of data elements =-=[14, 6]-=-. Now, if a new host is added to this system, then only data elements need to be reassigned which will be stored on the new host. This feature is called consistency. Such distributed hash tables are u...

Citation Context

...opies for each peer wi. This is not feasible, if maxj∈V {wj}/ minj∈V {wj} is too large. Furthermore, interesting nodes with small weights increases the number of copies of all nodes. Brinkmann et al. =-=[5]-=- presented a scheme to overcome this problem, but not as elegant as the original weighted consistent hashing and still uses a large number of copies. Furthermore, small disks are under-utilized and th...

Citation Context

..., that identifies responsible nodes to determine the next hop for route completion. Our association of hash values to hash locations works with a concept that is similarly to the scheme introduced by =-=[4]-=-, yet improves on the balancing properties and can be evaluated more efficiently by this distributed network. In [5] Brinkmann et al. the authors introduced several criteria a placement scheme needs t...

Citation Context

... yet using the height Hi = −(ln(1 − di(x))/wi for finding the host vi for the data element minimizing this term. The Linear Method comes historically first and is already mentioned and implemented in =-=[2]-=-. We know that the Logarithmic Method outperforms the Linear Method. Yet, knowing the Linear Method helps to understand the Logarithmic Method. Therefore and because of its simplicity and elegance we ...

Citation Context

...pplication field are Storage Area Networks (SAN), to overcome problems induced by huge RAID arrays. Here, the task is to distribute data on multiple heterogeneous disks that act like one virtual disk =-=[3]-=-. In this paper we consider the more general case of weighted consistent hashing and compared to the original approach every host vi comes now with a positive weight wi. Let W = P i∈V wi be the overal...

Citation Context

...ons has a fragmentation of qkn for n nodes. The Logarithmic Method using k partitions has a fragmentation of 2kn − 1. Proof. This theorem follows by applying results from Davenport Schinzel Sequences =-=[1]-=-. The standard choice of parameters is q = O(1) and k = O(log n) to achieve constant precision with high probability. Therefore both methods provide a fragmentation of O(n log n). 6.2 Double Hash Func...