Abstract

Abstract

Abstract

Abstract
Data base clusters based on share-nothing replication techniques are currently widely accepted as a practical solution to scalability and availability of the data tier. A key issue when planning such systems is the ability to meet service level agreements when load spikes occur or cluster nodes fail. This translates into the ability to provision and deploy additional nodes. Many current research efforts focus on designing autonomic controllers to perform such reconfiguration, tuned to quickly react to system changes and spawn new replicas based on resource usage and performance measurements. In contrast, we are concerned about the inherent impact of deploying an additional node to an online cluster, considering both the time required to finish such an action as well as the impact on resource usage and performance of the cluster as a whole. If noticeable, such impact hinders the practicability of self-management techniques, since it adds an additional dimension that has to he accounted for. Our approach is to systematically benchmark a number of different reconfiguration scenarios to assess the cost of bringing a new replica online. We consider factors such as: workload characteristics, incremental and parallel recovery, flow control and outdatedness of the recovering replica. As a result, we show that research should be refocused from optimizing the capture and transmition of changes to applying them, which in a realistic setting dominates the cost of the recovery operation.

Abstract
Although epidemic or gossip-based multicast is a robust and scalable approach to reliable data dissemination, its inherent redundancy results in high resource consumption on both links and nodes. Tins problem is aggravated in settings that have costlier or resource constrained links as happens in Cloud Computing infrastructures composed by several interconnected data centers across the globe. The goal of this work is therefore to improve the efficiency of gossip-based reliable multicast by reducing the load imposed on those constrained links. hi detail, the proposed CLON protocol combines an overlay that gives preference to local links and a dissemination strategy that takes into account locality. Extensive experimental evaluation using a very large number of simulated nodes shows that this results in a reduction of traffic in constrained links by an order of magnitude, while at the same time preserving the resilience properties that make gossip-based protocols so attractive.

Abstract
Gossip-based protocols have been gaining an increasing interest from the research community due to the high resilience to node churn and high scalability, thus making them suitable to modern large-scale dynamic systems. Unfortunately, these properties come at the cost of redundant message transmissions to ensure bimodal delivery to all interested peers. In systems with high message throughput, those additional messages could pose a significant burden on the excess of required bandwidth. Furthermore, the overlays upon which message dissemination takes place are oblivious to the underlying network, or rely on posterior optimizations that bias the overlay to mimic the network topology. This contributes even more to the required bandwidth as 'undesirable' paths are chosen with equal probability among desired ones. In a Cloud Computing scenario, nodes tend to be aggregated in sub-nets inside a data-center or in multiple data-centers, which are connected by costlier, long-distance links. The goal of this work is, therefore, to build an overlay that approximates the structure of the physical network, while ensuring the connectivity properties desirable to ensure reliable dissemination. By having each node judiciously choose which nodes are in its dissemination list at construction time, i.e. by giving preference to local nodes, we are able to significantly reduce the number of messages traversing the long-distance links. In a later stage, this overlay shall be presented as a service upon which data dissemination and management protocols could be run. Copyright 2009 ACM.