An In-Network Replica Selection Framework for Distributed Data Stores with Critical Latency

PROJECT TITLE :

An In-Network Replica Selection Framework for Latency-Critical Distributed Data Stores

ABSTRACT:

It is common for the performance of individual servers in distributed data stores to vary, and this is especially true when the servers are hosted in a cloud-based environment. Because of this, the replica chosen to fulfill a reading request will have an immediate impact on the response latency. However, there are many challenges associated with replica selection in latency-critical data stores (e.g., key-value stores). These types of data stores typically deal with data of a small size, and customers are required to make independent replica selections. Even the most advanced algorithm for replica selection currently available (C3) still has a lot of room to grow in terms of its latency performance. According to the results of our experiments, using the ideal replica selection (Oracle) results in a reduction of approximately 34-60 percent of the 99th latency when compared with using C3. Within the context of this article, "NetRS" refers to a solution that we propose to address the fundamental factors that prevent replica selection algorithms from being effective. The NetRS framework enables in-network replica selection for distributed data stores. This functionality is provided by the framework. It does this by taking advantage of newly developed network devices such as programmable switches and network accelerators to select replicas for requests. In addition to supporting a wide variety of replica selection algorithms, NetRS is also well-suited to the network topology of contemporary data centers. According to the extensive evaluations that we have conducted, when compared with the traditional method in which clients select replicas for requests, the mean latency can be reduced by up to 50.3% using NetRS, and the 99th latency can be reduced by up to 56.3% using NetRS. In addition, NetRS is able to effectively cut response latency even when unexpected events (such as changes in workload or failures of network devices) and network congestion occur.

Did you like this research project?

To get this research project Guidelines, Training and Code... Click Here