Research

Technical Report Results

Technical Report TR722:
A Survey on SEDA and Using SEDA to Optimize HDFS Read Operation

Isuru Suriarachchi
(Oct 2015), 8
[Written for PhD qualifying exam]
Abstract:
Handling high concurrency is a critical issue in almost all internet services. A web server is a perfect example where a large amount of concurrent users are served by managing limited resources. There has been number of different architectures proposed to optimize the resource usage and maximize the throughput in such applications. Thread-per-request model and event-driven model are two heavily used such architectures. In 2001, Welsh et al. pointed out the performance issues in those two architectures under high concurrency and proposed Staged Event Driven Architecture (SEDA) to address the issues. SEDA uses a hybrid of thread pools and eventing concept to utilize the advantages of both techniques. Since the introduction, there have been number of applications of SEDA in highly concurrent systems. In this paper, first we present few most interesting applications of SEDA selected through a survey. Then we discuss SOR-HDFS which is an application of SEDA to improve HDFS Write operation. There we try to highlight the shortcomings and restrictions as well in SOR-HDFS design. Finally we present a design to use SEDA for HDFS Read operation and discuss how it improves read performance.

Available as:
  • PDF (317 KBytes)