|Title:||Efficient Distributed File System (EDFS)
|Author(s):||Kassa, Debessay Fesehaye; Malik, Rahul; Nahrstedt, Klara
|Subject(s):||Cloud architecture, resource allocation, distributed namenode
|Abstract:||Large scale distributed systems such as cloud computing
applications are becoming very common. These applications
come with increasing challenges on how to transfer
and where to store and compute data. The most prevalent
distributed file systems to deal with these challenges is the
Hadoop File System (HDFS) which is a variant of the Google
File System (GFS). However HDFS has two potential problems.
The first one is that it depends on a single name node
to manage almost all operations of every data block in the
file system. As a result it can be a bottleneck resource and
a single point of failure. The second potential problem with
HDFS is that it depends on TCP to transfer data. As has
been cited in many studies TCP takes many rounds before
it can send at the full capacity of the links in the cloud. This
results in low link utilization and longer download times.
To overcome these problems of HDFS we present a new distributed
file system. Our scheme uses a light weight front
end server to connect all requests with many name nodes.
This helps distribute load of a single name node to many
name nodes. Our second contribution is to use an efficient
protocol to send and route data. Our protocol can achieve
full link utilization and hence decreased download times.
Based on simulation our protocol can outperform HDFS and
|Genre:||Presentation / Lecture / Speech
|Date Available in IDEALS:||2010-07-11