Browse Prior Art Database

Method for Efficient Image Distribution over a Cloud Environment

IP.com Disclosure Number: IPCOM000233940D
Publication Date: 2014-Jan-03
Document File: 6 page(s) / 129K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is a method to improve the efficiency and increase the distribution speed of installation images over a cloud environment. Using the nearest existing machines as a starting point to boot, new machines for installation use just small pieces of information, retrieved from the nearest host, to generate an image installation.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 30% of the total text.

Page 01 of 6

Method for Efficient Image Distribution over a Cloud Environment

Cloud computing is becoming an important market share to the Information Technology companies and making this service more efficient is fundamental to a company that wishes to become a competitive cloud computing service provider.

One of the main problems in making cloud computing more efficient is the way that installation images are distributed. If those images are not being distributed in an efficient way, some of the more important features of cloud computing can be degraded, such as the agility in scaling an environment. Another problem caused by

bad image distribution is the traffic congestion when transferring large images through the network. In this case, even the already installed machines can suffer performance degradation.

A common solution for this problem is to bring the image source near the target host or try to distribute the image downloads among a number of hosts .

Following are some of the current solutions:


Binary Tree Distribution: In this method, all compute nodes are arranged in a balanced binary tree with the source node as its root. The balanced tree property guarantees that the depth of the leaves differs by at least one . All transfers must be

synchronized to avoid a transfer on a child node starting before the data from the parent is available. The main problem of this method is to keep the binary tree balanced. When higher installation demands arrive, this can be a bottleneck. Even

with the transfer being near (avoiding transferring from large distances), this method must copy the whole image from a source host.

Peer-to-peer Distribution: This method uses the BitTorrent protocol to transfer images from a source host. The node hosting the source file starts a tracker that coordinates the distribution of the file. A file containing meta-data about the source file is generated and must be distributed to all clients. Clients, also called peers, connect to the tracker that indicates from which other peers' pieces of the file are available for download. The problem with this method can be the congestion caused by too many installations running at the same time or when just a few machines can be used as source to image downloads.

Network File System (NFS): This is the traditional way to distribute a number of virtual machine (VM) Images in a cloud site. Basically, the central NFS server stores the VM images and the worker nodes retrieve copies on demand. This leads to multiple point-to-point transfers. Furthermore, when multi-gigabytes files are accessed by a larger number of nodes at the same time , NFS shows erratic behavior.

The novel contribution is based on a premise that in a cloud computing environment the installation images have a lot of parts in common (e.g., the operating system kernel, program binary, etc.). In general, only the data used by the programs is changed over time. This way, the machines that will be installed can use as...