Discipline: Computer Sciences and Information Management
Subcategory: Computer Science & Information Systems
Pedro Ledesma - California State University Dominguez Hills
Co-Author(s): Dr.Bin Tang, Dr.Mohsen Beheshti
Data centers are becoming increasingly popular for their flexibility and processing capabilities in the modern computing environment. With datacenter networks increasing in size, full-bandwidth connectivity between all pairs of servers is becoming difficult and expensive to scale. Therefore, a fat tree is an important implementation for a datacenter because looking at it from a bandwidth perspective, this topology has identical bandwidth at any bisections and each layer has the same aggregated bandwidth. As for the money side, a fat tree topology can be built by using cheap devices with static capacity where each port supports same speed as end host and all devices can transmit at line speed if packets are distributed uniform along available paths.
In this paper, we discuss the implementation of a fat tree topology in a SDN cloud data center called CloudSimDN. This framework is put through a series of use cases to send packets of different data sizes to and from a specific pair of virtual machines using different scenarios of bandwidth and workloads between links. These use-cases demonstrate the effectiveness that a tree topology can have on energy consumption and transmission time when a packet has to travel through certain links to get to its destination node. CloudSimSDN is only tested through a tree topology, this is not a realistic topology for a real-world data center design. The fat tree is used for its non-blocking nature, providing many redundant paths between any 2 hosts.
This paper utilizes CloudSimSDN, a simulation framework for SDN-enabled cloud environments based on CloudSim. The type of fat tree that is used in these use cases is a K-ary fat tree. Which is a three-layer physical topology (edge, aggregation and core), where each pod consists of (k/2)2 servers and 2 layers of k/2 k-port switches. Each edge switch connects to k/2 servers and k/2 aggregate switches. Each aggregate switch connects to k/2 edge & k/2 core switches. Each (k/2)2 core switches connects to k pods. A virtual topology is also used in order to place virtual machines in their respective host. Both the physical topology and virtual topology are in JSON format. In order to make this simulation happen, we also need a CSV workload file which will generate traffic to be able to analyze the results of this topology.
Since this fat tree enables full bisection bandwidth at level of the topology, it is predicted that all packet transfers will occur with no complications. We will utilize best-fit and worst-fit VM allocation algorithms in order to fully optimize our fat tree topology and determine which one can produce better results. We will measure the amount of energy used for a packet to go from a source node to the destination node along with the transmission time as well.
The reason for the implementation of the fat tree topology is because common data center designs do come with their faults. One of the biggest faults that common data center designs have is oversubscription. This is where connecting multiple devices to the same switch port to optimize the switch use. This can cause multiple devices connected to the same switch port to contend for that port?s bandwidth, resulting in poor response time. This is why a fat tree topology is a better design for datacenter since it stays consistent when it comes to bandwidth and eliminates congestion. In our future work we want to be able to create VM replication. This where a virtual machine on the network is replicated and placed in the physical machine that is closest to the source virtual machine. This is very efficient because instead of the workloads traveling throughout the network for a specific node, it will be able to travel to the closest physical machine and therefore use less energy.
Funder Acknowledgement(s): National Science Foundation
Faculty Advisor: Bin Tang, btang@csudh.edu
Role: I did all the research, my mentor and advisor would lend me a hand when I would get stuck but this was all done by me.