A comparison of latency and jitter on OpenFlow and Ethernet pipelines to evaluate Software Defined Networking (SDN) controllers for Disaster Recovery Exercises (DREs).

摘要

Availability of mission critical applications is one of the most important requirements for modern enterprises. Data centers, where these applications run, are often very complex systems comprised of network, compute, and storage resources. These resources typically consist of both physical and virtual components. In order to ensure high-availability of mission-critical applications, most businesses conduct Disaster Recovery Exercises (DREs). One of the main goals of the DREs is to ensure the business can recover their applications and production data in case of any failures.;One of the more challenging aspects of conducting DREs is to ensure isolation of the resources being used for the DRE from the production resources. If not planned carefully, it is possible to impact the production applications up to and including corruption of the production data.;While it is possible to have a complete physically separate DRE area comprised of duplicate hardware and software, many organizations find it cost prohibitive. For this reason, it is common for organizations to temporarily reassign a subset of resources for DREs. After the DRE is completed, the resources are reassigned back to the production environment.;The temporary reassignment of resources to the DRE area currently is done by physically recabling resources from the production network to the DRE area. While this meets the objective of isolation between the two areas, it can drastically increase the amount of time a resource is unavailable for production activities and also increases the operational expense (OPEX) of conducting the DRE.;This study presents a principled approach to DREs that takes advantage of a recent development in computer networking called Software Defined Networking (SDN). Traditional network devices contain both the control plane and data plane in the same physical device. SDN abstracts the control plane from the physical network device and centralizes it in a SDN controller. Network administrators can use an SDN controller to centrally reconfigure the forwarding path of network devices instead of manual configuration of each separate network device.;In this study, research is conducted on OpenFlow and Ethernet pipelines. Specifically, equivalent tests are run between OpenFlow and Ethernet pipelines to determine the differences in Jitter and Latency and an evaluation on the use of SDN controllers for DREs is made. The data from this research suggests it is possible to virtually isolate production resources for DREs utilizing SDN Controllers. Latency is the response time or how much time it takes for a packet of data to reach its destination. Jitter (latency variation) is the variation in the delay of packets reaching its destination. To this purpose, the researcher developed an application, SDN Rapid Adjustable Networking (RAN), that's purpose, is to rapidly isolate virtual and physical resources needed for a DRE via an SDN controller. This application reduces the OPEX and capital expenses (CAPEX) required for DREs by drastically reducing the time to setup and teardown an isolated DRE environment and by temporarily repurposing production resources to the DRE environment. This application is designed to be placed in a traditional Ethernet network comprised of traditional Ethernet switches.;The SDN RAN Application is written in Python and takes advantage of the OpenDaylight Controller. OpenDaylight is a Linux Foundation Collaborative Project that has an open platform for network programmability that enables SDN. Specifically, SDN RAN is able to work in conjunction with the OpenDaylight Controller to isolate physical and virtual production resources in a matter of seconds for DRE testing and repurpose them back to production after the DRE is completed. The uniqueness of this approach is using an SDN controller on a traditional Ethernet network with traditional Ethernet switches instead of an SDN network comprised of switches that utilize SDN protocols. This approach allows SDN automation capability to be used in an existing traditional Ethernet network comprised of traditional Ethernet switches and override the forwarding path of data packets on only the network resources that have been selected to participate in the DRE.;Through the experimental study, the primary inferential statistical analysis tool that was used was the Independent Samples Mann-Whiney U test. The data showed distributions of Latency for OpenFlow and Ethernet were not similar, as assessed by visual inspection. However, Latency for OpenFlow (mean rank = 104.90) and Ethernet (mean rank = 96.10) were not statistically significantly different.;The data further showed distributions of Jitter for OpenFlow and Ethernet were similar, as assessed by visual inspection. Furthermore, median jitter for OpenFlow (.04400) and Ethernet (.03650) was not statistically significantly different. The data from this research suggests it is possible to automatically isolate production network resources for the purpose of disaster recovery exercises in a traditional Ethernet network using the open source version of the OpenDaylight SDN controller.