|Issue:||Asia-Pacific I 2009|
|Topic:||Testing the next generation data centre|
|Title:||Senior VP Product Development|
Victor Alston is the Senior VP Product Development at Ixia. Mr Alston oversees engineering, product strategy, product development, partnership and marketing activities for Ixia worldwide. He joined Ixia as Vice President of Engineering. Prior to that, Mr Alston was Senior Director of Engineering at SAP, a publicly traded enterprise software company. Mr Alston honed his entrepreneurial skills as a co-founder and Vice President of Engineering for two successful enterprise software companies that were later sold to Computer Associates and Oracle Corporation. Before starting those two companies, Mr Alston rose through the ranks at Oracle in the RDBMS and PLSQL group. Victor Alston received both his bachelor’s and master’s degrees in Computer Science from Stanford University.
Next generation data centres must simultaneously increase performance, efficiency and their capacity to deliver converged services while ensuring improved users experience, quality of service and return on investment. To meet these demands, next generation data centres increasingly rely upon higher speed Ethernet networks, virtualized servers, converged application servers and specialized network devices for load balancing, deep packet inspection and security. The resulting increase in the complexity of these centres challenges the ability of those deploying them to maintain an uninterrupted, high-performance operation.
Enterprises and service providers continuously face customer demands for additional services and capacity, but must simultaneously reduce their infrastructure and recurring costs – especially in the area of energy consumption. To meet these challenges, they have taken steps to develop next generation data centres characterized by: higher speed Ethernet networks; virtualized servers; converged application servers; and specialized network devices for load balancing, deep packet inspection and security. Next generation data centre operators face significant pre-deployment testing challenges to ensure uninterrupted, high-performance operation. A pre-deployment lab is the norm for most data centres, allowing exhaustive testing without affecting live network operation. Within the next generation data centre, multiple one Gigabit links have given way to a smaller number of ten Gigabit connections, and this holds true for external data centre connections as well. Larger data centres are finding multiple external ten G connections necessary to link to partner data centres, information suppliers and the Internet in general. The largest data centres are now looking for higher speed Ethernet connections to reduce the number of internal ten G networks that they operate. The IEEE 802.3ba task force1 is working toward defining standards for 40 and 100 Gigabit Ethernet networks in preparation for their expected deployment in the next decade. In fact, early pre-standard products have already been demonstrated and tested2. With the expansion of data centre capacity and the increased cost of power and air conditioning, controlling the energy consumption of servers and networking equipment has become essential. Many network vendors claim to have ‘green’, lower-power devices, but no standardized rating system that relates energy consumption to units of networking work accomplished yet exists. In order to compare apples to apples, one needs a metric that correlates energy used per Mbps of transmitted data. The ECR Initiative3 (the Energy Consumption Rating [ECR] Initiative) is one organization leading the way toward the definition of such a standard. The figure below describes the ECR measurements for a large-scale router. The ECR (Energy Consumption Rating) and ECRW (Energy Consumption Rating Weighted) values describe energy efficiency in terms of watts consumed per Gigabit per second of forwarded data. Server virtualization takes advantage of the substantial increase in processing power in modern CPUs to reduce the number of independent servers in the data centre. Virtual servers substantially decrease the infrastructure costs associated with rack space, power feeds and interfaces – especially network connections. This concentration, however, adds to test complexity. Virtual machine migration from one server blade to another, for example, will result in dropped connections or long pauses. Virtualization also results in a combination of real and virtual network switches, as shown below. It is essential that standard network performance indicators be measured for both the virtual and physical switch components. Distinct Fibre Channel networks have long been used to support storage area networks (SANs) but Fibre Channel over Ethernet (FCoE) – coupled with higher network speeds – has now made them an unnecessary duplication of infrastructure elements. This includes computer and switch interfaces as well as cable plants. FCoE is rapidly supplanting dedicated Fibre Channel SANs, allowing converged network and data storage traffic to be carried over the same Ethernet infrastructure. One of the prime drivers for data centre expansion has certainly been the convergence of voice, video and data services. These converged data services have resulted in both additional usage and bandwidth requirements. Each of the voice, video and data elements will have a unique set of requirements for minimum bandwidth and maximum permissible latency, jitter and loss. In most cases, these requirements can be satisfied by setting appropriate quality of service (QoS) values in the packet stream for the type of traffic. QoS values, however, have proven insufficient to distinguish the myriad of data services – some of which embody voice or video streams. More sophisticated devices, able to perform deep packet inspection (DPI), are needed to further prioritize data services and enforce security. Converged services require that a variety of voice, video and data servers work together using standard network protocols. Where multiple vendors are involved issues of interoperability can arise. These issues can be minimized through the use of conformance and system functional testing. Proper provisioning of data and network services requires accurate assessment of the data centre’s ability to handle multiple traffic flows of all types. This includes delivery of multiple services on multiple networks to multiple distinct user communities. A user community is characterized by a number of employees or customers using a set of services. Such communities often change their usage over time. For example, office workers are active from 9-5, service provider customers exhibit peak usage after school and in the evening, and enterprise batch jobs run in the early morning. Data centre capacity measurement requires modelling of these scenarios. All of these next generation data centre characteristics add up to a substantial set of requirements for network testing. The test platforms for next generation data centres must provide substantial test services, including: emulation of multiple user/subscriber communities to properly characterize capacity; ability to transmit and receive substantial network traffic through a variety of interfaces with a variety of speeds. Ethernet speeds to ten Gbps are essential; simultaneous emulation of FCoE network traffic; measurement of energy consumption in networking terms: watts per Mbps; ensure interoperability of network components through conformance testing; ensure proper prioritization of voice, video and data services under heavy load; and measure the overhead associated with server virtualization. Next generation data centres are simultaneously increasing performance, efficiency and converged service delivery while ensuring that users experience improved quality of service and increased return on investment. We are at a major inflection point in data centre technology, proving that next generation technology will deliver. Next generation data centres, however, create a substantial set of requirements for network test equipment vendors. The leading manufacturer of test systems for IP networks, have worked diligently to satisfy these requirements.