Manageable Data Center Internet on Wide Area Networks
By Daniel Yu, Enterprise Networking Product Mgmt Dept, Huawei Enterprise Business Group
Huge traffic increases in data centers require high-performance data processing, fueling the deployment of new data centers and the integration of small- and medium-sized data centers into centralized computing and storage resources. These new data centers have higher bandwidth requirements, more complicated Internet architectures, and require more complex service bearer management. Also, the "cloud" technique used for IT resource reintegration, distribution, and delivery extends to the construction of different enterprise data centers. Therefore, the problem of how to eliminate area differences, ensure service continuity, and simplify network monitoring becomes very important.
The technologies and protocols deployed in existing data center networks have been developed over the last 30 years where all network intelligence addressing switches or routers is governed by more than 6,000 protocols. New requirements are met with yet another protocol added to the stack. Understandably, the deployment of new services was a constant challenge. SDN resolves this issue by making networks directly programmable from a separate control plane. SDN-based solutions for controlling WAN traffic and simplifying network monitoring will have widespread application.
SDN deployments for data center networks have increased resource utilization, enabled greater traffic volume, and reduced operating expenses. The trade-off is an increased requirement for network monitoring to guarantee service stability.
The Huawei solution proposes full-path management and full-network optimization, rather than partial-FIB (Forwarding Information Base) path optimization, by using network virtualization on cloud platforms for complex, multi-tenant, multi-path networks. The proposed data center Internet is designed to solve the problems caused by high traffic volumes over complicated topologies. In terms of network monitoring, Huawei promotes the unification of quality metrics and fault location into service flows with the goal of implementing real-time monitoring and service quality reports. In turn, we expect to eliminate point-by-point service deployments, segment-by-segment monitoring, and simulated service measurements.
SDN Traffic Control on the Optical-Transmission-Based Data Center Internet
Most data centers using optical transmission will employ Automatically Switched Optical Network (ASON) devices. The ASON optical and electrical layers are most often separated; however, in configurations such as Shared Risk Link Group (SRLG) and distance, they are often dependent upon each other. If and when the two layers fail to communicate, tens of thousands of SRLGs must be manually configured at great expense and potential risk of error. This manual reservation operation is complicated and inflexible.
Path Computation Elements (PCEs) have been introduced to link the optical and electrical layers and calculate multi-layer end-to-end paths. PCE results deliver end-to-end path configurations to each transmission network element such that all devices on an optical network have received a single set of instructions. A PCE central controller provides standard User Network Interfaces (UNIs) to implement communication between Internet Protocol (IP) devices (i.e. routers) and optical networks. The Path Computation Element Protocol (PCEP) is used to communicate between the PCE central controller and the transmission network elements. PCEP is used to detect and report network resource status in real time, and delivers computation results.
SDN Traffic Control Implementation on the IP-Based Data Center Internet
An IP-based data center Internet uses Layer 2 or 3 networking, and will run a traditional IP signaling protocol, such as Border Gateway Protocol (BGP), Open Shortest Path First (OSPF), or Reservation Protocol-Traffic Engineering (RSVP-TE) over the control plane. Each network element uses its own IP FIB table to compute the shortest-path next hop in order to create a complete forwarding path. This path computation method is sub-optimal on networks with large numbers of service flows as the consumption of bandwidth for the forwarding path calculation leads to improper flow distribution with some links congested and others idle. Manual configurations, time-consuming and difficult, are not best choice for a quick resolution. As a result, deploying standalone computation elements becomes the only solution.
The Huawei solution substitutes stand-alone PCEs for traditional routing protocols. PCEs collects information for each network element in real time (i.e. delay, jitter, remaining bandwidth) and allow users to configure time periods, service attributes, and other parameters. The PCE then computes an optimal path using actual conditions. After path computation, the PCE uses the PCEP protocol to deliver the result to all other network elements. After a PCE is deployed, a network element's control plane only needs to receive control information from the PCE, so it will be able to concentrate on rapid and efficient data forwarding.
Service-Flow-Based Network Monitoring
Live networks currently use simulation tests to monitor and detect network quality. This method is implemented by deploying probes on each network element and advertising probe packets to the entire network, and is used to monitor single and specific packet types. For example, the Y.1731 technique is dedicated to Ethernet VLAN (Virtual Local Area Network) packets, RFC 6375 is dedicated to MPLS-TP packets, and RFC 6374 is dedicated to MPLS LSP packets. Therefore, the simulation test method does not apply to multipoint-to-multipoint or Equal-Cost Multi-path (ECMP) scenarios – and, in many cases, is insufficient to provide accurate monitoring results or fails to monitor network quality. In addition, the simulation test method encounters many problems on hierarchical networks (such as L2 + L3) and cannot simultaneously monitor primary and backup links.
With ten years of experience in IP device development, Huawei proposes its proprietary Packet Conservation Algorithm for Internet (iPCA) technique. iPCA can implement proactive and real-time quality detection, as well as real-time fault locating. iPCA is capable of labeling, measuring, and collecting statistics on real service packets, thus preventing the accuracy issues found in simulation tests. iPCA also supports IP packets and can distinguish between different types of encapsulated packets and thus collect statistics on a specific type of service packet. These advantages enable iPCA to be applied to various tunnel encapsulation scenarios and networks. With iPCA enabled on a network, all devices on the network periodically report statistics to the central control unit, and all devices use the uniform packet format to report statistics. The central control unit can then calculate packet loss and jitter for the entire network instantly, improving fault isolation efficiency.
Using SDN to control traffic and monitor network quality is just the first step for deploying SDN on the data center Internet. The development of SDN and controller technologies acts as a foundation for the development of all-service-based traffic control and flexible traffic scheduling and management.