FibeAir IP-10: Enhancing Burst Resiliency in TCP-based LTE Networks, Apuntes de Ingenieria Eléctrica

¡Descarga FibeAir IP-10: Enhancing Burst Resiliency in TCP-based LTE Networks y más Apuntes en PDF de Ingenieria Eléctrica solo en Docsity! TECHNICAL BRIEF | FibeAir IP-10 Burst Resiliency September 2012 1 FibeAir IP-10 Burst Resiliency Introduction Demand for greater data rates continues to increase unabated. Operators and users must be aware of several limitations to meeting this demand. The most important of these is to shape the bandwidth offered to the network to match the bandwidth profile of the service level agreement (SLA). While we want to think that pushing higher data rates equates to greater throughput, this is not always the case. In fact, forcing on an Ethernet service a data rate that is greater than the service’s capacity will result in a poor user experience as packets may be dropped and re-transmissions may be required resulting in net throughputs much lower than expected. In Long Term Evolution (LTE) networks running the TCP protocol, instantaneous congestion issues can arise frequently resulting in instantaneous, large bursts that contribute further to throughput reduction. Ceragon has developed a diverse and flexible solution to instantaneous bursts within its FibeAir® IP-10 enhanced traffic manager. The Enhanced QoS innovation employs Weighted Random Early Detect (WRED) and Configurable Buffer Lengths to make bandwidth use as efficient as possible in order to deliver the best possible user experience. The Challenge In some cases, users perceive that they are receiving lower throughput than they expect. In actuality, many IP and Application Layer factors affect the network operator’s efficiency when utilizing an Ethernet service and most of these are under their own direct control. Intelligent use of the FibeAir IP-10 alleviates these problems. TECHNICAL BRIEF | FibeAir IP-10 Burst Resiliency September 2012 2 The most important problem discussed in this paper is rate adaptation—connecting a network element line, with a high data rate limited only by physical interface and cabling, to a radio interface where the data rate is limited by the channel bandwidth (spectrum efficiency) resulting in the drop of frames. As a consequence, the overall capacity is reduced. An excellent example of the need for addressing rate adaptation scenarios that are characterized by bursty traffic is in LTE networks. LTE technology was developed with the objective of high data rates, low-latency and packet-optimized radio access technology. Level Standard Uplink (bps) Downlink (bps) 2G GSM 9.6k 14.4k 2.5G Edge 384.k 513k 3G UMTS 384k 2.0M 3G+ HSDPA/HSUPA 5.8M 14.4M 3G+ HSPA+ 11M 42M 4G LTE 50M 100M Table 1. The Peak Data Rate of Various Mobile Standards Even though it was not originally designed for real-time applications, TCP is the most widely used protocol for wire and wireless networks. In its early days, TCP was applied to wired networks, but later showed poor performance over wireless channels mainly due to high error rates. In order to compensate for these errors, LTE employs many error recovery techniques at the Link Layer; these partially overlap with error recovery performed at the Transport Layer of TCP/IP. TECHNICAL BRIEF | FibeAir IP-10 Burst Resiliency September 2012 5  TCP burst size grows linearly as the bit rate and latency increase. However, large bursts are never recommended as they may affect other flows that will need to wait for the whole burst to be transmitted.  As the number of TCP flows increases, network efficiency improves since, when several TCP flows share the same limited bandwidth (radio), the throughput of each flow is smaller and likewise the burst length that it can create. In addition, when there are several TCP flows, statistically, the timing of transmit and receive of each flow is different so bursts don’t necessarily occur at the same time. As the number of TCP flows decreases, the effect of large bursts and concomitant congestion increases. The worst case of inefficiency due to congestion is when only one TCP flow is used. FibeAir IP -10 handles the bursty traffic using the enhanced traffic manager as described at the next section. Ceragon’s Solution for Overcoming Burst Situations Shaping the traffic prior to injecting it into the network element is always the preferable solution and is an MEF (10.2) recommendation. Since this is not always possible, Ceragon implements a solution based on the FibeAir IP-10 Enhanced Traffic Manager feature marketed as Enhanced QoS. Enhanced QoS provides several benefits using its policer, classification, advanced scheduling, monitoring and other techniques. These and other IP-10 enhanced QoS features enable operators to provide differentiated services with strict SLA while maximizing network resource utilization. Two Enhanced QoS features are directly related to overcoming burst problems. WRED Support Used with tail-drop for congestion management, the Weighted Random Early Detect (WRED) mechanism can increase capacity utilization of TCP traffic by eliminating the phenomenon of global synchronization. Global synchronization occurs when TCP flows sharing bottleneck conditions receive loss indications at around the same time. This can result in periods during which link bandwidth utilization drops significantly as a consequence of simultaneous falling to ‘slow start’ of all the TCP flows. WRED eliminates the occurrence of traffic congestion peaks by restraining the transmission rate of the TCP flows. Each queue occupancy level is monitored by the WRED mechanism and randomly selected frames are dropped before the queue becomes overcrowded. TECHNICAL BRIEF | FibeAir IP-10 Burst Resiliency September 2012 6 Eight Priority Queues with Configurable Buffer Length The handling of burst effects, usually for best-effort class of service, is performed using FibeAir IP-10’s advanced large buffer allocation mechanism which allows specification of the buffer size of each of the eight priority queues according to the specific service. The total amount of memory dedicated for the queue buffers is 4Mb while the size of each queue can be set in granularities of 0.5Mb. The default is 0.5Mbit for each queue. The buffer size must be calculated carefully and based on two considerations:  Latency – If low latency is required (dropping frames in the queue is preferable to increasing latency), small buffer sizes are preferred. The actual effective buffer size of the queue can be less than 0.5Mbit by setting WRED tail-drop curve.  Throughput immunity to fast bursts – When the traffic is characterized by fast bursts, it is recommended to increase the buffer sizes of the priority queues to prevent packet loss. Of course, this will come at the expense of a possible increase in the latency. Burst sizes can be configured as a tradeoff between the latency and immunity to bursts according application requirements. Practical Example The first step is to calculate the maximum buffer size according to the following formula: (1) Buffer Size [Bytes] = (Flow Bandwidth [bps] * Link Delay [sec]) / 8 The example considers a typical case of LTE service and demonstrates the use of formula (1). The example shows how to buffer the TCP data while maintaining different services. Common values for LTE service suggest: (2) Link Delay < 20ms, TCP bandwidth = 100Mbps Applying the maximum latency scenario, substituting (2) into (1) yields: (3) Buffer Size = 250KBytes The large buffer size calculated in (3) is easily supported by the IP-10’s buffers which can accommodate up to 4Mbps or burst lengths of 500KB. Planning for burst handling is done according to Enhanced QoS: TECHNICAL BRIEF | FibeAir IP-10 Burst Resiliency September 2012 7  A 250KB buffer size will be allocated to the TCP flow to fill the data stream (lowest priority queue).  The remaining 250KB can be allocated between up to four queues (with higher priorities) and may be dedicated for management protocols, synchronization packets, Pseudowire services (voice & signaling), etc. Synchronization & management 0.5Mbit buffer Voice and Signaling 0.5Mbit buffer Highest priority Low priority Other 1Mbit buffer Data (TCP) 2Mbit buffer Figure 3: Granularity of the Buffer Allocation Summary Bursty traffic is an emerging impediment to maximizing throughput in multi-carrier LTE- Advanced networks that employ the TCP protocol. In this paper, we defined the nature of the problem of bursts and explained how they occur in TCP. Ceragon has solved the problem in its FibeAir IP-10 family with its innovative technology called Enhanced QoS which allows network planners to handle any type of TCP-oriented traffic in order to maximize the utilization of network resources.