Bandwidth and Service Level Agreement What is Bandwidth? Bandwidth quantifies the data rate at which a network link or a network path can transfer. Network providers lease links to customers and usually charge based on bandwidth purchased. Service level agreements (SLAs) between providers and customers often define service in terms of available bandwidth at key interconnection (network boundary) points. Carriers plan capacity upgrades in their network based on the rate of growth of bandwidth utilization of their users. Bandwidth is also a key concept in content distribution networks, intelligent routing systems, end-to-end admission control, and video/ audio streaming. – The bandwidth of a link – The bandwidth of a sequence of successive links, or end-to-end path. – The maximum possible bandwidth a link or path can deliver (capacity) – The maximum unused bandwidth at a link or path (available bandwidth) – The achievable throughput of a bulk transfer TCP connection The Internet is largely a commercial infrastructure in which users pay for their access to an Internet Service Provider (ISP) and from there to the global Internet. It is often the case that the performance level (and tariff) of these network connections is based on their bit rate, or “network bandwidth,” since more bandwidth normally means higher throughput and better quality of service. Network operators commonly use tools such as MRTG to monitor the utilization of their links with information obtained from the router management software. These techniques are based on counters maintained by routers, and they are normally accurate. Users need to check whether they get the access bandwidth that they pay for and whether the network “clouds” that they use are sufficiently provisioned. ISPs also need bandwidth monitoring tools in order to plan their capacity upgrades and to detect congested or underutilized links. Bandwidth: parameters “C”:Capacity Number of bytes that can be sent over a link per unit of time “A”:Available Bandwidth Number of bytes that be sent over a link considering the current cross traffic “C-A”:Utilized Bandwidth Number of bytes that is sent over the link right now Definitions Path P a sequence of links from sender S to receiver R Capacity C minimum transmission rate among all links in P Available bandwidth A minimum available bandwidth among all links in P Narrow link the link with minimum capacity Tight link the link with minimum available bandwidth Bandwidth related Metrics Capacity – individual links and end-to-end paths Available Bandwidth – individual links and end-to-end paths Bulk Transfer Capacity (BTC) – End-to-end paths Links at the data link layer (layer 2) Segments a physical point-to-point link a virtual circuit a shared access local area network Links at the IP layer (layer 3) Hops a sequence of one or more segments, connected through switches, bridges, or other layer 2 devices. Capacity A layer 2 link, or segment, can normally transfer data at a constant bit rate, which is the transmission rate of the segment. – 10 Mb/s on a 10BaseT Ethernet segment – 2.048 Mb/s on a E1 segment – The transmission rate of a segment is limited by both the physical bandwidth of the underlying propagation medium as well as its electronic or optical transmitter/receiver hardware. At the IP layer a hop delivers a lower rate than its nominal transmission rate due to the overhead of layer 2 encapsulation and framing. The IP layer capacity depends on the size of the IP packet relative to the layer 2 overhead. The nominal capacity of a segment CL2 IP packet of size LL3 bytes The transmission time for an IP packet Δ L3 The total layer 2 overhead (in bytes) HL The capacity of that segment at the IP layer CL3 The capacity the hop can deliver to the IP layer is 7.24 Mb/s for 100-byte packets, and 9.75 Mb/s for 1500-byte packets. CL2 is 10 Mb/s and HL2 is 38 bytes (18 bytes for the Ethernet header, 8 bytes for the frame preamble, and the equivalent of 12 bytes for the interframe gap) The capacity of a hop as the bit rate, measured at the IP layer, at which the hop can transfer MTU-sized IP packets. The capacity of the ith hop Ci The Number of hops in the path H The hop with the minimum capacity is the narrow link on the path. The fraction of segment capacity delivered to the IP layer, as a function of packet size. Available Bandwidth The available bandwidth of a link relates to the unused or spare capacity of the link during a certain time period. Ai = (1 – u i )Ci Ci the capacity of hop i ui the average utilization of that hop in the given time interval T he average available bandwidth Ai of hop I A = min Ai The hop with the minimum available bandwidth is called the tight link of the end-to-end path i= 1, …,H Even though the capacity of a link depends on the underlying transmission technology and propagation medium, the available bandwidth of a link additionally depends on the traffic load at that link, and is typically a time-varying metric. Instantaneous utilization for a link during A pipe model with fluid traffic for a three hop a time period (0,T). network path Since the average available bandwidth can change over time, it is important to measure it quickly. In contrast, the capacity of a path typically remains constant for long time intervals (e.g., until routing changes or link upgrades occur) The capacity limiting link narrow link The available bandwidth limiting link tight link. TCP Throughput TCP is the major transport protocol in the Internet, carrying almost 90 percent of the traffic Several factors may influence TCP throughput transfer size type of cross traffic (UDP or TCP) number of competing TCP connections TCP socket buffer sizes at both sender and receiver sides congestion along the reverse path the size of router buffers capacity and load of each link in the network path Variations in the specification and implementation of TCP use of selective ACKs vs. cumulative ACKs selection of the initial window size The throughput of a small transfer such as a typical Web page primarily depends on the initial congestion window, round-trip time (RTT), and slow-start mechanism of TCP, rather than on available bandwidth of the path. The throughput of a large TCP transfer over a certain network path can vary significantly when using different versions of TCP even if the available bandwidth is the same. Bulk Transfer Capacity BTC is the maximum throughput obtainable by a single TCP connection. The connection must implement all TCP congestion control algorithms The BTC and available bandwidth are fundamentally different metrics. BTC is TCP-specific, whereas the available bandwidth metric does not depend on a specific transport protocol. The BTC depends on how TCP shares bandwidth with other TCP flows, while the available bandwidth metric assumes that the average traffic load remains constant and estimates the additional bandwidth a path can offer before its tight link is saturated. Bandwidth Estimation Techniques variable packet size (VPS) the capacity of individual hops probing, packet pair/train dispersion (PPTD) end-to-end capacity self-loading periodic streams (SLoPS) end-to-end available bandwidth. trains of packet pairs (TOPP) end-to-end available bandwidth. VPS – to measure the RTT from the source to each hop of the path as a function of the probing packet size. – VPS uses the time-to-live (TTL) field of the IP header to force probing packets to expire at a particular hop. – The router at that hop discards the probing packets, returning ICMP time- exceeded error messages back to the source. The source uses the received ICMP packets to measure the RTT to that hop. – The RTT to each hop consists of three delay components in the forward and reverse paths: serialization delays, propagation delays, and queuing delays – VPS probing may yield significant capacity underestimation errors if the measured path includes store-and- forward layer 2 switches – Such devices do not generate ICMP TTL-expired replies because they are not visible at the IP layer. Variable Packet Size method Per-Hop capacity “C” estimation Use of IP TTL field Receive ICMP Time Exceeded Problems: Level 2 store-and-forward devices Variation in ICMP generation delays More … Bandwidth Estimation Techniques Packet Pair/Train Dispersion Probing (PPTD) – The source sends multiple packet pairs to the receiver. – Each packet pair consists of two packets of the same size sent back to back. – The dispersion of a packet pair at a specific link of the path is the time distance between the last bit of each packet. – PPTD probing techniques typically require double-ended measurements, with measurement software running at both the source and the sink of the path. – monitoring variations in the one-way delays of the probing packets. Self-Loading Periodic Streams (SLoPS) – The source sends a number K ≈ 100 of equal-sized packets (a periodic packet stream) to the receiver at a certain rate R – The sender attempts to bring the stream rate R close to the available bandwidth A, following an iterative algorithm similar to binary search. – The sender probes the path with successive packet trains of different rates, while the receiver notifies the sender about the one-way delay trend of each stream. – The sender also makes sure that the network carries no more than one stream at any time. Trains of Packet Pairs (TOPP) – TOPP sends many packet pairs at gradually increasing rates from the source to the sink. – TOPP assumes that the packet pair will arrive at the receiver with the same rate it had at the sender Bandwidth Estimation Tools (Publicly available) Pathchar Algorithm sending packets of varying sizes and measuring their round trip time. The pathchar program uses an active algorithm that sends packets varying in size from 64 bytes to the path MTU with a stride of 32 bytes. The number of different packet sizes pa/thchar sends is s = [MTU/32] – 1 For Ethernet , the MTU Size 1500 bytes, s is 45 it sends p packets per size for every hop. In the default configuration, p = 32. It must wait for each packet it sends to be acknowledged before sending the next packet. It must wait for each packet it sends to be acknowledged before sending the next packet. For a 10-hop Ethernet network with an average round trip latency of l0 ms, pathchar would run in 144 seconds. The average bandwidth used for probing a particular hop is average packet size/ round-trip latency pathchar will send 10 MB of data on a 10-hop network regardless of the bandwidth of the network, since it only depends on the number of hops, the path MTU, and p. If the path MTU is high and one of the early hops is a low bandwidth network link, such as a 56K modem, then pathchar can consume most of the bandwidth of that link for an extended amount of time. SLA A common complaint about the Internet is that it is slow. Some of this slowness is due to properties of the end points, like slow servers, but some is due to properties of the network, like propagation delay and limited bandwidth. Propagation delay can be measured using widely deployed and well understood algorithms implemented in tools like ping and traceroute. Today’s Internet only provides best-effort service, where the network treats all traffic in exactly the same way. Traffic is processed as quickly as possible, but there is no guarantee as to timeliness or actual delivery. When the load level is low, the network delivers a high quality service. The best-effort Internet does not deny entry to traffic, so as the load levels increase, the network congestion levels increase, and service-quality levels decline uniformly. No matter how much bandwidth the networks can provide, new applications will be invented to consume them; therefore, mechanisms will still be needed to provide QoS. Even if bandwidth will eventually become abundant and cheap, it is not going to happen soon. For now, some simple mechanisms are definitely needed in order to provide QoS on the Internet. In order for a customer to receive differentiated services from its Internet service provider (ISP), it must have a service level agreement (SLA) with its ISP. An SLA basically specifies the service classes supported and the amount of traffic allowed in each class. The more competitive a particular service’s market, the more comprehensive and stringent, or tight, the commitments or service level agreements (SLAs) offered for the service. An SLA can be static or dynamic. Static SLAs are negotiated on a regular (e.g., monthly or yearly) basis. Customers with dynamic SLAs must use a signaling protocol (e.g., RSVP) to request services on demand. SLA commitments are based on delay, jitter (delay variation), packet loss rate, throughput, availability, and per-flow sequence preservation. SLA A service level agreement (SLA) is a contract between an Internet service provider (ISP) and its customer. SLAs obligate service providers to maintain a certain grade or level of service. Service providers are keen on offering SLAs because SLAs permit differential treatment of the customer traffic. – Economic benefit to the service providers – the customer wants SLA guarantees because they can ensure the rigid level of performance they pay for, and be compensated for the lack thereof. – Customers can choose the level that suits their need, and not have to pay a premium for unnecessary features The guarantee is usually limited to within the boundary of a single ISP only. the maximum bandwidth the minimum connection availability – The connection availability is defined as the fraction of time a connection is capable of transferring data between both end points. Asymmetric bandwidth – Downlink and uplink bandwidth are normally different. – Downlink bandwidth is larger than uplink bandwidth in general. Service Classification At the ingress of the ISP networks, packets are classified, policed, and possibly shaped. The SLA includes: Classification Policing Shaping rules used at the ingress routers The amount of buffering space needed for the above operations – When a packet enters one domain from another domain, its DS field may be re-marked as determined by the SLA between the two domains. Premium service – for applications requiring low-delay and low-jitter service Assured service – for applications requiring better reliability than best-effort service Olympic service, which provides three tiers of services: Gold Silver Bronze with decreasing quality Differentiated services only defines DS fields and PHBs (Per-Hop Behaviors) It is ISPs’ responsibility to decide which services to provide Assured service is intended for customers that need reliable services from their service providers, even in times of network congestion. Customers will have SLAs with their ISPs. The SLAs will specify the amount of bandwidth allocated for the customers. Customers are responsible for deciding how their applications share that amount of bandwidth. SLAs for assured service are usually static, meaning that customers can start data transmission whenever they want without signaling their ISPs. Aggregate over provisioning of bandwidth represents an expensive option for the service provider and can be difficult to ensure in all cases. Core Service Classes and SLA specification Common among QoS-enabled IP services that SPs offer today is support for service classes designed to meet the needs of three aggregate traffic types. Real-time This class targets applications such as VoIP and video. SPs define service for this class in terms of low delay and jitter (typically less than 5 ms within the backbone), and close to zero loss. The class might include a commitment for per-flow sequence preservation. Business data This class represents business critical interactive applications such as IBM’s System Network Architecture; Systems, Applications, and Programming Facilities’ real-time system, version three (SAP R/3); Telnet; and possibly intranet Web applications. SPs define service for this class in terms of defined delay and close to zero loss. The class might include a commitment for per flow sequence preservation. Standard. This class represents all traffic not classified as real-time or business. SPs define service for this class in terms of a loss rate; it might also include a commitment for per-flow sequence preservation. Because delay and jitter are unimportant for this service, they are not defined Core Service Classes and SLA specification For real-time traffic, the SP uses a strict priority-queuing behavior to ensure the lowest delay and jitter service. Once this class is served, the SP allocates the remaining bandwidth, with a minimum assurance of 90 percent going to the business data class and 10 percent to the standard class. Because the real-time and business data loads are expected to be less than their available class capacity, these classes effectively experience zero loss. A holistic per-class capacity-planning process is essential to ensure that this is actually the case. The capacity-planning process might take into account single or multiple network component (link and node) failures, depending on the SP’s particular goals. Assuming the routers use a work-conserving scheduler, the standard class can reuse all unallocated or unused interface capacity once the real-time and business data classes have been serviced. when congestion occurs the loss is restricted to the standard class, thereby assuring the SLAs for the real-time and business classes. Classification and policing are done at the ingress routers of the ISP networks. If the assured service traffic does not exceed the bit rate specified by the SLA, they are considered in profile; otherwise, the excess packets are considered out of profile. all packets, in and out, are put into an assured queue (AQ) to avoid out of order delivery. Third, the queue is managed by a queue management scheme called random early detection (RED) with In and Out - RIO RED is a queue management scheme that drops packets randomly. This will trigger the TCP flow control mechanisms at different end hosts to reduce send rates at different time. By doing so, RED can prevent the queue at the routers from overflowing, and therefore avoid the tail-drop behavior (dropping all subsequent packets when a queue overflows). Tail-drop triggers multiple TCP flows to decrease and later increase their rates simultaneously. It causes network utilization to oscillate and can hurt performance significantly CORVIL Bandwidth BANDWIDTH ALLOCATION IN IP NETWORKS Allocating adequate bandwidth is necessary to provide the network performance required for applications. Corvil Bandwidth -a way to determine the minimum bandwidth required to deliver traffic within customer-specified quality of service (QoS) targets, with statistical reliability Statistical Reliability Relation between a network performance target and the resources required to meet it and quantify that target. Propagation and serialization delay in a network, and some packet loss due to bit errors on links As all applications can tolerate some level of loss, provision the network to provide a level of performance commensurate with application needs. No more than one packet out of every 10,000 will be dropped. No more than one packet out of every 1000 will be delayed by more than 20 ms. Bandwidth, Statistical Multiplexing, and QoS Mechanisms Bandwidth - routers and switches - Simple Network Management Protocol (SNMP) MIBs that offer traffic statistics such as the average bit rate over five-minute windows. Loss and jitter levels that packet traffic experiences on the bit rates in the traffic at the ms level Statistical Multiplexing - The difference between the bandwidth requirement of the aggregate and the sum of the per-stream bandwidth requirements In a circuit-switched network, each stream needs a separate circuit The statistical sharing of resources happens automatically in packet-based networks QoS Mechanisms Shaping and policing provide separation between different service instances priority and weighted-fair schedulers allow the services to share bandwidth efficiently The Fundamental Equation of Network Quality Network bandwidth, traffic load, and QoS goals are intrinsically linked. Changes in one affect the relationship between the other two. For example, the bandwidth required to meet a delay target depends not only on the load on the network, but also on whether it carries VoIP, video, or data traffic. What quality does the network provide to the traffic it carries? Quality = fQ(Network, Traffic) – What network resources are needed for the traffic to achieve the quality it requires? Bandwidth = fB(Traffic, Quality) – how much traffic can be carried before the resulting quality degrades excessively? Traffic = fT(Network, Quality) CORVIL BANDWIDTH FOR IP NETWORKS Bandwidth = fB(Model, Quality) The Corvil Traffic Descriptor (CTD) is a compact encoding of the distribution of bit and packet rates in a traffic aggregate over any given time window. This descriptor is all that is needed to calculate the resource requirements of packet traffic. Hybrid fiber coaxial (HFC) technology At the head end of the network, signals from various sources, such as traditional satellite services, analog and digital services using WANs, and internet service provider (ISP) services using a backbone network, are multiplexed and up-converted from electrical radio frequency (RF) to an optical signal. The signal is brought to a fiber node via a pair of optical fibers, where communication is one way on the optical fiber, and then distributed via a single coaxial cable to the customer premises. The mode of transmission over the coaxial cable is duplex. a pair of optical fibers from the cable modem termination system (CMTS) at the head end to the fiber node, each carrying one- way traffic in opposite directions. The optical signal is down-converted to RF at the fiber node and travels over the coaxial cable in a duplex mode. The coaxial cable is a shared access medium and designed to carry signals up to tens of kilometers by amplification of signal in both directions. A duplex mode of communication is achieved by transmitting the downstream signal in high-frequency band, 50– 860 MHz, and the upstream signal in the low-frequency band of 5–42 MHz. The downstream signal includes analog cable television spectrum. The signal that goes from head end to the customer premises is called the downstream signal and is transmitted in broadcast mode. The signal traversing from the customer premises to the head end is called the upstream signal. It is not in broadcast mode, but is generally a combination of time division multiplexing and random access protocol. At the customer premises, there is a network interface unit (NIU), which is the demarcation point between the customer network and service provider network. The RF signal is split at NIU. The TV signal is directed to the TV monitor and the multimedia signal to the cable modem. The cable modem converts the multimedia RF signal to Ethernet output. Data, voice- over-IP, and IPbased video stream services are carried as multimedia signals and support multimedia services to the subscribers. The multimedia signal is carried over an analog channel of 6 MHz in the downstream and upstream directions. Based on the volume of traffic, multiple channels could be assigned to carry multimedia services.