QoS MechanismsQoS 機制

QoS mechanisms refer to the set of tools and techniques to manage network resources and are considered the key enabling technology for network convergence. The objective of QoS mechanisms is to make voice, video and data convergence appear transparent to end users. QoS mechanisms allow different types of traffic to contend inequitably for network resources. Voice, video, and critical data applications may be granted priority or preferential services from network devices so that the quality of these strategic applications does not degrade to the point of being unusable. Therefore, QoS is a critical, intrinsic element for successful network convergence.

QoS 機制是指用來管理網路資源的一套工具與技術，被視為實現網路匯流（convergence）的關鍵技術。QoS 機制的目標是讓語音、視訊與資料的匯流對終端使用者而言呈現透明化。QoS 機制允許不同類型的流量以不公平的方式競爭網路資源。語音、視訊及關鍵資料應用程式可從網路裝置獲得優先或優先服務，使這些策略性應用程式的品質不至於降級到無法使用的程度。因此，QoS 是成功實現網路匯流的關鍵且內在的元素。

Cisco network devices can provide a complete toolset of QoS features and solutions for addressing the diverse needs of voice, video and multiple classes of data applications. QoS mechanisms allow complex network control and predictable service for a variety of networked applications and traffic types. They can effectively control bandwidth, delay, jitter, and packet loss. By ensuring the desired results, the QoS mechanisms lead to efficient, predictable services for business-critical applications.

Cisco 網路裝置可提供完整的 QoS 功能與解決方案工具集，以滿足語音、視訊及多種資料應用程式類別的多元需求。QoS 機制能為各種網路應用程式與流量類型提供複雜的網路控制與可預測的服務。它們可有效控制頻寬、延遲、抖動與封包遺失。透過確保預期結果，QoS 機制為關鍵業務應用程式帶來高效且可預測的服務。

In any network in which networked applications require differentiated levels of service, traffic must be sorted into different classes upon which QoS is applied. Classification and marking are two critical functions of any successful QoS implementation. Classification allows network devices to identify traffic as belonging to a specific class with specific QoS requirements, as determined by an administrative QoS policy. After network traffic is sorted, individual packets are marked (also called colored) so that other network devices can apply QoS features uniformly to those packets in compliance with the defined QoS policy.

在任何需要為網路應用程式提供差異化服務等級的網路中，流量必須先分類為不同的類別，才能套用 QoS。分類與標記是任何成功 QoS 實作的兩項關鍵功能。分類讓網路裝置能依管理員制定的 QoS 政策，識別出流量屬於具有特定 QoS 需求的特定類別。流量分類完成後，個別封包會被標記（也稱為著色），使其他網路裝置能依已定義的 QoS 政策，對這些封包一致地套用 QoS 功能。

A classifier is a tool that inspects packets within a flow to identify the type of traffic that the packet is carrying. Traffic is then marked so that a policy enforcement mechanism will implement the policy for that type of traffic.

分類器（classifier）是一種工具，用來檢查流量中的封包，以識別封包所承載的流量類型。流量經識別後會被標記，以便政策執行機制能對該類型的流量執行相對應的政策。

Commonly used traffic descriptors include CoS at layer 2, incoming interface, IP Precedence, Differentiated Services Code Point (DSCP) at layer 3, source or destination address, and application. After the packet has been classified, the packet is then available for QoS handling on the network.

常用的流量描述符包括第 2 層的 CoS、輸入介面、IP 優先權（IP Precedence）、第 3 層的差異化服務代碼點（DSCP），以及來源或目的位址、應用程式。封包經分類後，即可在網路中進行 QoS 處理。

Using packet classification, you can partition network traffic into multiple priority levels or classes of service. When traffic descriptors are used to classify traffic, the source agrees to adhere to the contracted terms and the network promises a QoS. Different QoS mechanisms, such as traffic policing, traffic shaping, and queuing techniques, use the classification of the packet to ensure adherence to this agreement.

透過封包分類，你可以將網路流量劃分為多個優先等級或服務類別。當使用流量描述符來分類流量時，來源方同意遵守約定的條款，網路則承諾提供對應的 QoS。不同的 QoS 機制，例如流量管制、流量整形與佇列技術，都會使用封包的分類結果來確保遵守此協議。

Classification should take place at the network edge, typically in the wiring closet, in IP phones or at network endpoints. It is recommended that classification occur as close to the source of the traffic as possible.

分類應在網路邊緣進行，通常是在配線間、IP 電話或網路端點上。建議分類盡可能靠近流量來源進行。

The concept of trust is key for deploying QoS. When an end device (such as a workstation or an IP phone) marks a packet with CoS or DSCP, a switch or router has the option of accepting or not accepting values from the end device. If the switch or router chooses to accept the values, the switch or router trusts the end device. If the switch or router trusts the end device, it does not need to do any reclassification of packets coming from this interface. If the switch or router does not trust the interface, it must perform a reclassification to determine the appropriate QoS value for the packets coming from this interface. Switches and routers are generally set to not trust end devices and must specifically be configured to trust packets coming from an interface.

信任（trust）的概念是部署 QoS 的關鍵。當終端裝置（例如工作站或 IP 電話）以 CoS 或 DSCP 標記封包時，交換器或路由器可以選擇接受或不接受來自該終端裝置的值。若交換器或路由器選擇接受這些值，即表示交換器或路由器信任該終端裝置。若交換器或路由器信任該終端裝置，就不需要對來自該介面的封包進行重新分類。若交換器或路由器不信任該介面，就必須執行重新分類，以判斷來自該介面封包的適當 QoS 值。交換器與路由器通常預設不信任終端裝置，必須特別設定才能信任來自某介面的封包。

Marking is related to classification and allows network devices to classify a packet or frame, based on a specific traffic descriptor.

標記與分類相關，讓網路裝置能依特定流量描述符對封包或框架進行分類。

Marking a packet or frame with its classification allows network devices to easily distinguish the marked packet or frame as belonging to a specific class. After the packets or frames are identified as belonging to a specific class, other QoS mechanisms can use these markings to uniformly apply QoS policies.

以類別標記封包或框架，可讓網路裝置輕易分辨該封包或框架屬於特定類別。封包或框架被識別為屬於特定類別後，其他 QoS 機制即可利用這些標記一致地套用 QoS 政策。

Ultimately, there are various Layer 2 and Layer 3 mechanisms that are used in the network for marking traffic.

最終，網路中會使用多種第 2 層與第 3 層機制來進行流量標記。

Layer 3 packet marking with IP Precedence and DSCP is the most widely deployed marking option because Layer 3 packet markings have end-to-end significance. Layer 3 markings can also be easily translated to and from Layer 2 markings.

使用 IP 優先權與 DSCP 進行的第 3 層封包標記，是最廣泛部署的標記方式，因為第 3 層封包標記具有端到端的意義。第 3 層標記也可輕易與第 2 層標記互相轉換。

DSCP is encoded in the header of both IPv4 and IPv6 packets.

DSCP 編碼於 IPv4 與 IPv6 封包的標頭中。

DiffServ uses the DiffServ (DS) field in the IP header to mark packets according to their classification. The DS field occupies the eight-bit ToS field in the IPv4 header or the Traffic Class field in the IPv6 header.

DiffServ 使用 IP 標頭中的 DiffServ（DS）欄位，依分類結果標記封包。DS 欄位佔用 IPv4 標頭中 8 位元的 ToS 欄位，或 IPv6 標頭中的流量類別欄位。

Within a network, different forms of connectivity can have significantly different costs for an organization. Because WAN bandwidth is relatively expensive, many organizations would like to limit the amount of traffic that specific applications can send. This is especially true when enterprise networks use internet connections for connectivity to remote sites and the extranet. Downloading nonbusiness-critical images, music, and movie files can greatly reduce the amount of bandwidth that is available for other mission-critical applications. Traffic policing and traffic shaping are two QoS techniques that can limit the amount of bandwidth that a specific application, user, or class of traffic can use on a link.

在網路中，不同形式的連線對組織而言可能有截然不同的成本。由於 WAN 頻寬相對昂貴，許多組織希望限制特定應用程式可傳送的流量。在企業網路使用網際網路連線連接遠端站點與外部網路時，這一點尤其重要。下載非業務關鍵的圖片、音樂與影片檔案，可能會大幅減少其他關鍵業務應用程式可用的頻寬。流量管制（policing）與流量整形（shaping）是兩種 QoS 技術，可限制特定應用程式、使用者或流量類別在鏈路上可使用的頻寬。

You can use traffic policing to control the maximum rate of traffic that is sent or received on an interface. Traffic policing is often configured on interfaces at the edge of a network to limit traffic into or out of the network. You can use traffic shaping to control the traffic going out an interface in order to match its flow to the speed of the remote target interface and to ensure that the traffic conforms to policies contracted for it.

你可以使用流量管制來控制介面上傳送或接收流量的最大速率。流量管制通常設定在網路邊緣的介面上，以限制進出網路的流量。你可以使用流量整形來控制介面輸出的流量，使其流量與遠端目標介面的速度相符，並確保流量符合為其約定的政策。

Policers make instantaneous decisions and are thus optimally deployed as ingress tools. The logic is that if you are going to drop the packet, you might as well drop it before spending valuable bandwidth and CPU cycles on it. However, policers can also be deployed at egress to control the bandwidth that a particular class of traffic uses. Such decisions sometimes cannot be made until the packet reaches the egress interface.

管制器會做出即時決策，因此最適合部署為入口工具。其邏輯是：既然要捨棄封包，不如在耗費寶貴的頻寬與 CPU 資源之前就先捨棄它。不過，管制器也可以部署在出口，用來控制特定類別流量所使用的頻寬。有時這類決策要到封包抵達出口介面時才能做出。

When traffic exceeds the allocated rate, the policer can take one of two actions. It can either drop traffic or re-mark it to another class of service. The new class usually has a higher drop probability, which means packets in this new class will be discarded earlier than packets in classes with higher priority.

當流量超出配置的速率時，管制器可採取兩種動作之一：捨棄流量，或將其重新標記為另一個服務類別。新類別通常有較高的捨棄機率，這表示這個新類別中的封包，會比優先等級較高類別中的封包更早被捨棄。

Shapers are commonly deployed on enterprise-to-service provider links on the enterprise egress side. Shapers ensure that traffic going to the service provider does not exceed the contracted rate. If the traffic exceeds the contracted rate, it would get policed by the service provider and likely dropped.

整形器通常部署在企業對服務提供商鏈路的企業出口端。整形器可確保傳送到服務提供商的流量不超過約定速率。若流量超過約定速率，就可能被服務提供商管制並極可能被捨棄。

While policers can cause a significant number of TCP re-sends when traffic is dropped, shaping involves fewer TCP re-sends. Policing does not cause delay or jitter in a traffic stream, but shaping does.

雖然管制器在捨棄流量時可能導致大量 TCP 重傳，但整形涉及的 TCP 重傳較少。管制不會造成流量的延遲或抖動，但整形會。

Traffic-policing mechanisms such as class-based policing also have marking capabilities in addition to rate-limiting capabilities. Instead of dropping the excess traffic, traffic policing can alternatively mark and then send the excess traffic. Excess traffic can be re-marked with a lower priority before the excess traffic is sent out. Traffic shapers, on the other hand, do not re-mark traffic; these only delay excess traffic bursts to conform to a specified rate.

類別式管制（class-based policing）等流量管制機制，除了速率限制能力外，也具備標記能力。流量管制可以不捨棄超額流量，而是改為標記後再傳送超額流量。超額流量在傳送出去之前，可以被重新標記為較低優先權。相對地，流量整形器不會重新標記流量，只會延遲超額流量突發，使其符合指定速率。

Congestion occurs anytime an interface is presented with more traffic than it is able to transmit. Aggressive traffic can fill interface queues and starve more fragile flows such as voice, video, and interactive traffic. The results can be devastating for delay-sensitive traffic types, making it difficult to meet the service-level requirements that these applications require. There are many congestion management techniques available on Cisco platforms that can provide effective means to manage software queues and to allocate the required bandwidth to specific applications when congestion exists.

只要介面收到的流量超過其可傳送的量，就會發生壅塞。過於激進的流量可能會填滿介面佇列，並排擠語音、視訊與互動式流量等較脆弱的流量。對延遲敏感的流量類型而言，這種情況可能造成嚴重後果，使其難以滿足所需的服務等級要求。Cisco 平台上有許多壅塞管理技術，可有效管理軟體佇列，並在發生壅塞時將所需頻寬分配給特定應用程式。

Whenever a packet arrives at an exit interface faster than it can exit, the potential for congestion exists. If there is no congestion, packets are sent when they arrive at the exit interface. If congestion occurs, congestion management tools are activated.

每當封包抵達出口介面的速度快過其可離開的速度時，就存在發生壅塞的可能性。若沒有壅塞，封包抵達出口介面時即會被傳送。若發生壅塞，壅塞管理工具就會啟動。

The scheduling tools that you use for QoS deployments therefore offer a combination of these algorithms and various ways to mitigate their downsides. This combination allows you to best tune your network for the actual traffic flows that are present.

因此，你在 QoS 部署中使用的排程工具，會結合這些演算法並以各種方式緩解其缺點。這種組合可讓你依照實際存在的流量狀況，最佳化你的網路。

Queuing algorithms are one of the primary ways to manage congestion in a network. Network devices handle an overflow of arriving traffic by using a queuing algorithm to sort traffic and determine a method of prioritizing the traffic onto an output link. Each queuing algorithm was designed to solve a specific network traffic problem and has a particular effect on network performance.

佇列演算法是管理網路壅塞的主要方式之一。網路裝置透過佇列演算法對流量進行排序，並決定將流量優先送上輸出鏈路的方法，以處理湧入流量的溢出。每種佇列演算法都是為解決特定的網路流量問題而設計，對網路效能有其特定影響。

With CBWFQ, you define the traffic classes based on match criteria, including protocols, Access Control Lists (ACLs), and input interfaces. Packets satisfying the match criteria for a class constitute the traffic for that class. A queue is reserved for each class, and traffic belonging to a class is directed to that class queue.

使用 CBWFQ 時，你可依比對條件（包括通訊協定、存取控制清單（ACL）與輸入介面）定義流量類別。符合某類別比對條件的封包即構成該類別的流量。每個類別都會保留一個佇列，屬於該類別的流量會被導向該類別佇列。

After a class has been defined according to its match criteria, you can assign characteristics to it. To characterize a class, you assign it the minimum bandwidth that it will be delivered during congestion.

類別依其比對條件定義完成後，你可以為其指派特性。為某類別設定特性時，你會指派它在壅塞期間應獲得傳送的最小頻寬。

To characterize a class, you also specify the queue limit for that class, which is the maximum number of packets allowed to accumulate in the class queue. Packets belonging to a class are subject to the bandwidth and queue limits that characterize the class. After a queue has reached its configured queue limit, enqueuing of additional packets to the class causes tail drop or random packet drop to take effect, depending on how the class policy is configured.

為類別設定特性時，你也要指定該類別的佇列限制，也就是該類別佇列中允許累積封包的最大數量。屬於某類別的封包受該類別所設定的頻寬與佇列限制約束。佇列一旦達到設定的佇列限制，再將封包排入該類別，就會觸發尾端捨棄（tail drop）或隨機封包捨棄，視類別政策的設定而定。

For CBWFQ, the weight for a packet belonging to a specific class is derived from the bandwidth that you assigned to the class when you configured it. Therefore, the bandwidth assigned to the packets of a class determines the order in which packets are sent. All packets are serviced fairly based on weight; no class of packets may be granted strict priority. This scheme poses problems for voice traffic, which is largely intolerant of delay, especially jitter.

對 CBWFQ 而言，屬於某特定類別封包的權重是依你在設定該類別時指派的頻寬而定。因此，指派給某類別封包的頻寬決定了封包被傳送的順序。所有封包都依權重公平地被服務，沒有任何類別的封包能獲得嚴格優先權。這種方式對語音流量而言會有問題，因為語音流量對延遲、尤其是抖動的容忍度很低。

The LLQ brings strict priority queuing to CBWFQ. Strict priority queuing allows delay-sensitive data such as voice to be dequeued and sent first (before packets in other queues are dequeued), giving delay-sensitive data preferential treatment over other traffic.

LLQ 為 CBWFQ 帶來了嚴格優先權佇列。嚴格優先權佇列讓語音等對延遲敏感的資料能優先被取出並傳送（早於其他佇列中封包被取出之前），使對延遲敏感的資料獲得優於其他流量的優先處理。

Congestion is a normal occurrence in networks. Whether congestion occurs as a result of a lack of buffer space, network aggregation points, or a low-speed wide-area link, many congestion management techniques exist to ensure that specific applications and traffic classes are given their share of available bandwidth when congestion occurs. When congestion occurs, some traffic is delayed or even dropped at the expense of other traffic. When drops occur, different problems may arise that can exacerbate the congestion, such as retransmissions and TCP global synchronization in TCP/IP networks. Network administrators can use congestion avoidance mechanisms to reduce the negative effects of congestion by penalizing the most aggressive traffic streams as software queues begin to fill.

壅塞是網路中的正常現象。無論壅塞是由於緩衝空間不足、網路匯聚點，還是低速廣域鏈路所造成，都存在許多壅塞管理技術，可確保特定應用程式與流量類別在發生壅塞時獲得應有的可用頻寬份額。發生壅塞時，部分流量會被延遲，甚至以犧牲其他流量為代價而被捨棄。發生捨棄時，可能引發其他問題，進而加劇壅塞，例如 TCP/IP 網路中的重傳與 TCP 全域同步。網路管理員可使用壅塞避免機制，在軟體佇列開始填滿時懲罰最激進的流量流，以降低壅塞的負面影響。

TCP has built-in flow control mechanisms that operate by increasing the transmission rates of traffic flows until packet loss occurs. When packet loss occurs, TCP drastically slows down the transmission rate and then again begins to increase the transmission rate. Because of TCP behavior, tail drop of traffic can result in suboptimal bandwidth utilization. TCP global synchronization is a phenomenon that can happen to TCP flows during periods of congestion because each sender will reduce the transmission rate at the same time when packet loss occurs

TCP 具有內建的流量控制機制，其運作方式是持續提高流量流的傳輸速率，直到發生封包遺失為止。發生封包遺失時，TCP 會大幅降低傳輸速率，然後再度開始提高傳輸速率。由於 TCP 的這種行為，流量的尾端捨棄可能導致頻寬使用效率不佳。TCP 全域同步是一種現象，可能發生在壅塞期間的 TCP 流量中，因為當封包遺失發生時，每個傳送端會同時降低傳輸速率。

Congestion avoidance techniques are advanced packet-discard techniques that monitor network traffic loads in an effort to anticipate and avoid congestion at common network bottleneck points.

壅塞避免技術是一種進階的封包捨棄技術，會監控網路流量負載，以嘗試預測並避免在常見網路瓶頸點發生壅塞。

Queues are finite on any interface. Devices can either wait for queues to fill up and then start dropping packets, or drop packets before the queues fill up. Dropping packets as they arrive is called tail drop. Selective dropping of packets while queues are filling up is called congestion avoidance. Queuing algorithms manage the front of the queue, and congestion mechanisms manage the back of the queue.

任何介面上的佇列容量都是有限的。裝置可以等待佇列填滿後才開始捨棄封包，也可以在佇列填滿之前就先捨棄封包。封包一到達就被捨棄稱為尾端捨棄（tail drop）；在佇列填滿過程中選擇性地捨棄封包則稱為壅塞避免。佇列演算法管理佇列的前端，壅塞機制則管理佇列的後端。

Randomly dropping packets instead of dropping them all at once, as it is done in a tail drop, avoids global synchronization of TCP streams. One such mechanism that randomly drops packets is random early detection (RED). RED monitors the buffer depth and performs early discards (drops) on random packets when the minimum defined queue threshold is exceeded.

與尾端捨棄一次性全部捨棄不同，隨機捨棄封包可避免 TCP 流量的全域同步。其中一種隨機捨棄封包的機制是隨機早期偵測（Random Early Detection，RED）。RED 會監控緩衝區深度，並在超過所定義的最小佇列門檻時，對隨機封包執行早期捨棄。

Cisco IOS Software does not support pure RED, but does support WRED. The principle is the same as with RED, except that the traffic weights skew the randomness of the packet drop. In other words, traffic that is more important will be less likely to be dropped than less important traffic.

Cisco IOS 軟體不支援純粹的 RED，但支援 WRED。其原理與 RED 相同，差別在於流量權重會使封包捨棄的隨機性產生偏斜。換句話說，較重要的流量被捨棄的機率會低於較不重要的流量。

The idea behind using WRED is both to maintain the queue length at a level somewhere between the minimum and maximum thresholds and to implement different drop policies for different classes of traffic. WRED can selectively discard lower-priority traffic when the interface becomes congested, and it can provide differentiated performance characteristics for different classes of service.

使用 WRED 的用意，一是將佇列長度維持在最小與最大門檻之間的某個水準，二是為不同類別的流量實作不同的捨棄政策。當介面發生壅塞時，WRED 可以選擇性地捨棄優先權較低的流量，並為不同服務類別提供差異化的效能特性。

The figure shows how WRED is implemented, as well as the parameters that WRED uses to influence packet-drop decisions.

此圖顯示 WRED 的實作方式，以及 WRED 用來影響封包捨棄決策的參數。

The router constantly updates the WRED algorithm with the calculated average queue length, which is based on the recent history of queue lengths.

路由器會持續以計算出的平均佇列長度更新 WRED 演算法，此平均值是依最近的佇列長度歷史記錄計算而得。

When a packet arrives at the output queue, the QoS marking value is used to select the correct WRED profile for the packet. The packet is then passed to WRED for processing. Based on the selected traffic profile and the average queue length, WRED calculates the probability for dropping the current packet (Probability Denominator). If the average queue length is greater than the minimum threshold but less than the maximum threshold, WRED will either queue the packet or perform a random drop. If the average queue length is less than the minimum threshold, the packet is passed to the output queue.

當封包抵達輸出佇列時，會使用 QoS 標記值來選擇適用於該封包的正確 WRED 設定檔。接著封包會被傳遞給 WRED 進行處理。根據所選的流量設定檔與平均佇列長度，WRED 會計算捨棄目前封包的機率（機率分母）。若平均佇列長度大於最小門檻但小於最大門檻，WRED 會將封包排入佇列或執行隨機捨棄。若平均佇列長度小於最小門檻，封包會被傳遞至輸出佇列。

If the queue is already full, the packet is tail-dropped. Otherwise, the packet will eventually be transmitted out onto the interface.

若佇列已滿，封包會被尾端捨棄。否則，封包最終會被傳送到該介面上。

Increasing the bandwidth of WAN links can be expensive. An alternative is to use QoS techniques to improve the efficiency of low-bandwidth links, which, in this context, typically refer to links with speeds less than or equal to 768 kbps. Header compression and payload compression mechanisms reduce the sizes of packets, reducing delay and increasing available bandwidth on a link. Other QoS link efficiency techniques, such as Link Fragmentation and Interleaving (LFI), allow traffic types, such as voice and interactive traffic, to be sent either ahead or interleaved with larger, more aggressive flows. These techniques decrease latency and assist in meeting the service-level requirements of delay-sensitive traffic.

提升 WAN 鏈路的頻寬可能成本高昂。另一種方式是使用 QoS 技術來改善低頻寬鏈路的效率，在此情境下，低頻寬鏈路通常是指速率小於或等於 768 kbps 的鏈路。標頭壓縮與承載資料壓縮機制可縮小封包大小，減少延遲並增加鏈路上的可用頻寬。其他 QoS 鏈路效率技術，例如鏈路分片與交錯（Link Fragmentation and Interleaving，LFI），可讓語音與互動式流量等流量類型，在較大、較激進的流量之前或與其交錯傳送。這些技術可降低延遲，有助於滿足對延遲敏感流量的服務等級要求。

While many QoS mechanisms exist for optimizing throughput and reducing delay in network traffic, QoS mechanisms do not create bandwidth. QoS mechanisms optimize the use of existing resources, and they enable the differentiation of traffic according to a policy. Link efficiency QoS mechanisms such as payload compression, header compression, and LFI are deployed on WAN links to optimize the use of WAN links.

雖然存在許多用於最佳化吞吐量並降低網路流量延遲的 QoS 機制，但 QoS 機制並不會創造頻寬。QoS 機制是最佳化既有資源的使用，並依政策實現流量的差異化處理。承載資料壓縮、標頭壓縮與 LFI 等鏈路效率 QoS 機制，會部署在 WAN 鏈路上，以最佳化 WAN 鏈路的使用。

Payload compression increases the amount of data that can be sent through a transmission resource. Payload compression is primarily performed on Layer 2 frames and therefore compresses the entire Layer 3 packet. The Layer 2 payload compression methods include Stacker, Predictor, and Microsoft Point-to-Point Compression (MPPC).

承載資料壓縮可增加透過傳輸資源傳送的資料量。承載資料壓縮主要在第 2 層框架上執行，因此會壓縮整個第 3 層封包。第 2 層承載資料壓縮方法包括 Stacker、Predictor 與 Microsoft 點對點壓縮（MPPC）。

Compression methods are based on eliminating redundancy. The protocol header is an item of repeated data. The protocol header information in each packet in the same flow does not change much over the lifetime of that flow. Using header compression mechanisms, most header information can be sent only at the beginning of the session, stored in a dictionary, and then referenced in later packets by a short dictionary index. Cisco IOS header compression methods include TCP header compression, Real-Time Transport Protocol (RTP) header compression, class-based TCP header compression, and class-based RTP header compression.

壓縮方法的基礎在於消除冗餘。通訊協定標頭是一種重複出現的資料項目。同一流量中每個封包的通訊協定標頭資訊，在該流量的生命週期內變化不大。使用標頭壓縮機制，大部分的標頭資訊只需在會話開始時傳送一次，儲存於字典中，之後的封包只需以簡短的字典索引來參照即可。Cisco IOS 的標頭壓縮方法包括 TCP 標頭壓縮、即時傳輸協定（RTP）標頭壓縮、類別式 TCP 標頭壓縮，以及類別式 RTP 標頭壓縮。

It is important to note that Layer 2 payload compression and header compression are performed on a link-by-link basis. These compression techniques cannot be performed across multiple routers because routers need full Layer 3 header information to be able to route packets to the next hop.

值得注意的是，第 2 層承載資料壓縮與標頭壓縮是以逐段鏈路為基礎執行的。這些壓縮技術無法跨越多台路由器執行，因為路由器需要完整的第 3 層標頭資訊才能將封包路由到下一跳。

LFI is a Layer 2 technique in which large frames are broken into smaller, equally sized fragments and then transmitted over the link in an interleaved fashion with more latency-sensitive traffic flows (like Voice over IP). Using LFI, smaller frames are prioritized, and a mixture of fragments is sent over the link. LFI reduces the queuing delay of small frames because the frames are sent almost immediately. Link fragmentation, therefore, reduces delay and jitter by expediting the transfer of smaller frames through the hardware transmit queue.

LFI 是一種第 2 層技術，將大型框架切割成較小、大小相等的片段，再以交錯方式與延遲敏感度較高的流量流（例如 VoIP）一起透過鏈路傳送。使用 LFI 後，較小的框架會被優先處理，並在鏈路上傳送混合的片段。由於框架幾乎能立即被傳送，LFI 降低了小型框架的佇列延遲。因此，鏈路分片可透過加速小型框架通過硬體傳輸佇列，降低延遲與抖動。