To perform this task, the control entity/program has to develop a view of the network
What Do They Do? | 11
topology that satisfies certain constraints. This view of the network can be programmed
manually, learned through observation, or built from pieces of information gathered
through discourse with other instances of control planes, which can be through the use
of one or many routing protocols, manual programming, or a combination of both.
The mechanics of the control and data planes is demonstrated in Figure 2-2, which
represents a network of interconnected switches. At the top of the figure, a network of
switches is shown, with an expansion of the details of the control and data planes of two
of those switches (noted as A and B). In the figure, packets are received by switch A on
the leftmost control plane and ultimately forwarded to switch B on the righthand side
of the figure. Inside each expansion, note that the control and data planes are separated,
with the control plane executing on its own processor/card and the data plane executing
on a separate one. Both are contained within a single chassis. We will discuss this and
other variations on this theme of physical location of the control and data planes later
in the chapter. In the figure, packets are received on the input ports of the line card
where the data plane resides. If, for example, a packet is received that comes from an
unknown MAC address, it is punted or redirected (4) to the control plane of the device,
where it is learned, processed, and later forwarded onward. This same treatment is given
to control traffic such as routing protocol messages (e.g., OSPF link-state advertisements).
Once a packet has been delivered to the control plane, the information contained
therein is processed and possibly results in an alteration of the RIB as well as the
transmission of additional messages to its peers, alerting them of this update (i.e., a new
route is learned). When the RIB becomes stable, the FIB is updated in both the control
plane and the data plane. Subsequently, forwarding will be updated and reflect these
changes. However, in this case, because the packet received was one of an unlearned
MAC address, the control plane returns the packet (C) to the data plane (2), which
forwards the packet accordingly (3). If additional FIB programming is needed, this also
takes place in the (C) step, which would be the case for now the MAC addresses source
has been learned. The same algorithm for packet processing happens in the next switch
to the right.
The history of the Internet maps roughly to the evolution of control schemes for managing
reachability information, protocols for the distribution of reachability information,
and the algorithmic generation of optimized paths in the face of several challenges.
In the case of the latter, this includes an increasing growth of the information base used
(i.e., route table size growth) and how to manage it. Not doing so could result in the
possibility of a great deal of instability in the physical network. This in turn may lead to
high rates of change in the network or even nonoperation. Another challenge to overcome
as the size of routing information grows is the diffusion of responsibility for
advertising reachability to parts of the destination/target data, not only between local
instances of the data plane but also across administrative boundaries.
12 | Chapter 2: Centralized and Distributed Control and Data Planes
Figure 2-2. Control and data planes of a typical network
In reality, the control plane for the Internet that was just discussed is some combination
of layer 2 or layer 3 control planes. As such, it should be no surprise then that the same
progression and evolution has taken place for both layer 2 and layer 3 networks and the
protocols that made up these control planes. In fact, the progression of the Internet
happened because these protocols evolved both in terms of functionality and hardware
vendors learned how to implement them in highly scalable and highly available ways.
A layer 2 control plane focuses on hardware or physical layer addresses such as IEEE
MAC addresses. A layer 3 control plane is built to facilitate network layer addresses such
as those of the IP protocol. In a layer 2 network, the behaviors around learning MAC
addresses, the mechanisms used to guarantee an acyclic graph (familiar to most readers
through the Spanning Tree Protocol), and flooding of BUM (broadcast, unicast unknown,
and multicast) traffic create their own scalability challenges and also reveal their
scalability limitations. There have been several iterations or generations of
standards-based layer 2 control protocols whose goals were to address these and other
What Do They Do? | 13
issues. Most notably, these included SPB/802.1aq from the IEEE and TRILL from the
IETF.
As a generalization, though, layer 2 and layer 3 scaling concerns and their resulting
control plane designs eventually merge or hybridize becau
ภารกิจนี้ เอนทิตี/โปรแกรมควบคุมมีการพัฒนามุมมองของเครือข่ายพวกเขาทำอะไร? | 11โครงสร้างที่เป็นไปตามข้อจำกัดบางอย่าง มุมมองของเครือข่ายสามารถตั้งโปรแกรมหรือด้วยตนเอง เรียนรู้ผ่านการสังเกต สร้างจากชิ้นส่วนของข้อมูลที่รวบรวมผ่านวาทกรรมกับอินสแตนซ์อื่น ๆ ของเครื่องบินควบคุม ซึ่งได้ผ่านการใช้หนึ่ง หรือหลายโพรโทคอสายงานการผลิต เขียนโปรแกรมด้วยตนเอง หรือใช้ทั้งสองอย่างกลศาสตร์ของเครื่องบินควบคุมและข้อมูลจะแสดงในรูป 2-2 ที่แสดงเครือข่ายสวิตช์เชื่อมต่อกัน ที่ด้านบนของตัวเลข เครือข่ายสวิตช์จะมี ขยายรายละเอียดของเครื่องบินควบคุมและข้อมูลของทั้งสองสวิตช์เหล่านี้ (บันทึกเป็น A และ B) ในรูป แพคเก็ตที่รับ โดย A สลับบนตัวควบคุมซ้ายแหวน และส่งต่อกับสวิตช์ B สุด righthand ด้านของตัวเลข ภายในแต่ละการขยายตัว ทราบว่า จะแยกเครื่องบินควบคุมและข้อมูลกับการควบคุมเครื่องบินดำเนินการบนตัวเองที่ตัวประมวลผลบัตรข้อมูลเครื่องบินดำเนินในหนึ่งแยกต่างหาก ทั้งสองอยู่ภายในตัวถังเดียว เราจะพูดคุย และรูปแบบอื่น ๆ ในรูปแบบนี้ของที่ตั้งทางกายภาพของข้อมูลและควบคุมเครื่องบินในภายหลังในบทนี้ ในรูป ได้รับแพคเก็ตพอร์ตอินพุตของบัตรรายการซึ่งเครื่องบินข้อมูลอยู่ ถ้า เช่น แพคเก็ตรับที่มาจากการไม่ทราบที่อยู่ MAC เป็น punted หรือเปลี่ยนเส้นทาง (4) การควบคุมเครื่องบินอุปกรณ์where it is learned, processed, and later forwarded onward. This same treatment is givento control traffic such as routing protocol messages (e.g., OSPF link-state advertisements).Once a packet has been delivered to the control plane, the information containedtherein is processed and possibly results in an alteration of the RIB as well as thetransmission of additional messages to its peers, alerting them of this update (i.e., a newroute is learned). When the RIB becomes stable, the FIB is updated in both the controlplane and the data plane. Subsequently, forwarding will be updated and reflect thesechanges. However, in this case, because the packet received was one of an unlearnedMAC address, the control plane returns the packet (C) to the data plane (2), whichforwards the packet accordingly (3). If additional FIB programming is needed, this alsotakes place in the (C) step, which would be the case for now the MAC addresses sourcehas been learned. The same algorithm for packet processing happens in the next switchto the right.The history of the Internet maps roughly to the evolution of control schemes for managingreachability information, protocols for the distribution of reachability information,and the algorithmic generation of optimized paths in the face of several challenges.In the case of the latter, this includes an increasing growth of the information base used(i.e., route table size growth) and how to manage it. Not doing so could result in thepossibility of a great deal of instability in the physical network. This in turn may lead to
high rates of change in the network or even nonoperation. Another challenge to overcome
as the size of routing information grows is the diffusion of responsibility for
advertising reachability to parts of the destination/target data, not only between local
instances of the data plane but also across administrative boundaries.
12 | Chapter 2: Centralized and Distributed Control and Data Planes
Figure 2-2. Control and data planes of a typical network
In reality, the control plane for the Internet that was just discussed is some combination
of layer 2 or layer 3 control planes. As such, it should be no surprise then that the same
progression and evolution has taken place for both layer 2 and layer 3 networks and the
protocols that made up these control planes. In fact, the progression of the Internet
happened because these protocols evolved both in terms of functionality and hardware
vendors learned how to implement them in highly scalable and highly available ways.
A layer 2 control plane focuses on hardware or physical layer addresses such as IEEE
MAC addresses. A layer 3 control plane is built to facilitate network layer addresses such
as those of the IP protocol. In a layer 2 network, the behaviors around learning MAC
addresses, the mechanisms used to guarantee an acyclic graph (familiar to most readers
through the Spanning Tree Protocol), and flooding of BUM (broadcast, unicast unknown,
and multicast) traffic create their own scalability challenges and also reveal their
scalability limitations. There have been several iterations or generations of
standards-based layer 2 control protocols whose goals were to address these and other
What Do They Do? | 13
issues. Most notably, these included SPB/802.1aq from the IEEE and TRILL from the
IETF.
As a generalization, though, layer 2 and layer 3 scaling concerns and their resulting
control plane designs eventually merge or hybridize becau
การแปล กรุณารอสักครู่..
