Inside router of network equipment hard core technology (10) disassembly of Cisco asr9900 (4)

First summarize the contents of the first three chapters ：

CISCO ASR99 As a high-end router , Usually in the enterprise headquarters /DC exit 、 Wan core , And the edge of the operator city (Metro Edge) As BRAS/SR Use , therefore , The whole machine adopts distributed and HA Design , Include NP+FIA+Switch Fabric Line cards and Switch Fabric+ The switching network board composed of arbitration unit . Specially , Distributed switching network arbitration and VoQ The design of exchange , Avoid congestion in the switching network , It also avoids packet loss caused by single port congestion on the line card .

that ,ASR9900 Distributed in the control plane HA How is the implementation done ？

Pictured ,ASR9900 The control plane of is controlled by the master CPU、 Line card CPU as well as NP The control plane package processing unit of .

When routing protocol 、BFD、 Data packets such as spanning tree protocol are sent to NPU when , First of all, by LPTS（Local Packet Transport Service） To deal with . Because the processing capacity of the control plane is much lower than that of the forwarding plane , therefore ,LPTS It will limit the current according to certain strategies , Avoid sending too many local data packets to the master .

about BFD This kind of quantity is large , Frequent cycles (3.3ms), Simple processing ( The event will be triggered if it is not received for several consecutive cycles ) Data packets of ,NP It can be processed in local hardware , Do not send it to the master .

ARP,ICMP,OAM As well as the two-layer spanning tree protocol and other local control packets, the online card CPU Upper processing .

What control package needs to be in the master CPU What about processing? ？

Of course, it is a control package involving the whole situation . Such as routing protocol 、MPLS LDP,PIM,HSRP and VRRP etc. .

Their working interfaces are divided as follows ：

Remember what we mentioned ,FIB The generation process of table items ？

For distributed forwarding routers , actually ,FIB The table item has software FIB(SW FIB) And hardware FIB(HW FIB) The difference between . Why do we need such a design ？

original , If all line cards are transferred to the master CPU Initiate a query , master control CPU The burden will be heavier . therefore , In master control CPU After learning the route , Will be based on RIB Select the appropriate routing table entry , Generating software FIB, And distributed to all line cards CPU. Dang line card NP When you receive a packet that cannot be forwarded , Online card CPU Query software FIB, Only when the query cannot be found will the master query be launched . such , You can use the CPU Share the master CPU Send out FIB The job of , Reflect the advantages of distributed systems .

For layer 2 Forwarding , Routers work in a similar way to switches , Need to know about the unknown MAC Study .ASR9900 Of MAC Learning is divided into two steps ：

1. NP I learned the unknown layer-2 data packet MAC;
2. NP Yes MAC The table item goes through Switch Fabric Flood to all in the card of this line NP, And other line cards NP.

The control plane has another important task , Is dealing with OAM/BFD A kind of heartbeat message .

because BFD There are a lot of packets , With 3.3ms Periodometer , Every second 300 individual . If the whole machine has 480 Interface , The whole machine needs to be processed every second 144000 individual BFD message . We need a distributed approach , Avoid this 144K individual BFD All messages are in the master CPU Handle .

Pictured , In master control (RP) On the configuration BFD after , master control (RP) Will maintain a BFD The sessions table , And distribute it to the line card . Of each line card NP The chip will handle BFD conversation , And in BFD When the state changes , Report to the master (RP).RP Will target BFD State switching route .

because NP It has powerful packet processing capability , stay Tomahawk and LightSpeed The line card supports NP Handle BFD.

Every BFD Hello stay NP Top processing ,NP Found three consecutive BFD The loss of , Will report the line card CPU Handle .

Current issues ：

At one ASR9906 Dual master control is configured on ,4 Block line card and 5 Block switching network board , that , The function of the control plane consists of several CPU Share ？

Answer questions left over from last issue ：

since ASR9900 Yes N+1 Redundant switching network board , And between the same line card NP It also requires interworking on the switching network board , Why design on online card Switch Fabric？

because NP To the interface Serdes Capacity and to Switch Fabric The exchange capacity of is equal , therefore , If there is no Switch Fabric Provide redundant bandwidth to the switching network board , that , Unable to switch the network board N+1 redundancy .