当前位置:网站首页>Network equipment hard core technology insider router Chapter 11 Cisco asr9900 disassembly (V)
Network equipment hard core technology insider router Chapter 11 Cisco asr9900 disassembly (V)
2022-07-27 15:26:00 【User 8289326】
In the topic of the previous few days , We disassembled ASR9900 Forwarding line card 、 Switch network board 、NP chip 、 Control plane implementation .
In the design of large-scale distributed systems , Another key point is the allocation of resources .
For the forwarding plane , The most important resource is the table item resource , That is to say TCAM Allocation of resources .
Let's go over it again ASR9900 The forwarding process :
After entering the line card, the data packets are classified , Determine whether to point to the control plane or forward .
after , yes ACL and QoS classification , Give Way NP Inquire about TCAM, Make sure it matches ACL and QoS Strategy .
next ,NP Start performing the most critical steps : Post search , In this step ,NP Will be in TCAM Query in FIB surface .
there FIB It is a generalized forwarding information table entry , Not limited to LPM( The longest match ) Three layer routing , There will be MPLS Tag forwarding table and layer 2 MAC Forwarding table , These two items belong to LEM( Exactly match ).
Besides ,MPLS FEC( Forwarding equivalence class ) Then there may be based on LEM Of (VLL/VPLS), It may also be based on LPM Of .(MPLS L3 VPN).
These items are in TCAM Store and retrieve .
For outgoing direction , Save FIB Table lookup , But according to the data packet header Metadata Forward the packet to the corresponding port , And find out the direction ACL/QoS Strategy and implementation . These also need access TCAM.
NP Put the above TCAM The query action is abstracted as these steps :
- Build query key values . For example, the query key value of the three-layer routing forwarding table is prefixed +Token.Token yes VRF A related one ID; and ACL The query key value of can be IP Prefix , It can also be UDP/TCP Port, etc .
- Bitmask (Bit Mask). Mentioned earlier ,TCAM Query for , Some bit by Not Care, Here's the picture :
- choice TCAM Bank.LightSpeed and Tomahawk On the card ,TCAM With BANK Form organization , Every time BANK It can be configured to have 512 Entry x 160bit The way , It can also be configured as 128 Entry x 640bit The way . Multiple BANK It can be configured as a BANK Group (Bank Set). stay Tomahawk On the card , share 9 individual BANK Group . among 5 There are two business groups , Such as physical interfaces and interface aggregation groups ,VLAN, Tunnel and other resources ;2 A for 160bit Group , be used for IPv4 FIB and ACL;2 A for 640bit Group , be used for IPv6 FIB and IPv6 ACL. And for LightSpeed Line card , Tier 2 business does not need TCAM Resources .NP The chip has built in SRAM And hash engine implementation LEM( Exact match search ), Only FIB and ACL Need to consume TCAM resources . The following table is different TCAM The purpose of the partition .
actually , Whether on the router or switch ,TCAM Resources are very valuable . Generally used for high-speed table lookup FIB Table and ACL surface .
because IPv6 The address itself needs to be occupied 128bit, therefore ,IPv6 Table entry in TCAM In general, it is 640bit Unit allocation .
Yes ASR9K The disassembly of is over . Let's summarize our experience :
- ASR9900 Adopt multi-level switching architecture , Use on the line card CISCO Self developed NP Implement forwarding ;
- NP Only packet processing and QoS Ability to manage . Table entries are stored in TCAM in ,VoQ The exchange adopts FIA Realization ;
- master control CPU And line card CPU Jointly build a distributed control plane , Control message distributed processing ;
- TCAM according to IPv4 or MPLS/IPv6 organization , Yes 160bit/640bit Two ways of organizing ;
Tomorrow we will disassemble the implementation of another router .
Current issues :
Why? LEM The table may not need to be in TCAM Storage in ?
Problems left over from the previous period :
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 :6 star . Dual master and 4 The block line card shares the function of the control plane , And the exchange network board CPU Only realize arbitration and chip status monitoring .
边栏推荐
猜你喜欢

Watermelon book machine learning reading notes Chapter 1 Introduction

Selenium reports an error: session not created: this version of chromedriver only supports chrome version 81

泛型

华为鸿蒙模拟器去除顶部导航栏方法

周鸿祎:数字安全能力落后也会挨打
Principle of MOS tube to prevent reverse connection of power supply

工具 - markdown编辑器常用方法

适配验证新职业来了!华云数据参与国家《信息系统适配验证师国家职业技能标准》编制

Dialog manager Chapter 3: create controls

After configuring corswebfilter in grain mall, an error is reported: resource sharing error:multiplealloworiginvalues
随机推荐
Comparison of advantages and disadvantages between instrument amplifier and operational amplifier
With just two modifications, apple gave styleganv2 3D generation capabilities
Code coverage statistical artifact -jacobo tool practice
LeetCode 81. 搜索旋转排序数组 II 二分/medium
LeetCode 90. 子集 II 回溯/medium
LeetCode 341.扁平化嵌套列表迭代器 dfs,栈/ Medium
LeetCode 1143. 最长公共子序列 动态规划/medium
Leetcode-1737- minimum number of characters to change if one of the three conditions is met
What is the breakthrough point of digital transformation in the electronic manufacturing industry? Lean manufacturing is the key
Overview of wechat public platform development
How to edit a framework resource file separately
网络设备硬核技术内幕 路由器篇 7 汤普金森漫游网络世界(下)
Adaptation verification new occupation is coming! Huayun data participated in the preparation of the national vocational skill standard for information system adaptation verifiers
基于FIFO IDT7202-12的数字存储示波器
EMC design scheme of USB2.0 Interface
Leetcode 244 week competition - post competition supplementary question solution [broccoli players]
The design method of integral operation circuit is introduced in detail
2022-07-27日报:IJCAI 2022杰出论文公布,大陆作者中稿298篇拿下两项第一
Deveco studio2.1 operation item error
Unity性能优化------渲染优化(GPU)之LOD(Level of detail)