As a new technology ,Segment Routing(SR) It's been very popular in recent years , It and another concept SDN Effective combination , Something very interesting has been achieved , for instance SR-TE.

But in fact ,SR In fact, it is SDN/SDWAN A natural way of thinking when developing to a certain stage .

SDWAN One of the most important things is “ How to define a path from any node to any node .”

Custom paths can be used in traditional link metrics based IGP Besides route selection, it is more flexible to measure and select the shortest path , Provide better transmission service for complicated business .

The implementation of custom path is no more than the following way :

* Tunnel mode .
* Loose source routing (loose source route).
* Strict source routing (strict source route).
* The way of cascading forward agents .
* tradition NAT The way .
* …
Under the sun , Nothing new . about overlay Mode SDWAN, Its custom path is actually a kind of loose source routing .

But this article starts with the tunnel mode as an example .

think about it , In the traditional IP In the network , How to build a tunnel :

It's a very simple operation , It only needs to be on the equipment at both ends of the tunnel Do some configuration that will do , This is also the daily work of many network workers . As for how the packets carried by the tunnel actually arrive at the opposite end , Then give it to me IGP bar .

IGP Is the general name of internal routing protocol , such as OSPF,IS-IS All of them , It's a distributed protocol , All routers work together to calculate the optimal path by exchanging information .

Now? , Enter SDWAN Tunnel scene , I need to be in tradition IP On the Internet overlay A network of tunnels , This means that multiple tunnels have to be connected from end to end , An interconnected structure is formed :
If the tunnel is still configured in the traditional way , Network workers and operation and maintenance people are not kneeling . tradition IP The purpose of the network tunnel is not to build this Large scale interconnection overlay network Of .

I actually built an interconnected one a few years ago overlay Tunnel network , When the centralized controller is about to be installed , I left … At that time SDN Ideas are spontaneous , So it's not a tall thing , As long as you're there and you're suffering from operations , You can imagine that, too .

Since we don't want to deploy human resources , So it's obvious to leave it to the machine , This is it. SDN The thought of .

Pull out a new machine , As controller , Collect network topology information , Only configuration choreography is needed on this controller , Then these configurations can be automatically distributed to each node router :

Free from manual work , But it's still messy .

IGP The calculation of optimal path is often based on some basic indicators , Like physical distance , bare RTT, Line configuration bandwidth, etc , however overlay The shortest path calculation of the network is much more flexible , This means that it will re converge relatively frequently .

If the path converges frequently , The mechanism in the figure above produces more control traffic , That is, southward flow , Frequent demolition and construction of tunnels , It will greatly affect the synchronization of the network .

Change your mind , Let the packet bring its own path information . such , Only the underlying standard is needed IP The routing can be completed, and the packets can be stored in the overlay Transportation on the Internet . This is it. SR The way of thinking :

The controller only sends path information to edge devices , The path of these paths is similar to a loose source route , with stack Forms pile up , And then according to the standard IP Hop by hop routing is OK , In a nutshell :

* pkt arrive A, Match to stream1, Stack label stack B-C,A according to IGP Query arrival B The next jump of , The result is A_n, Send to A_n.
* pkt arrive A_n,A_n See the B, query IGP, Direct connection B,pop fall B label , Send to B.
* pkt arrive B,B See the C, query IGP Find the arrival C The next jump of , The result is B_n, Send to B_n.
* …
You'll see , All it takes is IGP Find it , On and standard IP The route is the same , But obviously, just like the source routing, it limits the path through which the packets go .

Maybe you'll say , It's not easy , At least the device needs to be modified to support tag forwarding .

Specific to implementation , It's all kinds of things , But it's all once and for all , in addition , No equipment modification , I use standard NAT These logics are also implemented with source routing technology :
Um. , This is also an implementation of source routing .

But do not implement custom paths with source routing options , Because a lot of operators router will guard against this kind of behavior , Therefore, packets with the source routing option are discarded !

For operators , This is reasonable , You have to take the path it has planned for you unconditionally .

in fact , Operators may have thought that some Internet companies or folk craftsmen will hide all kinds of tricks in the transit traffic to build some gray things , So it's understandable to ban unnecessary options .

So far , I didn't mention it MPLS, Actually, I did it on purpose .SR The idea of thinking can be spontaneous , So you don't have to force SR and MPLS Connect , of course ,SR It can be borrowed directly MPLS Implementation of .

in my opinion ,MPLS It's an old technology , Many years ago , You have to design some complex technology to glue IP,ATM Such as a variety of network transmission protocols , Now , Basically IP agreement , rely on IP Agreements play all kinds of overlay, Simplicity is king !

If the SR and overlay Network integration ,SR Are all nodes of overlay Nodes of the network , So in general :

* SR Packet free underlay Hop by hop path .
* SR One of the data packets is specified overlay Hop by hop path of .
* SR Allowed in underlay do ECMP.

If you just look at it SR Technology itself , It's really nothing new , But with SDN Technology integration , You can achieve a very simple and clear custom routing , That's it SDWAN.

Easy deployment , Fast iteration is SDWAN Characteristics of , So it's doomed not to use tradition WAN The complex distributed protocol in , such as MPLS-LDP,IS-IS etc. , Those complex technologies are very unsuitable S(oftware/elf)D(efine), At best, it can be attributed to WAN The category of optimization .

conventional Underlay
WAN It is still dominated by traditional operators and equipment providers , After all, there are many of them , rich , It can fully interact with each other . However, for the freshmen SDWAN, It can let more folk craftsmen have more share .

Both in the existing traditional operators WAN Go up and do it overlay, Or self built computer room, white brand self research equipment interconnection , All belong to SDWAN, nothing less than SDWAN The two camps of , But they all face almost the same problems :

* Folk craftsmen camp : There is not enough money to buy equipment and bandwidth resources , There is not enough manpower for R & D and operation .
* Big Internet company camp : Business centered , There is not enough incentive to put money and people on the underlying infrastructure .
therefore ,SDWAN It has to be simple , Easy to deploy , Easy to operate and maintain ,SR Technology is in line with this :

Yes, of course ,SDWAN Not only SR One technology is available ,SDWAN no SR,SR Neither SDWAN. Since this article is a small talk SR, That must be more ink .

Zhejiang Wenzhou shoes wet , Rain and water will not make you fat .

©2019-2020 Toolsou All rights reserved,
Final review of database : Summary of comprehensive application questions use Python Make simple games Laplance operator ( Second derivative ) Convert hard disk to GPT Partition format Python Implementation of Hanoi Tower code about String How to create objects vue3 Learning journey 1—— establish vue3 project java String from back to front _Java String String summary use Python Write a story about plants versus zombies 【 Greedy Algorithm 】 Huffman coding problem