Research On The Traffic Engineering In Data Center Networks Based On Inband Network Telemetry

The population of planet Earth is growing,and with it,and even faster,the number of devices connected to the Internet is increasing.High-quality video,real-time transmission of high-quality video,financial transactions of giant companies and much more.All this is sent,processed and stored in data centers.From data centers,people expect reliability and speed.In this research work,we will study a method for optimization,and,accordingly,for increasing speed in data centers.To solve the problem of monitoring the network status,it was decided to use a computer network emulator-Mininet.This software uses hosts,switches,and also OpenFlow controllers.Using the simplest syntax in a primitive command interpreter,you can deploy networks from an arbitrary number of ho sts,switches in various topologies,all within a single virtual machine(VM).On all hosts,you can change the network configuration,use standard utilities(ipconfig,ping)and even access the terminal.You can add various rules to the switches and route traffic.Mininet has excellent compatibility with the Python programming language.Therefore,to create a network topology which is not in the list of standard topologies,we will use this function.Create twelve switches and eight hosts.Since for research in this work it is necessary to obtain as much data as possible,it was decided to use switches,which can be given a set of instructions.P4switch complies with these requirements.It should be noted that the replacement of the standard switches of the Mininet virtual network emulator is performed using the P4 programming language.Having created the network,the next question is how to collect telemetry data from the constructed network?The solution is the "Intracavitary network telemetry"already existing in the world.It remains to figure out what is INT.Intracavitary network telemetry("INT")is a structure designed to collect and report network status on the data plane without requiring intervention or work at the control level.Obviously,it is preferable for network devices to process any INT packets strictly within the fast path,often in the hardware forwarding plane.An ideal system would be able to process INT commands without increasing latency or slowing down forwarding performance,but in some cases it might be necessary to process INT packets outside the fast path.This slow processing may be a CPU-based control plane,some sideband,or an alternative hardware-assisted forwarding path or arbitrary INT resource.It should be noted that in the case where INT processing is performed outside the fast path,the device MUST still forward the original packet through the fast path(that is,without processing INT instructions).The implementation of this is not indicated,although it implies the ability to make a copy of the INT package for slow processing or a similar design.After processing the INT packet in the usual fast path the forwarding plane should generate a trigger for the slow path(for example,either a copy of the original INT packet or its digest).After receiving the trigger,the slow path should process the INT header accordingly,generate a new package called the "subsequent package" containing the results of the INT commands.The subsequent packet is forwarded separately.If the devices on the network perform slow INT processing,it is possible that one INT packet can spawn several subsequent packets-and,in turn,each of them can spawn more INT processing in the downstream direction.Care must be taken to prevent excessive replication.This method will allow data centers to increase productivity,which will affect the quality of the services provided and the benefits for the owner of the data center.