Thesis etd-07042016-164225 |
Link copiato negli appunti
Thesis type
Tesi di laurea magistrale
Author
IOZZELLI, YURI
URN
etd-07042016-164225
Thesis title
Performance improvements on the P4 software switch
Department
INGEGNERIA DELL'INFORMAZIONE
Course of study
COMPUTER ENGINEERING
Supervisors
relatore Prof. Rizzo, Luigi
relatore Prof. Lettieri, Giuseppe
relatore Prof. Lettieri, Giuseppe
Keywords
- SDN netmap optimization
Graduation session start date
22/07/2016
Availability
Full
Summary
The Programming Protocol-Independent Packet Processors (P4) is a
domain-specific language designed to allow programming of packet forwarding
dataplanes. It is used within a Software-Defined Networking (SDN) architecture,
and it is meant to replace the Openflow protocol, being more flexible and
powerful.
The reference software implementation of a P4 switch is currently used mainly
to test the correctness of P4 programs, before deploying them on hardware
switches. The software switch has poor performance in terms of throughput, and
this prevent it from being used in many different scenarios, like network
protocol experimentation and Virtual Machine interconnection.
The aim of this work is to analyze the bottlenecks of this implementation and
improve the performance, trying to reach a throughput with an order of
magnitude of 1 Mpps. Particular care has been used in making all the
modifications as compatible as possible with the existing code base, in order
to favour adoption of the improvements in existing and future applications.
domain-specific language designed to allow programming of packet forwarding
dataplanes. It is used within a Software-Defined Networking (SDN) architecture,
and it is meant to replace the Openflow protocol, being more flexible and
powerful.
The reference software implementation of a P4 switch is currently used mainly
to test the correctness of P4 programs, before deploying them on hardware
switches. The software switch has poor performance in terms of throughput, and
this prevent it from being used in many different scenarios, like network
protocol experimentation and Virtual Machine interconnection.
The aim of this work is to analyze the bottlenecks of this implementation and
improve the performance, trying to reach a throughput with an order of
magnitude of 1 Mpps. Particular care has been used in making all the
modifications as compatible as possible with the existing code base, in order
to favour adoption of the improvements in existing and future applications.
File
| Nome file | Dimensione |
|---|---|
| tesi.pdf | 904.05 Kb |
Contatta l’autore |
|