Optimization Of Software Framework To Improve Packet Processing Throughput Through Customized Configuration For Small Data Packet
This research investigates the method to improve the data transmission throughput in a networking system. The reason of this research proposal is because of computer network has grown tremendously in recent years and it requires to handle more complex data packet with different sizes from applicatio...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2016
|
| Subjects: | |
| Online Access: | http://eprints.usm.my/41319/1/SAM_POH_MUN_24_Pages.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | This research investigates the method to improve the data transmission throughput in a networking system. The reason of this research proposal is because of computer network has grown tremendously in recent years and it requires to handle more complex data packet with different sizes from applications. In this research, the throughput for various packet sizes used in a data transmission system were analyzed. The common network frameworks may not be efficient in handling the data traffic with random packet size especially when there are many small packet frames. Therefore, the optimization on the network framework is necessary to improve the packet processing throughput for small packet size. Investigation in methods to improve the local memory accessing rate in processor and network device were conducted. In addition, the duplication of memory address from system memory to cache is reduced by implementing Huge Page Table. Then, the impact on low throughput due to the context switching overhead originated from the system and device interrupts were analyzed. The interrupt polling mechanism were implemented on receive and transmit paths of the network driver for reducing the overheads. Another improvement introduced was by enabling the burst mode in the receive port. This will make the incoming packets being received and processed in bulk, and therefore removing the latency of processing each packet individually. The proposed improvements have been implemented on a Data Plane Development Kit (DPDK) framework and tested on the hardware using simulated data traffic. The results showed that the improved framework is able to achieve better network throughput by 61% compared to the conventional framework particularly for small packet size. |
|---|