Thesis_ECDS - CSIE -NCKU

advertisement
GFlow: Towards GPU-based HighPerformance Table Matching in
OpenFlow Switches
Author : Kun Qiu, Zhe Chen, Yang Chen, Jin Zhao, Xin Wang
Publisher : Information Networking (ICOIN), 2015 International
Conference on
Presenter: Tung-yin Chi
Date: 2015/3/25
Department of Computer Science and Information Engineering
National Cheng Kung University, Taiwan R.O.C.
Introduction

This paper investigates the acceleration of Software based
OpenFlow switches, equipped with commodity off-theshelf hardware, for high-performance table matching.

In our work, we leverage the power of GPUs to accelerate
table matching in software-based OpenFlow switches. We
propose GFlow, which can handle OpenFlow table
matching in a parallel fashion.

Based on our extensive evaluations, we can see the GFlow
is 8 to 10 times faster than existing GPU-based matching
algorithm.
National Cheng Kung University CSIE
Computer & Internet Architecture Lab
2
Data Structure: ItemGraph
National Cheng Kung University CSIE
Computer & Internet Architecture Lab
3
ItemGraph
National Cheng Kung University CSIE
Computer & Internet Architecture Lab
4
ItemGraph
1
2
3
4
National Cheng Kung University CSIE
Computer & Internet Architecture Lab
5
ItemGraph
1
2
3
4
National Cheng Kung University CSIE
Computer & Internet Architecture Lab
6
ItemGraph
1
2
3
4
National Cheng Kung University CSIE
Computer & Internet Architecture Lab
7
Matching on the ItemGraph
National Cheng Kung University CSIE
Computer & Internet Architecture Lab
8
The architecture of GFlow
National Cheng Kung University CSIE
Computer & Internet Architecture Lab
9
Parallel matching processing the GPU
National Cheng Kung University CSIE
Computer & Internet Architecture Lab
10
Experimental Environment






commodity PC
CPU : Intel I7 2600K @3.2GHz
• 4 cores/8 threads
Memory : 8GB DDR3-1333
GPU : Nvidia GTX 470 @607MHz
• 448 cores, 1280MB GDDR5 memory@3384MHz
OS : Fedora 18 (Kernel version 3.8.11-200)
The algorithm in GPU is implemented by
OpenCL
National Cheng Kung University CSIE
Computer & Internet Architecture Lab
11
Compared with Existing Approaches




Linear search used by OpenFlow vSwitch
Linear search with CPU parallel optimization
LightFlow, the GPU parallel optimization
GFlow, the work in this paper
National Cheng Kung University CSIE
Computer & Internet Architecture Lab
12
Experimental Result
National Cheng Kung University CSIE
Computer & Internet Architecture Lab
13
Experimental Result
National Cheng Kung University CSIE
Computer & Internet Architecture Lab
14
Experimental Result
National Cheng Kung University CSIE
Computer & Internet Architecture Lab
15
Download