fast forwarding table lookup exploiting gpu memory architecture author : youngjun lee,minseon...

15
Fast forwarding table lookup exploiting GPU memory architecture Author : Youngjun Lee,Minseon Jeong,Sanghwan Lee,Eun-Jin Im Publisher : Information and Communication Technology Convergence (ICTC), 2010 International Conference on Presenter: Gang Chi Date: 2015/1/7 Department of Computer Science and Information Engineering National Cheng Kung University, Taiwan R.O.C.

Upload: ashley-gilbert

Post on 29-Dec-2015

228 views

Category:

Documents


0 download

TRANSCRIPT

Fast forwarding table lookup exploiting GPU memory architecture

Author : Youngjun Lee,Minseon Jeong,Sanghwan Lee,Eun-Jin Im

Publisher : Information and Communication Technology Convergence (ICTC), 2010 International Conference on

Presenter: Gang Chi

Date: 2015/1/7

Department of Computer Science and Information Engineering National Cheng Kung University, Taiwan R.O.C.

Introduction

Packet Shader, a software router system, uses GPU to exploit GPU’s extensive parallelism, shows higher performance compared to other existing software routers.

However, Packet Shader does not utilize the memory architecture in the GPU system.

This paper provides a method to exploit GPU memory architecture.

National Cheng Kung University CSIE Computer & Internet Architecture Lab

2

GPU Memory Architecture

National Cheng Kung University CSIE Computer & Internet Architecture Lab

3

Type Location Access cycle Size

Global memory Off-chip >100 1~32GB per GPU

L1 cache On-chip 1~32 16 or 48KB per SMX

L2 cache On-chip 1~32 64KB per SMX

Registers On-chip n/a 32-bit x 65536 per SMX

GPU Memory Architecture

Memory type in GPU Global Memory : high bandwidth,large size,

but need more cycle to access. Constant memory ,texture memory : two

small pieces of memory and can be cached by on-chip read-only cache of each SMX.

National Cheng Kung University CSIE Computer & Internet Architecture Lab

4

Fast Forwarding Table Lookup

Store the whole forwarding table in the global memory.

Create index in the constant memory(64KB).

National Cheng Kung University CSIE Computer & Internet Architecture Lab

5

Fast Forwarding Table Lookup

Currently the number of active BGP entries is about 340,000.

The number of prefixes with length of 16 or less is 4.18%,24 is about 52%.Thus we expand all the prefixes of length less than 24 into multiple /24 prefixes.

The number of prefixes is still less than one million(needs 20 bits to identify).

National Cheng Kung University CSIE Computer & Internet Architecture Lab

6

Fast Forwarding Table Lookup

National Cheng Kung University CSIE Computer & Internet Architecture Lab

7

Fast Forwarding Table Lookup

National Cheng Kung University CSIE Computer & Internet Architecture Lab

8

Data Structure

National Cheng Kung University CSIE Computer & Internet Architecture Lab

9

Data Structure

National Cheng Kung University CSIE Computer & Internet Architecture Lab

10

Data Structure

National Cheng Kung University CSIE Computer & Internet Architecture Lab

11

Example

National Cheng Kung University CSIE Computer & Internet Architecture Lab

12

GPU Spec (1/2)

National Cheng Kung University CSIE Computer & Internet Architecture Lab

13

GPU Spec (2/2)

National Cheng Kung University CSIE Computer & Internet Architecture Lab

14

Exprimental Result

National Cheng Kung University CSIE Computer & Internet Architecture Lab

15