Reducing Server and Network Load with Shared Buffering

Reducing Server and Network Load with Shared BufferingSomaya Arianfar, Pasi Sarolahti, Jörg OttAalto University

10.12.2012CSWS’12, Nice, France

Reducing Server and Network Load with Shared Buffering

© Somaya Arianfar 2012

Outline

• Background and problem statement• Goals• Design

– Byte stream buffering– Store-me bits– Load Control Architecture – Receiver-based Transport Protocol

• Summary

2



Background

3

Server

Client 1

Client 2

Client 3

Bottleneck

Flow 1 proxy and serverFlow 2 Client 1 and proxy

Flow 3 Client 2 and proxy

Flow 4 Client 3 and proxy

Flow 5 Client 4 and server

Proxy Server

Client 4

Object A

Object A



Background

4

Client 1

Client 2

Client 3

Client 4

Server

CDN



Problem Statement

• Sudden traffic spikes

– Application specific solutions

– No CDN agreements => server goes down• E.g. links from Dig and Twitter

– Proxy’s violation from end-to-end argument• E.g. Security issues

5



Main goals

• Application independence

• Object size independence

• Segmentation and network independence

6



Design: Byte Stream Buffering

7



Design: Store-me bits and load control

8

Server

Client 1 PolicerBuffering node

Store-meStore-meAck Store-meStore-me OverloadAck, Overload Store-meOverloadAck, Overload



Design: Receiver-based Transport Protocol

• Stream label– 20-bytes unique Id

• Sequence Number– 64-bits offset

• Checksum– Cumulative from the beginning of the stream

9



Application Interface

10

Server

Client 1

Get(fd, “http”, “address.com”, <label> )

Request(“http”, <label>, 0-1460 )

Write(Data,0-1460 )

Data(“http”, <label>, 0,<data>)

Read(fd, <data>)



Initial Evaluations

11



Summary

• Storage sharing in the network

• Application independent stream buffering

• Load controlled architecture

• Pull-based transport protocol

12

Questions?

Thanks!

Reducing Server and Network Load with Shared Buffering

Documents

Transcript of Reducing Server and Network Load with Shared Buffering