7/28/2019 AWS Ac Ra Adserving 06

1/1

SystemOverview

3The click-through servers are a group ofAmazon EC2instances dedicated to collecting click-through data.

This information is contained in the log files of the click-through web servers, which are periodically uploaded toAmazon S3.

ADVERTISEMENTSERVING

Amazon

EC2

Amazon

EC2Sp

ot

Amazon

S3

Amazon

Impo

rt/Ex

port

AWS

Referen

ce

Archite

ctures

5Data processing results are pushed back into AmazonDynamoDB, a fully managed NoSQL database service

that provides fast and predictable performance with seamlessscalability. Amazon DynamoDB tables can store and retrieveany amount of data, and serve any level of request traffic,both of which are specific requirements for storing and quicklyretrieving visitors profile information.

The high availability and fast performance of AmazonDynamoDB enable ad server front-ends to serve requestswith predictable response time, even with high traffic volumesor large profiles data sets.

Amazon

EMR

3

2

AmazonD

ynam

oDB

Click-th

rough

Servers

5

AmazonS3Sta

ticFiles

Reposit

ory

Ad servinginfrastructure

Click-throughinfrastructure

SpotInstances

Amazon

Elastic

MapRedu

ce

Click

-through

Requests

Auto

Scaling

Profiles

Databa

se

Amazon

DynamoD

B

AdSer

vers

Amazon

EC2

Internet advertising services need to serve targeted advertisingand must do so under limited time. These are just two of multipletechnical challenges they face.

Amazon Web Services provides services and infrastructure tobuild reliable, fault-tolerant, and highly available ad servingplatforms in the cloud. In this document, we describe the two mainparts of such a system: ad serving infrastructure and click-throughcollection featuring a data analysis cluster.

4ElasticLoad

Balancing

Amazon

CloudFront

Content

Delivery

Networ

k

Visitor

Advertisement

(image/video)

ImpressionLogFiles

1When visitors load a web page, ad servers return apointer to the ad resource to be displayed. These

servers are running on Amazon Elastic Compute Cloud(Amazon EC2) instances. They query a data set stored in anAmazon DynamoDB table to find relevant ads depending onthe user's profile.

2Ad files are downloaded from Amazon CloudFront, acontent delivery service with low latency, high data-

transfer speeds, and no commitments. Log information fromdisplayed ads is stored on Amazon Simple Storage Service(Amazon S3), a highly available data store.

2

Click-th

rough

LogFile

s

4Ad impression and click-through data are retrieved andprocessed by an Amazon Elastic MapReduce cluster

using a hosted Hadoop framework to process the data in aparallel job flow. The cluster's capacity can be dynamicallyextended using Spot Instances to reduce the processingtime and the cost of running the job flow.

Elastic

LoadBalan

cing

Aut

oSc

aling

Amazon

Clo

udFr

ont

1