CouchConf Israel MediaMind Customer presentation

© 2011 MediaMind | A Division of DG | All rights reserved

December 2011

Efi Cohen | Vice President, Technology

MediaMind Overview

© 2011 MediaMind | A division of DG | All rights reserved

Search

Display

Networks

Emerging

ConsumersMedia Suppliers

Challenges in Migrating to Digital Advertising

?

Agencies

Industry Challenges

• Fragmentation

• Noise

• Inefficiency

Media Agency

Creative Agency

Advertisers

© 2011 MediaMind | A division of DG | All rights reserved

Media Agency

Creative Agency

Advertisers

Search

Emerging

Networks

Display

MediaMind: Addressing Digital Advertising Challenges

ConsumersAgencies MediaMind Media Suppliers

Resolves fragmentation

Overcome noise

Integrated reach

Impact & relevancy

Optimization Addresses inefficiency

© 2011 MediaMind | A division of DG | All rights reserved

\

How Is MediaMind Different?

Uniquely Positioned at a Critical Juncture Point

MediaEngagementData

Apply data directly to the consumer experience, across all touch points

© 2011 MediaMind | A division of DG | All rights reserved

Online Marketing Suite

MediaMind Blocks

MediaMind Workshop

MediaMind Analytics

Channel Connect

Smart Trading Smart Planning Smart Versioning

DeveloperTools

Tracking& Analytics

DemandSide

Platform

AdServing

Planning& Buying

DynamicCreative

Rich Media

Standard Serving

In-stream Video

MediaMind Mobile

© 2011 MediaMind | A division of DG | All rights reserved

Massive Scale

I’m still concerned about my account

I’m not convinced

Little better

Ok, that’s big

Ok Ok…I get it, scale isn’t

an issue

Countries Served : 63

Requests per Second: 65,000

Network usage: 20Gbps / 4PB (month)

Daily Impressions Served: 4.5 Billion

Daily log recors : 6 Billion / 500GB

Active Unique Users: 750 Million +

Advertisers: 9000 +

Up time: 99.99%

© 2010 MediaMind Technologies Inc. | All rights reserved

Global Infrastructure

BeijingNew York

AmsterdamNew Jersey

TokyoLos Angeles

Singapore

Media content servers owned by our CDN (AKAMAI) in more than 70 different countriesAd serving data centers in 7 locations (NJ, LA, Amsterdam x 2, Beijing, Tokyo and Singapore)Campaign Management and backend databases data centers in 2 locations (NJ, NY)

© 2011 MediaMind | A Division of DG | All rights reserved

Orit Alul | R&D Group Manager

December 2011

Real Time User DB

© 2011 MediaMind | A division of DG | All rights reserved

▸ What are our business requirements?

▸ What are our technical requirements?

▸ What are our assumptions?

▸ What is our solution?

▸ Q&A

Agenda

© 2011 MediaMind | A division of DG | All rights reserved

What are our business requirements?

▸ Unlimited user data storageAvoid http cookie limitations (such as: size, encoding, scale out)

▸ Real time bidder compatibilityProcess requests in less than 5ms

▸ Leverage our offline user data processing

▸ 3rd party data provides interoperabilityi.e. using the advertiser CRM user level information for retargeting and segmenting users

▸ Decrease the cost of traffic due to sending cookies back and forth

© 2011 MediaMind | A division of DG | All rights reserved

What are our technical requirements?

▸ Key/Value storeThe user id will be kept in the http cookie.

▸ Low latency of reads/writesOur web servers process requests in about 2-3ms.

▸ Get/Set relation of 1:1

▸ Horizontal scaleIn terms of size and performance.

▸ High Availability

Persistency and fully redundancy in both the DC level and across multiple DCs.

© 2011 MediaMind | A division of DG | All rights reserved

What are our assumptions?

1. We can afford a model of eventual consistency.

2. We can keep only the active users in memory.

Disk larger than memory attitude.

3. We can assume users' stickiness in the continent level.

© 2011 MediaMind | A division of DG | All rights reserved

What is our solution? – Architecture class Deployment Diagram

DC 1

.........

«web server»BS 1

RT Users DB 1

«web server»BS 2

«load balancer»LB 1

«load balancer»LB 2

«load balancer»LB m

«web server»BS n

RT Users DB 2

RT Users DB k

.....

....

© 2011 MediaMind | A division of DG | All rights reserved

What is our solution? - Software

1. Using Couchbase(Membase) server

2. Using C# Enyim Caching client

"Smart" client.

3. Adding performance counters stats service

To be aligned with our reporting and monitoring systems.

4. Adding DC replication (in process)

© 2011 MediaMind | A division of DG | All rights reserved

What is our solution? - Hardware

• A cluster of symmetric servers with the following setup each:6X 120 GB SSD drives2X 300 GB spinning disks96 GB RAME55 dual quad CPUOS: Windows server 2008 enterprise R2 x64

© 2011 MediaMind | A division of DG | All rights reserved

What is our solution? - Performance

▸ Average latency of ~0.4-0.7ms per operation (set/get)(Based on pilot running in one of DCs)

▸ Maximum throghput of 30K-35K operations per second per node.

(In our labs)

© 2011 MediaMind | A division of DG | All rights reserved

What is our solution? - Performance

Read/Write avg latency

© 2011 MediaMind | A division of DG | All rights reserved

What is our solution? - PerformanceRequests per sec Disk fetches per sec

CPU consumptionRead/Write avg latency

© 2011 MediaMind | A division of DG | All rights reserved

Questions?

© 2011 MediaMind | A division of DG | All rights reserved

Thank you!