CNN Case Study Deploying in a UNIX

CNN Case Study: Deploying ™ in a UNIX e. Directory Environment www. novell. com Steve Brunton Chief Engineer CNN Internet Technologies brunton@web. turner. com

Today’s Agenda • Defining “the personalization problem” • • that dot coms face CNN business and technical personalization requirements Reasons CNN chose Novell e. Directory™ for their solution Testing, implementing and supporting Novell e. Directory at CNN Summary, Q & A

Defining the personalization problem

What Is Personalization? In regards to web services • Targeting advertising • Customizing layouts • Delivering personalized content • “You don’t visit a web site, the web site visits you!”

Why Personalization? • Increase advertising revenue • Build brand loyalty • Maintain “stickiness” of browsers • Reduce the information overload by delivering only relevant content

Why Use a Directory Service? “The directory has spent the last 10 years in relative obscurity, as a sort of isolated application to generate user phone numbers and user IDs, or buried inside of things like a network operating system…. What changed all this was the Internet. What changed it even more than the Internet was the concept of personalization. ” — Richard Villars, IDC

CNN business and technical requirements

CNN Requirements • Extremely high-performance 4 Just downright FAST • Extremely scaleable 4 Handle more than 100 web servers with a few directory servers • Extremely flexible 4 Ability to change “how” the data is stored with no service interruption • Extremely fault tolerant 4 On line maintenance, fail-over, repair, determinism (isolationism)

CNN Requirements (cont. ) • 100% open standards-based 4 Fully RFC-compliant LDAPv 3 4 Easy to develop to (open SDK) • Buy, not build 4 Shrink-wrap vs. internally developed 4 No code to maintain internally • Cross-platform 4 Ability to choose best suited OS 4 Could develop and deploy on different platforms if desired

The Technology Selection • Understanding the CNN architecture and design philosophy • Understanding the performance and feature requirements • Evaluating the requirements against product capabilities • Capacity testing

Architecture Philosophy (Simple) Web Server Queues HTTP Requests Application Call Outs

Architecture Philosophy (Scaleable) Web Server Queues HTTP Requests Application Call Outs

The CNN “Challenge”

Performance Requirements • 2000 requests per second • Greater than 99% success of all requests • Less than 250 ms latency in the search • Load balance all requests • Graceful degradation of service 4 “World event” surge in traffic 4 More than two million requests in an hour

Feature Requirements • Ease of implementation and support 4 Start to finish in weeks 4 99. 999% uptime to date • Suitability to task 4 The technology matches the job • Ability to connect to anything 4 Via Dir. XML™, options are limitless • Quality of data 4 Integrity and speed of replication

Requirements vs. Capabilities • Open Standard (LDAP client) 4 Asynchronous support 4 Load balancing and determinism • Extreme Performance (e. Directory Cache) 4 Efficient data structure stored in memory and “managed” well • Ultimate Flexibility (e. Directory Database) 4 Prefer hierarchical (XML store) 4 Very efficient disk reads/writes 4 Ability to “back out” transactions

Division of Duties • CNN 4 LDAP client • Novell + + = Directory engine Directory cache Directory database Novell e. Directory

Testing Novell e. Directory

Novell Super. Lab Testing • Steps to building the “test” 4 Find “killer” directory, methodology and testing knowledge • Enter Novell Consulting 4 Find similar architecture to CNN Web Farm (to measure load/performance) • Enter Novell Super. Lab 4 Transplant to a pilot on the CNN Web Farm for “real-world” testing • Enter Novell Systems Engineering

Novell Super. Lab Testing • The Test 4 Fifty NT Workstations running ten processes each of an LDAP query 4 One single processor Net. Ware® Server hosting e. Directory with eight million objects • Results 4 More than 500 requests per second 4 Less than 60 ms latency with 100% hit rate 4 98 -100% CPU utilization until cache warmed— 4 -6% after

Implementing Novell e. Directory at CNN

CNN Novell Architecture (ad system helpers) Netscape Web Servers on Solaris (CNN Web Farm) (Cookie) HTTP Novell e. Directory on Net. Ware Development Server Internal Firewall (Thread-safe, load balanced NSAPI plug-in LDAP Client) Novell e. Directory on Solaris Master Replica Novell e. Directory on Solaris Partition R/W Replicas

Novell e. Directory @ CNN “Data Flow” 1. User accesses CNN web site 2. Cookie identifies the user 3. Web Server “looks up” profile from Novell e. Directory via LDAP 4. Web Server uses profile data (XML format) to customize the HTTP stream to the browser and/or make call out to ad injection system

Supporting Novell e. Directory at CNN

Supporting the Solution • Backup of the directory 4 Rely on inherent e. Directory replication 4 Scheduled export to LDIF • Directory maintenance 4 Automatic • Nightly reboots • NOT! • Remember, we’re running this on Net. Ware and Solaris!