A Reliable Internet J. Bradley Chen Chief Technology Officer Appliant, Inc.
© Appliant, Inc. All rights reserved. The Problem One of these things is not like the other… Reliability
© Appliant, Inc. All rights reserved. What’s the difference? Compared to “mature” consumer technologies, the Internet demonstrates: poor reliability poor reliability high complexity high complexity a pervasive lack of accountability a pervasive lack of accountability Many of these differences are related to the federated nature of the Internet.
© Appliant, Inc. All rights reserved. Accountability in a Federation Who’s responsible for your web experience? Example: Altavista web and content servers Lots of other equipment Akamai (image service) Doubleclick (ad service) Level3 (Altavista middle-mile) Sprint (Akamai middle-mile) Alternet (Doubleclick middle-mile) InterNAP (Doubleclick first-mile) (last-mile network link) Microsoft IE5.5 Browser on Windows NT
© Appliant, Inc. All rights reserved. Accountability: Solution Option 1: Test-Based Option 1: Test-Based Periodic experiments test the correct operation along a specific path through the system. Option 2: Monitor-Based Option 2: Monitor-Based Observation of ongoing system activity exposes system behavior and irregularities. These options are complementary.
© Appliant, Inc. All rights reserved. A Superficial Comparison CoverageRepeatabilityData Volume Test Only covers problems included in tests Complete coverage is extremely expensive GoodLow to moderate Monitor Can reveal any problem for any page delivered to an end-user Marginal coverage during periods of low activity Marginal for managing availability Good to poor; depends on traffic volume Can be huge!
Monitoring Example: Appliant’s Lateral Line Service
© Appliant, Inc. All rights reserved. Internet cloud Appliant Lateral Line End-users Internet cloud Firewall Loadbalancers Webservers Appservers Database Data Center Summarize Summarize Report Report Alert Alert Configure Configure Collect data Collect data Communicate Communicate CachingProxy Net
© Appliant, Inc. All rights reserved. System Architecture Data Center Web User RMON Legend:LL Core Data Store Schema+SP Web Server SMON APPD Transport Business Data Manager Rules Online Reporting Config Offline Reporting Gen Customer Site Export
© Appliant, Inc. All rights reserved.
Case Study A CDN Problem 27 January 2001
Case Study Documenting CDN Performance July 2001
© Appliant, Inc. All rights reserved. Parting Thoughts A reliable Internet is possible essential. A reliable Internet is possible essential. Progress demands more accountability. Progress demands more accountability. The Internet is a federation. Without accountability, it could be a long wait. Analysis of data from detailed system monitors can expose many of the problems we overlook today. Analysis of data from detailed system monitors can expose many of the problems we overlook today.
© Appliant, Inc. All rights reserved.