The Network Performance Advisor

1. The Network Performance Advisor J. W. Ferguson NLANR/DAST & NCSA

2. Acknowledgements • NLANR/DAST does its work on ‘The Advisor’ under a cooperative agreement with the National Science Foundation • The Internet2 E2E Initiative and the piPEs project have provided support and encouragement throughout • The Network Measurement Working Group of the Global Grid Forum, for the schema work being done to allow data sharing between measurement projects

3. Advisor Overview Targeted for end users & network, the Advisor measures, displays, and analyzes network metrics • Uses existing diagnostic tools by integrating them into a common framework • For network engineers & administrators, provides an easy to use interface to view network metrics, and customize which metrics you like to monitor • For end users, Advisor attempts to emulate a junior-level network engineer with its Analysis Engine • Uses the schema in development by the GGF’s Network Measurement Working Group. Will use newest version of this schema—demo at Boston GGF meeting in October. • Additional tools and analyses can be added easily • http://dast.nlanr.net/projects/advisor

4. Current Status • Version 2.1 released September 2005! • ‘Bundles’ included in release include Iperf, OWAMP, ping, top, pathchar, ifconfig, traceroute, netstat, and pathload • All code is accessible via anonymous CVS.

5. Architecture Performance Data Collector (PDC) • Gathers network performance data Performance Data Historical Archiver (PDHA) • Archives network performance data Analysis Engine • Analyzes network data • Provide plain text advice to solve problems or increase performance • User Interface • Expert Interface: table & tree of metrics • Map Interface: graphical display of network • Analysis Interface: interact with Analysis Engine • All components written in Java and use XML-RPC

6. Performance Data Collector (PDC) • Designed to be stand-alone, extensible, and portable • Elegantly handles platform differences and unavailability of any given measurement • Overview of Features • Uses bundlesto facilitate integration of performance data measurement tools • A collection of scripts or Java classes that describe: • How to invoke a measurement tool • What metrics the measurement tool measures • How to parse the measurement tool's output • Implements an XML-RPC interface • getAllMetrics: returns the list of metrics that may be measured • getMeasurement, getMeasurements, getAllMeasurements: returns an individual, a list, or all measurements given a remote host

7. PDC Features • All requests are fulfilled immediately without any caching • Activation • Allows cooperation of both ends through a mechanism called activation (i.e. for tools such as Iperf) • Security • Using SSL and username/password (more to come) • Autoupdating • Periodically updates the bundles (automatically) • User can set how often to check for updates • All system bundles updated • Tool to update bundles on demand • Global Grid Forum NM-WG Response Schema • Full Java class support for reading and writing in the schema • Bundle support for NM-WG Response Schema • New Version 2 of Schema necessitates code changes

8. Performance Data Historical Archiver Short to medium-term storage of PDC measurements • Overview of Features: • Act as a caching proxy for the PDC • Specify how much data to store, and for how long. • Utilizes an XML-RPC interface to retrieve data • Clients can retrieve old measurement results and the latest measurement results • Clients can force PDHA to request new measurements from PDC • To retrieve archived measurements specify an • interval: returns all measurements taken during the specified interval • list of timestamps: returns all measurements that match a timestamp in the list, with some amount of error allowed. • Allow different measurements to have different “lifetimes” • Stores data locally in an XML file • Ability to query third party databases with the NM-WG Request schema • Will support Version 2 of schema, demo at GGF in October

9. Analysis Engine Analyze metrics for a specific end-to-end path and give advice to solve performance or connectivity problems • Features: • “Test Definition Files” (TDFs) similar to PDC’s ADFs • TDFs consist of • detection rules: simple binary operations (AND, OR, NOT, etc) • synthetic metrics file name • problem descriptions, and solutions • Synthetic Metrics • operations on base metrics • may be written in any scripting language (and eventually Java) • Constructs a decision tree for each problem

10. Analysis Engine • Example TDFS include: • Duplex mismatch, incorrect buffer sizes, incorrect settings, congestion, general connectivity issues. • Future Development: • Modify Analysis Engine to use historical data • Engage the network engineering community to obtain more analysis test cases

11. Advisor GUI Displays Advisor’s collected network metrics and is capable of displaying outside metrics acquired in the GGF NM-WG schema. • Three main graphical displays • Metric Display GUI: Displays all metrics • uses a tree to organize metrics • uses a table to display metrics and corresponding information • Analysis GUI: Displays advice reported from the analysis engine • text display of analysis and advice • Map GUI: Visual display of the network and trouble spots • users will be able to click on a map to view measurements for specific areas along the path

12. Thank You ! dast.nlanr.net/projects/advisor