Initial work on quality of service and performance management of ATM networks was based on cell-level metrics such as cell loss ratio (CLR), cell delay variation (CDV), and cell transfer delay (CTD). It has now been realized that these metrices are not meaningful for most of Internet traffic. Users view of performance is frame level and so ATM Forum has recently started working on defining frame-level metrics. These metrics and how they will help service providers measure their own performance will be explained.

ATM networks are unique in the sense that receiving a beginning part (or cell) of the frame does not guarantee that the ending part (or cell) of the frame will be received at the same time. Thus, previously established latency metrics such as "First-bit in to First-bit out (FIFO)" can not be used. We have developed a new latency metric called "MIMO," which applies even when frames are discontiguous at the input or output or both.

This talk covers the following topics:

• Dictionary Definition
• Definition
• Other Networking Benchmarks
• Why Do This at ATM Forum?
• Cell Level vs Frame Level
• SCOPE: Goals
• Non-Goals
• OSU National ATM Benchmarking Lab
• Metrics
• Throughput
• Connection Configurations for Foreground Traffic
• Scaleable Test Configurations
• Scaleable Test Configurations II
• Latency Measurement: Overview
• Latency: Single Bit
• Latency: Single Cell
• Frame Latency
• Latency Definition: Requirements
• Simple Switches
• Zero-Delay Switch Input Speed = Output Speed
• Zero-Delay Switch Input Speed > Output Speed
• Zero-Delay Switch Input Speed < Output Speed
• Non-Zero Delay Switches
• Summary: Single Cell
• Discontigous Frames: Summary
• Frame Latency from Cell-level Data
• Scaleable Test Configuration
• Performance Metrics
• Summary
• References

Download complete presentation in Adobe Acrobat Format