System and Signal Monitoring for IPTV Set-

Top-Box Systems

Dávid Károly

Tivadar Szemethy

Árpád Bakay

The need for end to end monitoring Telcos are rolling out IPTV

3-play

Must ensure high availability, QoS, QoE

Importance of CPE monitoringLast mile problems

STB’s lack diagnostic interface

Contents

IPTV Network Architecture

Set top box architecture

MPEG-2 Transport Stream Monitoring

Implementation

Test results

Conclusions, future work

Customer PremisesVoD Cluster

Headend

IPTV Content Delivery System

Core Networ

k

STB

IAD

Access Network

Set top box architecture Hardware

RISC GPP Video decoding with DSP or ASIC Ethernet, SCART, IR interface

Software Embedded Linux Media player User interface, browser

Transport Stream Monitoring Raw UDP encapsulation

TS packetsHeader, Adaptation Field

PID’s, metadata, Elementary Streams

Timestamps: PCR, PTS, DTS

Transport Stream Monitoring (II) Characteristics to measure

Bandwidth, Packet loss Also per stream type (audio/video/metadata)

UDP-level Jitter, inter-arrival gap min/max/avgPCR characteristics

PCR Jitter PCR discontinuities, missing PCR

Naive implementation AF_PACKET raw sockets Packet filter Not feasible

InaccurateEats up CPU cycles, UI responsiveness is

hindered10 Mbps HDTV stream: 23% CPU usage

Packet capture inside the kernel Register a callback

Dev_add_pack()

Callback runs in softirq context

Gets skbuf with timestamp

Eliminates kernel to user copies

Results can be accessed via /proc filesystem

User mode portion SNMP agent based on Net-SNMP

Turn on/off with SNMP row creation/deletion Sends TRAP in cast of threshold violation

Repeatedly reads the proc-file Only aggregated values: averages, counters Read once per second

Can be remotely queried, turned on/off Also provides diagnostic parameters

Load average, free memory CPU load imposed by SNMP agent

Test results Tests conducted in two settings

Corporate intranet IPTV test network of a service provider

The approach is workable No extra CPU load imposed (even with HDTV, up to

10 Mbps) Accurate measured values, compared to a standalone

PC probe

Further findings

PCR behavior is codec-dependent

Usefulness of diagnostic parametersHeadend misconfiguration causes small

subtle errors (like metadata discontinuity)

This results in high STB load and frequent

reboots

Conclusions and future work Interpreting acquired data

Further experiments are needed Compare metrics with opinion score Data mining

Other parameters worth analyzing MDI per RFC1889 IGMP channel zap delay

Integrate into IAD or bridging probe

Thank you for your attention!Dávid Károly

David.Karoly at netvisor dot eu