Date post: | 16-Feb-2017 |
Category: |
Technology |
Upload: | maarten-verwaest |
View: | 2,503 times |
Download: | 0 times |
Background, challenges and Opportunities
Digital Media Prodution
© VRT-medialab: onderzoek en innovatie2
Table of Content
• 2004 – Background and Objectives• 2009 – Due Diligence
State of the technology Reconciliation of requirements Objectives and Challenges
– Operational Excellence– Scale up– Media Asset Management
• Additional requirements Broadband distribution Supply chain integration
• Technology trends
© VRT-medialab: onderzoek en innovatie3
2004...
© VRT-medialab: onderzoek en innovatie4
Background – Proof of Concept
medialab 5
Media FileMedia File
Media FileMedia File
Fully fragmented: Random access without using any Hard Disk intelligence
Cylinder
Sector
TracksHeads
SpindleArm
Platters
Arm assembly
Arm pivot
128 KB segment
4/8 KB segment Large segment size-> Higher efficiency Hard Disk
100+ HD’s// Striping:
Hard Disk Model
medialab 6
Cylinder
Sector
TracksHeads
SpindleArm
Platters
Arm assembly
Arm pivot
128 KB segment
4/8 KB segment Large segment size-> Higher efficiency Hard Disk
Full stroke8 KB segments
Full stroke128 KB segments
Full Stroke R/W 8KB – 128 KB Segments
Random8 KB segments
Random128 KB segments
Random R/W 8KB – 128 KB Segments
Inner track
R/W on the Inner Track
Outer track
R/W on the Outer Track
Hard Disk Model
medialab 7
Hard Disk Speed
15 Krpm calculated results10 Krpm calculated results128 KB segments8 KB segmentsTestresults
70 MB/s
50 MB/s
30 MB/s
10 MB/s
Read load types(1)Sequential Read
Outer tracks
(2)Sequential Read
Inner tracks
(3)Random Read
Tot. Fragmented disk128 KB / 8KB segments
(4)Worst case Read
Tot. Fragmented disk128 KB / 8KB segments
15Krpm – 10 K rpm 15Krpm – 10 K rpm 15Krpm – 10 K rpm 15Krpm – 10 K rpm
78.2 MB/s
71.5 MB/s
50.9 MB/s
39.4 MB/s
16.8 MB/s12.8 MB/s
0.91 MB/s 0.65 MB/s
12.2 MB/s8.9 MB/s
72.9 MB/s
66.7 MB/s
47.6 MB/s
35.4 MB/s
14.8 MB/s9.1 MB/s
0.92 MB/s 0.62 MB/s
DiskThroughput
=> 10 MB/s (SCSI/FC HD)(6.6MB/s for (S)ATA disk)
medialab
Storage Performance
3 Key Components:
MediaFile System
( GPFS, AVID-ISIS, Isilon, Omneon Media Grid, XSAN, … )
Hard Disk Network( Disk network, SAN )
100 %Efficiency
medialab 9
FC Network
Fileservers
Clients
Video
Proprietaire oplossingenSchaalbaarheid???
FC-client Based
Avid-Unity / Quantel / SGI-CXFS /
Thomson GV / …
IBM-GPFS / SGI-CXFS /HP-Lustre / ADIC-Stornext /
Nexsan / …
SAN
File servers cluster
Clients Clients Clients
Parallel accessfile system
MXF-AAF video/audio file
IP-client based
Media IP Network
GPFS (IBM)
(General Parallel File System)
Scalability
Reliability
Parallel Throughput
Storage Capacity
Media File System
© VRT-medialab: onderzoek en innovatie10StorageCapacity
StorageThroughput
Classical ITStorage
SD MediaStorageX
Architecture needs to be “tuned” for media production
File-based Media Production
© VRT-medialab: onderzoek en innovatie11
Digital Media Factory Storage Clusters
MCBE
GMI I
Taperobot
BUBE
Cluster3x4CPU
IP netwerk
Cluster3x2CPU
Cluster3x2CPU
Cluster3x2CPU
TSM2x2CPU
TSM1x2CPU
TSM1x2CPU
TSM2x2CPU
SATA mirror40 TBAudio
Other essence
FC mirror62,5 TB
News FeedsNews archive
FC mirror22,5 TBBrowse
SATA single56 TBWork
SANswitch
SANswitch
FC mirrorSATA mirrorSATA single
Taperobot
Taperobot
SANswitch
SANswitch
50 stromen x 50 Mbit/sReal-time
200 stromen x 1 Mbit/sBest-effort
Taperobot
StagingMCRT
Storage Cluster per Service Level
© VRT-medialab: onderzoek en innovatie12
Central storage
INGEST
MediaAsset Mgmt
IP network
CentralStorage
Draft Processing
Concurrentengineering
Browse Clients Processing duringingest/transfer
POSTProduction
Craft editors
Subtitling
Archiving
Analysis
Annotation
Follow-upcontinuity Editor-in-Chief
Post-sonorisation
PLAY-OUT
8 GB P2 card= 32 min videoDV25 (news)
Go Live - 17 June 2007
© VRT-medialab: onderzoek en innovatie13
Online searching, re-viewing and cutting
© VRT-medialab: onderzoek en innovatie14
One-click re-purposing of content
© VRT-medialab: onderzoek en innovatie15
Assertion
• There is a mismatch between business requirements and state-of-the-art technology;• Media technology suppliers don’t sufficiently understand the limitations of IT technology and conventional IT suppliers don’t have sufficient understanding of media production;• Proper research and development are essential...
© VRT-medialab: onderzoek en innovatie16
2009 - Reconciliation of Requirements
Production Supply Chain Automation Real-time co-production
Operational Excellence Simplify system architecture Standards implementation
Scale-out Extend to non-news Prepare for High-Definition
Non-linear distribution Integration of live and on demand Accurate Item-level metadata
© VRT-medialab: onderzoek en innovatie17
Objectives - Back-Office
Production Platform
Master Data Management
Production and distribution
Infrastructure – Networks and Storage
IngestMedia
Asset Mgnt
Editing
Playout
Mastering 2. Scale out Infrastructure Better applications reduce file transfers State-of-the-art technology - Cost reduction
1. Operational Excellence Less applications reduce system integration Metadata standards replace custom development
3. Master Data Management Rationale – Effective(!) re-use Structured Metadata Item-level metadata
© VRT-medialab: onderzoek en innovatie18
Challenge – Operational Excellence
Production Platform
Master Data Management
Infrastructure – Networks and Storage
Production and distribution
IngestMedia
Asset Mgnt
Editing
Playout
Mastering
© VRT-medialab: onderzoek en innovatie19
System Integration Approach
Assumptions Ad hoc plumbing (“data pump”) is a source of inconsistency; Collaborative production and re-use of material rely on a single central repository.
Supply Chain Engine
Issues None of the available state-of-the-art production applications have been designed to integrate
and neither they have been originally designed for file-based production; Commercially available MAM systems that lack essential functionality;
• While an Enterprise Service Bus is assumed to protect overall consistency and performance, it is sensitive to complexity as well and may become a source of instability on its own.
MAMsystem
The level of Operational Excellence that can be achieved by an Integrated System is determined by the extent to which individual applications have been designed to integrate and by the level of (metadata) standards implementation.
© VRT-medialab: onderzoek en innovatie20
Lessons Learned – Principles of System Integration
Everything should be made as simple as possible, but not one bit simpler.
© VRT-medialab: onderzoek en innovatie21
Challenge – Scale out Infrastructure
Production Platform
Master Data Management
Production and distribution
Infrastructure – Networks and Storage
IngestMedia
Asset Mgnt
Editing
Playout
Mastering
© VRT-medialab: onderzoek en innovatie2222
A single Ingest operation causes 36 file transfers!
GMII
Data TapeArchive
TSM2
TSM1
Data TapeBackup
TSM2
TSM1
IP network
Agility
Polopoly
Output
OT
OT
Nieuws Playout
Omneon
MER
Omneon
Ardome
MAM
ENCENCENCENCENCENCENCENCENCENC
Encoders
ENCENCENCENCENCENCENCENCENCENC
MCRT MCRT MCRT TSM
//news
//feeds /cut
/import
MCRT cluster
MCBE MCBE MCBE TSM
//lowres
//audio
//sec
MCBE cluster
BUBE BUBE BUBE TSM
/wip
/archive
/sec
BUBE cluster
IP network
Opslag//news//feeds Precut
/import
Ardex
Opslag/wip
/archive
Opslag//lowres//audio
Opslag/sec
Opslag//sec
DV25
D10
OPATOPAT
OPATOPATOPAT
OPATOPATOPATOPATOPAT
OP1A
OP1AOP1A
OP1A
OP1A
OP1A
OP1AOP1A
Lres
OP1A
Lres
OP1A OP1A
OPAT
OPAT
OP1A
OP1A
OPATOPATOPATOPAT
OPATOPATOPATOPAT
Browse Clients
Lres
Workflow vs Dataflow:
00010203040506070809101112131415161718192021222324252627282930313233343536
High Definition Media Production Infrastructure
© VRT-medialab: onderzoek en innovatie2323
High Definition Media Production InfrastructureHigh Definition Media Production Infrastructure
Agility
Polopoly
Output
OT
OT
Nieuws Playout
Omneon
MER
Omneon
Introduction: Issues – HD Digital Media Factory
HD Media Infrastructure Issues• Complex Workflow -> Inefficient Dataflows
• More media-formats: Transcoding becomes bottle-neck!
• Increased Storage Capacity Requirements: HD ~= SD x 2 -> HD Storage Architecture???
Backup/restore windows HD ~= SD x 2Reconsider # copies – storage protection
• Cost: Target HD ~= SD / 4 (not counting disk price erosion)
Geisha results IBC 2009
© VRT-medialab: onderzoek en innovatie24
HD Storage Architecture Requirements
StorageCapacity
StorageThroughput
HD MediaStorageX
Classical ITStorage
SD MediaStorageX
“Tuned” Fibre Channel Architecture
Geisha results IBC 2009
© VRT-medialab: onderzoek en innovatie25
High Definition Media Infrastructure
State-of-the-Art • Currently 160TB storage capacity (3200 hrs HD) per cluster• Demonstrated troughput 40Gbps• Built-in Disaster Recovery• CAPEX - 2.2€ per GB
© VRT-medialab: onderzoek en innovatie26
Objective 3 – Effective re-use
Production Platform
Master Data Management
Infrastructure – Networks and Storage
Production and distribution
Media Asset Management Issues• Production Environment
• No upfront selection causes information overload;• Multiple sources of more or less structured metadata;• Parallel sources cause dupplicates;• International context - multilingual
© VRT-medialab: onderzoek en innovatie27
Objective 3 – Effective re-use
Production Platform
Master Data Management
Infrastructure – Networks and Storage
Production and distribution
Media Asset Management Issues• Production Environment
• No upfront selection causes information overload;• Multiple sources of more or less structured metadata;• Parallel sources cause dupplicates;• International context - multilingual
• Professional Requirements• An item is relevant or it is not;• Retrieve all relevant items;• No false positives
• Opportunities• Recognise workflow and trace dependancies;• Integrate with Product Engineering (scripts,...);• Computer Vision techniques may help.
© VRT-medialab: onderzoek en innovatie28
The state of the Art…
archiefnummer : ALG 20010813 1fragmentnummer : 1 reeks : 1000 ZONNEN EN GARNALENbandnummer : E03024404formaat : DBCMfragmenttitel : 1000 ZONNEN & GARNALENbeeld : KL/PALPLUSfragmentduur : 18 20 tekst : 0'00" TOERISTISCH REPORTAGEMAGAZINE OVERZICHT ONDERWERPEN GENERIEK TOERISTISCH REPORTAGEMAGAZINE, OVERZICHT ONDERWERPEN 0'50" VANDAAG : KUNSTENAAR LUC HOFKENS ONTWIERP EEN OASE OP ZIJN DAKTERRAS IN BORGERHOUT DIE DOET DENKEN AAN DE GRAND CANYON INTERVIEW MET LUC EN ZIJN VROUW MARILOU BUITENBEELD DAK MET OMGEVING BUITENKANT ARBEIDERSWONING, PANO OVER ROTSWANDEN, KRATEN MET WATER, BEPANTING, FOTOALBUM MET VERLOOP WERKEN 4'00" JUNIOR : KLAARTJE ALAERTS, 13 JAAR WIL ASTRONAUTEN WORDEN ZE BEZOEKT HETEUROSPACE CENTER METRUIMTEVEREN, RAKETTEN SIMULATIE IN RUIMTEVEER, INTERVIEW, HEEFT EEN UFO GEZIEN MAAKT ZELF KLEIN RAKETJE, SCHIET HET AF 7'50" DE SCHEURKALENDER : ARCHIEF RECLAMEFILM IBM INTERVIEW MAURICE DE WILDE, EERSTE PERSOONLIJKECOMPUTER trefwoorden : BELGIE; BORGERHOUT; ARTIEST; OASE; KUNST; GRAND CANYON (NATUURGEBIED); DAK; TERRAS; INTERVIEW; EURO SPACE CENTER; RUIMTEVAART; PC; BOOTTOCHT; RIJKDOM; PASSAGIER; GASTRONOMIE; RESTAURANT; PERSONEEL; VAKANTIE; BINNENBEELD; SCHIP; BECKERS LEEN; VRT; LOTTO; RADIOOMROEPSTER; KLANKSTUDIO; UITVINDING; BARBECUE; BETONMOLEN; IBM; RECLAMESPOTrechthebbende : VRT
Opzoekscherm FILM Set: 16 Aantal: 1blz 1 van 3 trefwoorden: ibm and vrt archiefnummer: - uitzendjaar: maand: dag: fragmentnummer: fragmentduur: reeks: formaat: bandnummer: aflevering: afleveringsnummer: programma: uitzenddatum: fragmenttitel: tekst: kategorie: opnamedatum: opnamenummer: journalist: rechthebbende:
SETS The strings required for the operation are not defined
F11 F12 F13 F14 F17 F18 F19 F20 Ent Eindigen Sets Refset Toon Vorige Volg/Leeg Thesaurus Commando Opzoeken
© VRT-medialab: onderzoek en innovatie29
Shot Segmentation and Scene Recognition
© VRT-medialab: onderzoek en innovatie30
Face recognition and identification
© VRT-medialab: onderzoek en innovatie31
Media Asset Management - Opportunities
An integrated data-model and a consistent application framework are the key to production
automation and supply chain integration
!
(≠ Digitizing analogue and disintegrated workflows)
• Professional context - trace dependencies
• Scripts contain essential master data – integration with creative processes is a must
• Computer Vision does not replace formal annotation, but substantially accelerates the process and increases the accuracy
© VRT-medialab: onderzoek en innovatie32
Conclusion
Production Platform
Master Data Management
Production and distribution
Infrastructure – Networks and Storage
IngestMedia
Asset Mgnt
Editing
Playout
Mastering
2. Scale out Infrastructure Avoid file transfers Cheaper systems for cost reduction
1. Operational Excellence One set of numbers – reduce the number of apps Less apps reduce the need for system integration Replace custom development by proper standards
3. Master Data Management Trace dependancies Integrate with creative processes Explore and integrate automatic analysis tools
© VRT-medialab: onderzoek en innovatie33
© VRT-medialab: onderzoek en innovatie34
Broadband distribution
Production Platform
Master Data Management
Infrastructure – Networks and Storage
Production and distribution
D a t a G e n e r a l
D a t a G e n e r a l
D a t a G e n e r a l
D a t a G e n e r a l
D a t a G e n e r a l
D a t a G e n e r a l
MetaData
MetaData
Communication(Information)
D a t a G e n e r a l
Data
Distribution via DVB, IP or other
Video Audio
© VRT-medialab: onderzoek en innovatie35
2004 – Live Streaming
© VRT-medialab: onderzoek en innovatie36
Xmas 2007 – BBC iPlayer
© VRT-medialab: onderzoek en innovatie37
Broadcast...
A customer can have a car painted any colour that he wants so long as it is black.
(Henry Ford)
© VRT-medialab: onderzoek en innovatie38
Good, better,...
Anywhere• “Live streaming”• Radioplayer software application• Quick win – “in series” live transcoding
Anytime• “Podcast”, “iPlayer”• On demand complements live offering• Major impact – metadata power-lifting
The Ultimate Radio Exp - myRadio• Mobile - Anywhere, anytime and impulse-driven• Seamless integration of live and on demand• Cross-platform continuity?
© VRT-medialab: onderzoek en innovatie39
The Ultimate Radio Exp(eriment)
myRadio @time of writing• Excellent quality where possible, lower bit-rate if necessary• Non-linear distribution assumes broadband distribution• Open distribution framework• Http adaptive streaming enables bandwidth-friendly distribution
http://myradio.vrt.com
© VRT-medialab: onderzoek en innovatie40
Ongoing – Non-linear distribution
Seamless integration of Live and On demand• “Bookmark” by tagging• Pause, Back and Fast Forward • Radio search engine• Assemble myRadio
TO DO – Track each individual item• Audio segmentation • Speech recognition and indexing• Metadata harvesting (!!!)
© VRT-medialab: onderzoek en innovatie41
Business Case for Radio
IP Distribution• +/- 5.000.000 hours per dag• à 128 kbps, 250 TB, 90PB per year• à 70.000€ per PB• Total – 6.300.000€
Vgl. Homepage (deredactie.be)• 2MB• 128 seconden high-def audio(!)
© VRT-medialab: onderzoek en innovatie42
‘Supply Chain Integration’
D a t a G e n e r a l
Data
InformationCommunication
Distribution to TV, PC and mobile via DVB and IP
Video Audio
InformationMgnt
Infrastructure(network and storage)
Production and distribution
Media Asset Mgnt
© VRT-medialab: onderzoek en innovatie43
Trend – Production System Virtualisation
© VRT-medialab: onderzoek en innovatie44
Trend - From « metadata » to CAD/CAM
?
© VRT-medialab: onderzoek en innovatie45
Proof of Concept - Scoop
http://www.youtube.com(search for “scoop”, “ibbt”, “CAD”)
© VRT-medialab: onderzoek en innovatie46
Change...
© VRT-medialab: onderzoek en innovatie47 47
http://medialab.vrt.be
http://myradio.vrt.be