In a talk given to Nation Cable Television Association I describe a new model for caching and packaging content in an adaptive bit rate multi-format Content Delivery Network.
Intelligent Caching and Packaging in an ABR Multi-format CDN
1. Intelligent Caching and Packaging
in an ABR Multi-format CDN
Brian Tarbox
Distinguished Member of the Technical Staff
Motorola Mobility
2. Caching was not needed
when VOD was first launched
When VOD was in its infancy
• One central server supported an entire node
• Content was offered only in SD format
• A library consisted of hundreds of titles
• Caching was not necessary
VOD SERVER
3. EDGE CACHE
EDGE CACHE
EDGE CACHE
As VOD libraries grew, caching became key
Content grows
• Library with edge cache
• HD formatted content
• Thousands of titles
• LRU evolves to Intelligent Caching
ORIGIN SERVER
4. Caching Algorithms
Caching: predictive activity that
attempts to use past behavior to predict
future behavior
Critical decision in caching is determining
what content to remove
LRU removes content with oldest use time
regardless of usage pattern
Intelligent Caching tracks actual viewing
time in the aggregate, discounting passive
or initial viewing behavior
Involves multiple dynamic parameters
Draws statistical inferences on potential
for future likelihood of additional views
5. LRU
CACHE
ORIGIN SERVER
CDN
With introduction of CDNs, caching is rolled
back to LRU
Content Explodes
• CDNs introduced
• Adaptive bit rate chunks
• Millions of titles, billions of chunks
• Caching algorithm reverts to LRU
6. What affects predictability?
Diminishing Predictability
• Multiple formats
• Multiple bit rates
• nDVR unique copy
• Personalization
– Targeted advertising
– Blackout
Enhancing Predictability
• Content Affinity
– Based on existence of many
edge streamers
– All stream requests for the
same content are directed to
the same edge streamer
• Save disk space
• Save bandwidth
– Employs Cluster Manager to
route content requests
There are many influences on the predictability
of future viewing of content
7. ORIGIN SERVER
CDN
EDGE CACHE
EDGE CACHE
EDGE CACHE
EDGE CACHE
CLUSTER
MANAGER
Intelligent Caching with affinity
By streaming the same content
from the same cache, the counts
increase and network utilization is
decreased
8. Comparison:
Intelligent Caching vs. LRU in CDN
Intelligent Caching
• Uses content affinity to intelligently
store chunks at the edge
• Uses multiple parameters to assess
true value of chunks stored at the edge
• More efficient
LRU Caching
• No intelligence about content, its
placement or its usage
• Determines which chunks were least
recently used and removes them
CDN
INTELLIGENT
CACHE
CLUSTER
MANAGER
EDGE
SERVER
EDGE
SERVER
CLIENT
CLIENT
CLIENT
CLIENT
EDGE
SERVER
EDGE
SERVER
CLIENT
CLIENT
CLIENT
CLIENT
EDGE
SERVER
EDGE
SERVER
CDN
LRU
CACHE
9. Caching is also Affected by Packaging options
TRANSCODER PACKAGER ORIGIN CDN CLIENT
TRANSCODER PACKAGERORIGIN CDN CLIENT
TRANSCODER
TRANS
WRAPPERORIGIN CDN CLIENTCHUNKER
Common, cacheable Personalized, not cacheable