This session looks at the technical feasibility us using terrestrial delivery via the AWS cloud as an alternative to satellite transport. The key questions addressed include: can it be implemented without effecting the underlying media layers and can it be architected for low cost at scale. The presentation includes sample architectures and design considerations to achieve content distribution within acceptable latency parameters of satellite.

1. Shoot the Bird
Linear Broadcast Distribution on AWS
Konstantin Wilms
Amazon Web Services
© 2015, Amazon Web Services, Inc. or its Affiliates. All rights reserved©

2. So we had a few customer requests...

3. “We’re interested in examining alternatives to traditional
linear video distribution”

4. “Any solution will need to maintain
low latency high quality video”

5. “We shouldn’t have to sacrifice
performance or quality for cost savings”

6. And some of the reasons why?

7. “We’re out of capacity. We want to add more contribution
channels without impacting existing ones, utilize higher HD
bitrates across the board, ..”

8. Which started us down an interesting path...

9. Traditional multi-hop satellite distribution
First HopSecond Hop
Field Source / EncoderHeadend / ProcessingAffiliate Spoke / Decoder

10. “Broadcast distribution scales infinitely”
…but is not a pre-requisite for this use case

11. Can we implement a solution without affecting the
underlying media layers?

12. Can we architect for scale at low cost?

13. In a non-Cloud Solution …
 Multiple datacenter footprints
 High speed, costly IP transit
 Local ops staff to manage infrastructure
 A massive cap-ex outlay
 Development staff to build this out

14. AWS takes care of this
undifferentiated heavy lifting

15. Edge Locations
Availability Zone
Region
Dallas (2)
St.Louis
Miami
JacksonvilleLos Angeles (2)
Seattle
Ashburn (3)
Newark
New York (3)
Dublin
London (2)
Amsterdam (2)
Stockholm
Frankfurt (2)Paris (2)
Singapore(2)
Hong Kong (2)
Tokyo (2)
Sao Paulo
South Bend
San Jose
Palo Alto
Hayward
Osaka
Milan
Sydney
Madrid
Seoul
Mumbai
Chennai
Global Distribution Footprint

16. c
Massively Scalable Compute
 Compute Intensive
Intel ES-2666 v3 (Haswell) optimized specifically for EC2
 Memory Intensive
Lowest price point per GiB of RAM
 GPUs
1,536 CUDA cores
4GB of video memory
 Enhanced Networking
Higher PPS, Lower network jitter, low latency
 IO Intensive
SSD Storage, EBS Optimized
 High Storage
24 x 2000 GiB per instance
AMI
EBS
Instance
Store
Amazon
EC2
Instance
Size instance by
Application need

17. Launch a CloudFormation stack
with all the infrastructure
resources for a specific
project
Autoscale the stack as
appropriate
Automated Infrastructure
AMI
CloudFormation
Deploy
Template
CloudFormation
Terminate
Template

18. AWS Ecosystem (License included in hourly* pricing)
INGEST STORE MANAGE SECUREPROCESS
CREATE
MONETIZE
INTEGRATEDELIVER

19. So let’s take a closer look at the problem...

20. What if we evolved the second hop?
First HopSecond Hop
Field Source / EncoderHeadend / ProcessingAffiliate Spoke / Decoder

21. What if we Evolved the Second Hop?
 Approach
 Up/downlink: dedicated and
internet-based IP links
 Direct Connect
 For ‘uplink’
 For stream consumption
 Concerns
 FEC
 ~500ms + RTT latency
Second Hop
Headend / ProcessingAffiliate Spoke / Decoder
Direct Connect
Secure VPN
Route53

22. Looking at Bandwidth & Transport
Satellite
 $3-5000/Mhz/mo* (~$30k/20Mbits*) +
Spoke costs
 Fixed b/w ceiling cap
AWS
 b/w to deliver an HD stream ~ $500/mo*
 Pay as you go model
FEC
 Can be implemented on UDP layer
 ARQ, SRT, LD for jitter/latency/reliability
 Sub 1Gb Direct Connect (100Mbps)
 Availability stream ingest (1:1, 1:N)
Second Hop
Headend / Processing
Affiliate Spoke / Decoder
$
$
$
! !
!

23. Multi-hop Distribution with AWS
First Hop
Field Source / EncoderHeadend / ProcessingAffiliate Spoke
Ingest
Fan Out
Egress
Scale Out
Multi-Region, Multi-AZ
Cellular
Internet
Direct Connect
Secure VPN
Internet
S3
Glacier
Route53

24. Multi-hop Distribution with AWS
Affiliate Spoke
Ingest
Fan Out
Egress
Multi-Region, Multi-AZ
Direct Connect
Secure VPN
S3
Glacier
Route53
 Additional Workflows
 Transient infrastructure
 Templatize Environments
for Quick POCs
 CloudBursting
(utilizing on-prem)
 Additional Regions

25. Some notable features …

26. 10GbpsNetwork
placement
groups
 Capacity plan for hundreds of live
HD streams and contribution silos
 Low latency high throughput
 Combine with regional replication
and Route53 for true nearest-
neighbor latency
Highly Scalable Infrastructure

27. c4
g2m3
High Capacity Egress
GPU Transcode
IngestEncoder
Broadcast
Decode
Low Bitrate
Proxy
 Fan out / fan in
 Size workflow to compute
 Flexible multi-format
HLS w/ Cloudfront CDN
MPEG-UDP w/FEC
Dedicated Pipe
Multi-Path Distribution

28. Amazon Glacier (Life Cycle Policies)
Amazon S3
 Segment media into S3
 Periodically archive to Glacier
 Time-windowed hot content with
infinite cold store
 Store/Retrieve to local edit
stations via high-speed partner
appliances
 Affiliates can make use of
storage infrastructure
(transcode)
Media Lifecycle Management

29. Let’s look at a real implementation…

30. Proof of Concept
Deployed in one afternoon into AWS VPC
Co-ordinated cross-country by a team of 3 – headend
operations, en/decoder manufacturer, and AWS
6Mbps 1080p60 MPEG-UDP w/FEC (SRT) stream
Distribution over public internet
200ms encoder to AWS, AWS to decoder latency
Lower measured latency than existing satellite 2nd hop
40 day ingress uptime with no video dropouts