1© Copyright 2018 NETINT Technologies Inc. All rights reserved.
Encoding at Scale for
Live Video Streaming
November 13, 2018
Ray Adensamer
Director of Marketing, NETINT Technologies
ray.adensamer@nct.group
Compressed video with lower bitrate reduces CDN costs
2© Copyright 2018 NETINT Technologies Inc. All rights reserved.
Video Streaming Delivery Architecture
Lower Bitrates Reduce CDN Costs
3© Copyright 2018 NETINT Technologies Inc. All rights reserved.
▪ Example: Monthly distribution cost for 1080p30 continuous live video stream
• Compares CDN costs for H.264 vs H.265 encoded video of similar quality
▪ CDN distribution costs $0.02 per GB1
Average viewership 1,000,000 100,000 10,000 1,000 1
CDN Costs for 8 Mbps (H.264 AVC, 1080p30) $ 51,840,000 $ 5,184,000 $ 518,400 $ 51,840 $ 51.84
CDN Costs for 4 Mbps (H.265 HEVC 1080p30) $ 25,920,000 $ 2,592,000 $ 259,200 $ 25,920 $ 25.92
CDN Savings $ 25,920,000 $ 2,592,000 $ 259,200 $ 25,920 $ 25.92
1: AWS Cloudfront CDN on-demand pricing (over 5PB/month) – October 2018
Drivers for Encoding and Transcoding Capacity Growth
▪ 15x Growth in Live Video1, 2016-2021
Encoding Capacity Growth will be higher:
▪ More high compression codecs
will require more encoding processing power
• H.265 HEVC = ~10x more processing vs H.264 AVC
▪ More high resolution videos
will require more encoding processing power
▪ User Generated Content (UGC) growth
• Each UGC input stream requires encoding
4© Copyright 2018 NETINT Technologies Inc. All rights reserved.
1) Cisco Visual Networking Index (VNI) 2016-2021
https://www.cisco.com/c/en/us/solutions/collateral/service-provider/visual-networking-index-vni/complete-white-paper-c11-481360.html#_Toc484813989
Drivers for Reducing Encoding Costs
5© Copyright 2018 NETINT Technologies Inc. All rights reserved.
Sports News &
Events
User Generated
Content (UGC)
Celebrities
Sources of Live Streaming Content
CDN Savings > Encoding Costs CDN Savings < Encoding Costs
Encoding Infrastructure
can profitably support
more live streams
with fewer viewers
2
Improved Margins
1
Technology Alternatives for Real-time Video Encode/Transcode
6© Copyright 2018 NETINT Technologies Inc. All rights reserved.
H.265 encoding with
Intel QSV
• Improved density
H.265 encoding on GPU
• Improved density and
capacity
• Form factors and power
consumption make GPU
hard to scale
H.265 encoding on FPGA
• Improved density and
capacity
• Better power efficiency
• Requires packaging of
FPGA somewhere inside
host server or
proprietary hardware
H.265 encoding on ASIC
(Codensity G4 SoC)
• Best density, Lower cost
• Best real-time performance
80% less power
x265 software encoding on CPU
• Lowest density and capacities
• Huge resource requirements are
economic limitation to
scalability
• Highest $ per stream
C5 (CPU) C5 (CPU) P3 or G3 (GPU) F1 (FPGA)
Common AWS EC2 Instance Types for Cloud Encoding
❑ Scalable
❑ Flexible
❑ Encoding Quality
7© Copyright 2018 NETINT Technologies Inc. All rights reserved.
Feature Benefit
Video Transcoding • Scalable H.264/H.265 <-> H.264 H.265 Encoding and Transcoding
High Capacity • 4K UHD resolution @ 60 fps*, or
• 8x 1080p streams in single U.2 module*
Low Cost • More capacity, at lower cost, than software encoding on server
• 80% less power than alternative solutions
Low Latency • Predictable low latency, even under high simultaneous loads
Introducing
Codensity™ T400 Video Transcoder
Codensity T400 Video Transcoder
Scalability through innovations in SSD cloud storage
▪ HDD -> SSD
▪ SATA/SAS -> NVMe
• NVMe is extendible for other applications
▪ U.2 is common form factor for SSDs
▪ Solution scales using NVMe storage servers
• But instead of adding U.2 SSD modules,
you add U.2 T400 transcoder modules
▪ In this example configuration,
• 10x T400 modules installed in 1U server
• 80x 1080p30 H.265 sessions* (in 1RU!)
8© Copyright 2018 NETINT Technologies Inc. All rights reserved.
NVMe = Non-Volatile Memory express = High throughput, low latency protocol developed for SSD control
Codensity T400 Video Transcoder – Encoding Quality
Codensity T400 delivers high real-time video encoding quality,
at much higher density compared to alternative solutions
9© Copyright 2018 NETINT Technologies Inc. All rights reserved.
Source: NETINT internal benchmarking and testing. August, 2018
Codensity T400 Video Transcoder – Software Integration
▪ FFmpeg integration achieved
by installing FFmpeg Codec Lib
and SDK into host server
• Seamlessly abstracts FFmpeg
video transcoding functions
from 1 or more T400
transcoder modules
▪ T400 video transcode
functions controlled through
standard NVMe protocol
10© Copyright 2018 NETINT Technologies Inc. All rights reserved.
PCIe 4.0 x4 Interface
ASIC Video Processor
H.265/HEVC
Encode/Decode
4K @ 60 fps
H.264/AVC
Encode/Decode
4K @ 60 fps
Codensity™ T400 Video Transcoder
Device Driver
Codensity™ Xcoder Codec Lib *
libavcodec
FFmpeg
U.2 Plug-In Module
installed in 2.5” NVMe Bay
NVMe
IOCTL
Xcoder lib API
Qty=1 Host Server
H.264 / H.265 Inputs
(File or Real-Time Streaming)
H.264 / H.265 Outputs
Qty=1..N
Early customer validation
“Codensity T400 promises to deliver
improved density and performance,
which will correspondingly reduce
encoding processing hardware and
costs.”
11© Copyright 2018 NETINT Technologies Inc. All rights reserved.
“Our first impressions from early
evaluations of the Codensity T400
solution in our labs indicate high
encoding quality – equivalent or
better than our encoding outputs
today.”
Dr. Lei SUN, Principal Scientist
Wangsu Science & Technology
Yongjun YU, Director
YY.Com (NASDAQ: YY)
http://about.yy.com/
▪ Live streaming video volumes will grow 15x between 2016 and 2021
▪ Encoding capacity will need to grow even faster
• Due to UGC growth, increasing video resolution, high compression codecs
▪ Today’s encoding solutions are inefficient, limiting scalability
▪ Codensity T400 Video Transcoders deliver high quality, scalability and
performance at lower cost compared to alternative solutions
• First to combine innovations from SoC and cloud storage technologies
▪ Already in early customer trials today
▪ For more information, visit www.netint.ca
Summary
12© Copyright 2018 NETINT Technologies Inc. All rights reserved.
13© Copyright 2018 NETINT Technologies Inc. All rights reserved.
Ray Adensamer
Director of Marketing, NETINT Technologies
ray.adensamer@nct.group
Questions?
Visit www.netint.ca for more information,
or download our whitepaper
“Encoding at Scale for Live Video Streaming”

Encoding at Scale for Live Video Streaming

  • 1.
    1© Copyright 2018NETINT Technologies Inc. All rights reserved. Encoding at Scale for Live Video Streaming November 13, 2018 Ray Adensamer Director of Marketing, NETINT Technologies ray.adensamer@nct.group
  • 2.
    Compressed video withlower bitrate reduces CDN costs 2© Copyright 2018 NETINT Technologies Inc. All rights reserved. Video Streaming Delivery Architecture
  • 3.
    Lower Bitrates ReduceCDN Costs 3© Copyright 2018 NETINT Technologies Inc. All rights reserved. ▪ Example: Monthly distribution cost for 1080p30 continuous live video stream • Compares CDN costs for H.264 vs H.265 encoded video of similar quality ▪ CDN distribution costs $0.02 per GB1 Average viewership 1,000,000 100,000 10,000 1,000 1 CDN Costs for 8 Mbps (H.264 AVC, 1080p30) $ 51,840,000 $ 5,184,000 $ 518,400 $ 51,840 $ 51.84 CDN Costs for 4 Mbps (H.265 HEVC 1080p30) $ 25,920,000 $ 2,592,000 $ 259,200 $ 25,920 $ 25.92 CDN Savings $ 25,920,000 $ 2,592,000 $ 259,200 $ 25,920 $ 25.92 1: AWS Cloudfront CDN on-demand pricing (over 5PB/month) – October 2018
  • 4.
    Drivers for Encodingand Transcoding Capacity Growth ▪ 15x Growth in Live Video1, 2016-2021 Encoding Capacity Growth will be higher: ▪ More high compression codecs will require more encoding processing power • H.265 HEVC = ~10x more processing vs H.264 AVC ▪ More high resolution videos will require more encoding processing power ▪ User Generated Content (UGC) growth • Each UGC input stream requires encoding 4© Copyright 2018 NETINT Technologies Inc. All rights reserved. 1) Cisco Visual Networking Index (VNI) 2016-2021 https://www.cisco.com/c/en/us/solutions/collateral/service-provider/visual-networking-index-vni/complete-white-paper-c11-481360.html#_Toc484813989
  • 5.
    Drivers for ReducingEncoding Costs 5© Copyright 2018 NETINT Technologies Inc. All rights reserved. Sports News & Events User Generated Content (UGC) Celebrities Sources of Live Streaming Content CDN Savings > Encoding Costs CDN Savings < Encoding Costs Encoding Infrastructure can profitably support more live streams with fewer viewers 2 Improved Margins 1
  • 6.
    Technology Alternatives forReal-time Video Encode/Transcode 6© Copyright 2018 NETINT Technologies Inc. All rights reserved. H.265 encoding with Intel QSV • Improved density H.265 encoding on GPU • Improved density and capacity • Form factors and power consumption make GPU hard to scale H.265 encoding on FPGA • Improved density and capacity • Better power efficiency • Requires packaging of FPGA somewhere inside host server or proprietary hardware H.265 encoding on ASIC (Codensity G4 SoC) • Best density, Lower cost • Best real-time performance 80% less power x265 software encoding on CPU • Lowest density and capacities • Huge resource requirements are economic limitation to scalability • Highest $ per stream C5 (CPU) C5 (CPU) P3 or G3 (GPU) F1 (FPGA) Common AWS EC2 Instance Types for Cloud Encoding ❑ Scalable ❑ Flexible ❑ Encoding Quality
  • 7.
    7© Copyright 2018NETINT Technologies Inc. All rights reserved. Feature Benefit Video Transcoding • Scalable H.264/H.265 <-> H.264 H.265 Encoding and Transcoding High Capacity • 4K UHD resolution @ 60 fps*, or • 8x 1080p streams in single U.2 module* Low Cost • More capacity, at lower cost, than software encoding on server • 80% less power than alternative solutions Low Latency • Predictable low latency, even under high simultaneous loads Introducing Codensity™ T400 Video Transcoder
  • 8.
    Codensity T400 VideoTranscoder Scalability through innovations in SSD cloud storage ▪ HDD -> SSD ▪ SATA/SAS -> NVMe • NVMe is extendible for other applications ▪ U.2 is common form factor for SSDs ▪ Solution scales using NVMe storage servers • But instead of adding U.2 SSD modules, you add U.2 T400 transcoder modules ▪ In this example configuration, • 10x T400 modules installed in 1U server • 80x 1080p30 H.265 sessions* (in 1RU!) 8© Copyright 2018 NETINT Technologies Inc. All rights reserved. NVMe = Non-Volatile Memory express = High throughput, low latency protocol developed for SSD control
  • 9.
    Codensity T400 VideoTranscoder – Encoding Quality Codensity T400 delivers high real-time video encoding quality, at much higher density compared to alternative solutions 9© Copyright 2018 NETINT Technologies Inc. All rights reserved. Source: NETINT internal benchmarking and testing. August, 2018
  • 10.
    Codensity T400 VideoTranscoder – Software Integration ▪ FFmpeg integration achieved by installing FFmpeg Codec Lib and SDK into host server • Seamlessly abstracts FFmpeg video transcoding functions from 1 or more T400 transcoder modules ▪ T400 video transcode functions controlled through standard NVMe protocol 10© Copyright 2018 NETINT Technologies Inc. All rights reserved. PCIe 4.0 x4 Interface ASIC Video Processor H.265/HEVC Encode/Decode 4K @ 60 fps H.264/AVC Encode/Decode 4K @ 60 fps Codensity™ T400 Video Transcoder Device Driver Codensity™ Xcoder Codec Lib * libavcodec FFmpeg U.2 Plug-In Module installed in 2.5” NVMe Bay NVMe IOCTL Xcoder lib API Qty=1 Host Server H.264 / H.265 Inputs (File or Real-Time Streaming) H.264 / H.265 Outputs Qty=1..N
  • 11.
    Early customer validation “CodensityT400 promises to deliver improved density and performance, which will correspondingly reduce encoding processing hardware and costs.” 11© Copyright 2018 NETINT Technologies Inc. All rights reserved. “Our first impressions from early evaluations of the Codensity T400 solution in our labs indicate high encoding quality – equivalent or better than our encoding outputs today.” Dr. Lei SUN, Principal Scientist Wangsu Science & Technology Yongjun YU, Director YY.Com (NASDAQ: YY) http://about.yy.com/
  • 12.
    ▪ Live streamingvideo volumes will grow 15x between 2016 and 2021 ▪ Encoding capacity will need to grow even faster • Due to UGC growth, increasing video resolution, high compression codecs ▪ Today’s encoding solutions are inefficient, limiting scalability ▪ Codensity T400 Video Transcoders deliver high quality, scalability and performance at lower cost compared to alternative solutions • First to combine innovations from SoC and cloud storage technologies ▪ Already in early customer trials today ▪ For more information, visit www.netint.ca Summary 12© Copyright 2018 NETINT Technologies Inc. All rights reserved.
  • 13.
    13© Copyright 2018NETINT Technologies Inc. All rights reserved. Ray Adensamer Director of Marketing, NETINT Technologies ray.adensamer@nct.group Questions? Visit www.netint.ca for more information, or download our whitepaper “Encoding at Scale for Live Video Streaming”