Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.
Scaling Hotstar.com for
10Mn concurrency
CHALLENGES & LEARNINGS
Gaurav Kamboj
Cloud Architect at Hotstar
Let’s talk about Scale!
10Mn+
PEAK CONCURRENCY DURING
VIVO IPL FINAL
5.7 Tbps
PEAK BANDWIDTH CONSUMPTION
1Mn
PEAK REQUESTS...
Game Day
FACE THE REAL GAME BEFORE THE ACTUAL GAME
36,000
CPUs
80 Gbps
NETWORK OUT
72 TB
RAM
8 REGIONS
GEO DISTRIBUTED
Pro...
Race against time!
+ 500K
GROWTH RATE PER MINUTE
(CONCURRENCY)
FULLY BAKED AMIs
4 mins
APPLICATION BOOT-UP TIME
90 SECONDS...
Why we don’t use autoscaling
INSUFFICIENT
CAPACITY
ERRORS
GAME OF (AVAILABILITY) ZONES
STEP SIZE
AUTOSCALING GROUPS
Others challenges..
SINGLE
INSTANCE TYPE
PER AUTO SCALING GROUP
API
THROTTLING
RETRIES &
EXPONENTIAL
BACKOFF
PREWARM INFRA
BEFORE MATCH TIME
PROACTIVE SCALE UP
WITH BUFFER FOR SPIKES
SECONDARY
AUTO SCALING GROUP
SPOT FLEET
DIVERSIF...
MACHINE LEARNING
BASED AUTO SCALING
Future.. IPL 2019 onwards
INTELLIGENT
SCALE UP DECISIONS
RUN IPL & WORLD CUP 2019 ON
A...
tech.hotstar.com
Scaling hotstar.com for 10 million concurrent viewers
Upcoming SlideShare
Loading in …5
×

Scaling hotstar.com for 10 million concurrent viewers

246 views

Published on

Talk by Gaurav Kamboj, Cloud Architect at Hotstar on the topic "Scaling hotstar.com for 10 million concurrent viewers - challenges & learnings" at AWS Community Day Bengaluru 2018

Published in: Technology
  • Be the first to comment

  • Be the first to like this

Scaling hotstar.com for 10 million concurrent viewers

  1. 1. Scaling Hotstar.com for 10Mn concurrency CHALLENGES & LEARNINGS Gaurav Kamboj Cloud Architect at Hotstar
  2. 2. Let’s talk about Scale! 10Mn+ PEAK CONCURRENCY DURING VIVO IPL FINAL 5.7 Tbps PEAK BANDWIDTH CONSUMPTION 1Mn PEAK REQUESTS PER SECOND 10Bn+ CLICKSTREAM MESSAGES 100+ HOURS OF LIVE TRANSCODING EVERYDAY
  3. 3. Game Day FACE THE REAL GAME BEFORE THE ACTUAL GAME 36,000 CPUs 80 Gbps NETWORK OUT 72 TB RAM 8 REGIONS GEO DISTRIBUTED Project HULK ● Load Generation ● Performance & Tsunami Tests ● Chaos Engineering ● Traffic Pattern using ML
  4. 4. Race against time! + 500K GROWTH RATE PER MINUTE (CONCURRENCY) FULLY BAKED AMIs 4 mins APPLICATION BOOT-UP TIME 90 SECONDS REACTION TIME PUSH NOTIFICATIONS
  5. 5. Why we don’t use autoscaling INSUFFICIENT CAPACITY ERRORS GAME OF (AVAILABILITY) ZONES STEP SIZE AUTOSCALING GROUPS
  6. 6. Others challenges.. SINGLE INSTANCE TYPE PER AUTO SCALING GROUP API THROTTLING RETRIES & EXPONENTIAL BACKOFF
  7. 7. PREWARM INFRA BEFORE MATCH TIME PROACTIVE SCALE UP WITH BUFFER FOR SPIKES SECONDARY AUTO SCALING GROUP SPOT FLEET DIVERSIFIED INSTANCE TYPES Battle tested scaling strategy!
  8. 8. MACHINE LEARNING BASED AUTO SCALING Future.. IPL 2019 onwards INTELLIGENT SCALE UP DECISIONS RUN IPL & WORLD CUP 2019 ON AUTO PILOT MODE
  9. 9. tech.hotstar.com

×