How Condé Nast went form $$$ to $$$/4

1. Serverless Hamburg – 12 March 2018
Serverless Cost Optimization:
how Condé Nast went from $$$ to $$$/4
Marco Viganò
Digital CTO
Serverless On Stage #10
20 March 2018

2. Serverless Hamburg – 12 March 2018

3. Serverless Hamburg – 12 March 2018

4. Serverless Hamburg – 12 March 2018
CN.numbers // by month
30M Unique Visitors
250M Page Views
20% Desktop 80% Mobile
46% SEO 29% Social

5. Serverless Hamburg – 12 March 2018
CN.technologies

6. Serverless Hamburg – 12 March 2018
• Infrastructure scaling problems due to traffic boost
• Non optimal delivery and uptime
• Aggressive time to market
• No automation
• Costs
onPremise(CN) = Error 500 Internal Server Error
2013 / 2014

7. Serverless Hamburg – 12 March 2018
Wave 1: the pilot

8. Serverless Hamburg – 12 March 2018
CN.pilot === “Wired.it”
MigrationPreparationEvaluation Tuning
Pilot Cloud migrationEngagement of team

9. Serverless Hamburg – 12 March 2018
• Infrastructure migrated AS IS -> no optimization for the cloud
• 150 Server + 30 DB + more than 50 LB
• Application redundancy
• Costs explosion:
CN #epicfail
on premise +
cloud +
people +
external providers =
_______________
a lot of money!!!

10. Serverless Hamburg – 12 March 2018
Wave 2: consolidation
(start Q3 2014)

11. Serverless Hamburg – 12 March 2018
• Application !cloud optimized
• Destroy monoliths
• Make refactoring
• Automation
• CN Blueprint

12. Serverless Hamburg – 12 March 2018
CN.blueprint

13. Serverless Hamburg – 12 March 2018
end of wave 2 (2015): ROI

14. Serverless Hamburg – 12 March 2018
Wave 3: thinking Serverless
(start Q3 2014)

15. Serverless Hamburg – 12 March 2018
CN.Vogue().photovogue
• > 300,000 photographers
more than 800,000 photos
image size up to 50 Mb
The Challenge
• PV was launched in 2011: needs new
UI/UX and to be re-engineered
• Photos and users growing by the day: old
legacy IT infrastructure wasn’t able to
manage the website traffic
• We need to provision resources quickly:
problems in scaling
• We wanted to give both photographers and
editorial staff a better, faster experience
• Problems with large file upload

16. Serverless Hamburg – 12 March 2018
serverless(CN.Vogue().photovogue)
• Quicker provisioning of resources: from days to hours
• No scaling problem due to traffic boost
• Cost saving: cut 30% in comparison to the old infrastructure
• Enabling innovation: Devs / DevOps, are now focused on innovation not on manage old infrastructure
survival
• UX 90% faster: photographer and editorial team now have an excellent experience
old_windows_cluster(CN.Vogue().photovogue)

17. Serverless Hamburg – 12 March 2018
end of wave 3 (2016)

18. Serverless Hamburg – 12 March 2018
Wave 4: reserve capacity
(start in Q2 2016 – running in 2017/2018)

19. Serverless Hamburg – 12 March 2018
Predictable Workloads

20. Serverless Hamburg – 12 March 2018
• Reserved Instances / Committed use discounts
• 1 year / 3 years
• CN.Italy.saving[‘2016’] = 35%
• CN.Italy.saving[‘2017’] = 60%
• CN.Italy.saving[‘2018’] = VMs + DB + DWHPay as yo go Reserved Capacity

21. Serverless Hamburg – 12 March 2018
2017 costs = On Premise / 4

22. Serverless Hamburg – 12 March 2018
Wave 5: container
(start Q4 2017 - runninng)

23. Serverless Hamburg – 12 March 2018
• Build, Ship, and Run any App, Anywhere • Running containers across many different machines
• Scaling up or down by adding or removing containers
when demand changes
• Keeping storage consistent with multiple instances of
an application
• Distributing load between the containers
• Launching new containers on different machines if
something fails

24. Serverless Hamburg – 12 March 2018

25. Serverless Hamburg – 12 March 2018
Tips & Tricks

26. Serverless Hamburg – 12 March 2018
• VMs Autoscaling
• Caching: Varnish, Redis, memcache…
• Offload of static resources: CDN
• Infrastucture self healing
• Serverless
Tip&Tricks(CN); // costs optimization

27. Serverless Hamburg – 12 March 2018
Turn off the lights

28. Serverless Hamburg – 12 March 2018

29. Serverless Hamburg – 12 March 2018
Turn off the lights
25% 25% 25% 25%
• CPU from 8pm to 8am
• 0.2$/h
0.2$/h x 4VMs x 24h x 365day = 7008 $
• Turn of from 8pm to 8am
12h x 365day = 4380h saving = 876$
• 7008$ - 876$ = 6132$
• 12.5% Saving
33% 33% 33%

30. Serverless Hamburg – 12 March 2018

31. Serverless Hamburg – 12 March 2018
Last but not least
• Make investments on your team: training, summit, Meetup, certifications, R&D…
• Your team must be at the center of your Cloud Transformation
• No Team -> No Party -> No Saving!!!

32. Serverless Hamburg – 12 March 2018
summary

33. Serverless Hamburg – 12 March 2018
CN.costs.onCloud() = CN.costs.onPremise() / 4
CN.time_to_market.onCloud() = CN.time_to_market.onPremise() / 5

34. Serverless Hamburg – 12 March 2018
2013/2014
>150 servers!
30 Databases
2015: ROI!!!!
2016
Change Mindset:
Thinking Serverless
- Photovogue
- Starting reducing costs
From an angry CFO… to a happy CFO :)
2017
Infrastucture improvements
50 servers - 8 Databases
Costs = on premise / 4
On premise
2018
Continuos improvements:
Serverless *.*
Docker / K8

35. Serverless Hamburg – 12 March 2018
Thank You
Marco Viganò
@Sasha0423