Viele Unternehmen begegnen der Cloud mit gemischten Gefühlen. Ganz im Gegensatz zu Startups.
Wir sprechen über Erwartungen, Fehler, Erfahrungen und die Zukunft aus der Sicht von SPOUD AG, ein Startup welches ohne Cloud nicht existieren würde und selbst Cloud Services anbietet.
Von altem Wein in neuen Schläuchen, Geschwindigkeitsrausch, hohen Rechnungen, bis zu deploy-your-idea-in-15-minutes oder Serverless.
10. You don’t know what data assets your company has
You already know the Gap, if…
You have to go through an IT project to connect to data
You have no idea how good or bad data quality is
You don’t know who has access to sensitive data
Sensitive data ends up in test environments
Or if you create DB dumps to allow access to your data
You have nightmares about GDPR
20. Simple IoT Analytics - What we got
Webserver
Webserver
Webserver
Loadbalancer
- Auto Scaling
- Health Checks
- Monitor and Alerting
- Managed Database (Backups, Updates, …)
- Managed Server (Updates, Cleanup, …
21. From Zero to Cloud and Back
REINVENTING US!
AGILE!
#EVENT-DRIVEN
#MORE-DATA
#ELASTICSEARCH
#AWS-SERVICES
#1-MONTH
#KAFKA
#MORE-ANALYTICS
#REALTIME
#FLINK
1 2 3 4
April 2016
65 7
Today
22. More Data, more Analytics
- Instead of vertical scaling, redesign architecture
- Distributed Log, Message-Driven
- Use of Realtime Data Processing with Flink
- Use SaaS for everything with standard interfaces (e.g. MySQL, Redis, )
ETL
24. From Zero to Cloud and Back
OVERLOAD!
SCALE!
#HIGH-VVV
#GT-10-CUSTOMERS
#GT-2000-EVENTS-PER-SEC
#NOMAD
#ELASTIC-BEATS
#HA
#TERRAFORM
January 2017 Today
1 2 3 4 65 7
25. Handling Load
- 10 customers, 500 events/s on average, 2000 events/s peak
- High latency and backpressure! Cause? CPU or I/O? Which
application?
- CPU on Flink Node because bad partitioning (and bug)
- Vertical scaling as quick fix
- Horizontal scaling needs an even load distribution
- Partitioning is key
- Later, saturated network I/O on message broker
26. Management & Automation
- Using the existing AWS services for easy management no longer
reasonable
- Welcome to the container orchestration level
- Welcome to more complexity
~ 50 - 90 Instances
31. Cost Factors
- Guarantees are not free (e.g. Message delivery, availability, order)
- It’s too easy to use Cloud products :)
- No cleanup
- Flexibility has its price
- Most expensive technologies are Kafka, Flink, ElasticSearch
- Idle time, because no reasonable automatic downscaling
32. Cost Optimisation
- Clean Cloud
- Everyone should be aware of the costs and budget
- Know your cost indicator (e.g events/s and retention time)
- Emerging patterns and products (e.g. serverless)
- Automation - Auto Scaling
Cloud
33. From Zero to Cloud and Back
RESIST COZY LOCK-INS!
#FOCUS
#LOCK-IN
#ENABLER
#STANDARDS
1
August 2017
65 7
Today
2 3 4
34. Lock-Ins
- Do not depend on specific proprietary services for our base stack
- CH and On-Premise is around the corner
- Handle specific services as optional volatilities
35. From Zero to Cloud and Back
ENTERPRISES THINK DIFFERENT!
#EVALUATION
#CH-CLOUD
#ON-PREMISE
#ORGANIZATION
1
September 2017
65 7
Today
2 3 4
36. Evaluation Cloud Provider CH
- We are spoiled by GCP, AWS, AZ
- Be aware of legacy Hosting Business Models with a Cloud paint
- Focus on scalability, cost transparency and automation
- Our customers are ready to pay more
37. On-Prem?
- Container to the rescue (LCD)
- On-Prem is always highly specific
- Be aware of the organisational structures
38. From Zero to Cloud and Back
EVERYTHING!
CROSS-CLOUD!
#CROSS-SILO
#PRIVATE-CLOUD
#ON-PREMISE
#BRIDGE
#MULTI-CLOUD
#EDGE-COMPUTING
1 65 7
Today
2 3 4
39. Private Cloud, Multi Cloud, Everything
- We have to use different services to survive. Think AI.
- Focus on Data in-motion and Event-Driven Architectures
- Data Logistics is bridging the silos
40. Bridging through Data Logistics
Logistics Layer
Infrastructure 1 Infrastructure 2
Logistics Layer
Service AService B