Modular Monolith - a Practical Alternative to Microservices @ Devoxx UK 2024
Geocloud blue raster web mapping cloud deployment lessons from the field 2010-09-21
1. Web Mapping
Cloud Deployment-
Lessons from the Field
Presented by:
Michael Lippmann
Blue Raster
Applications Moving
to the Cloud
Location Intelligence
GeoCloud 2010
September 22, 2010
2. Challenge- Develop/Deploy in 30 days
Client
Health and Human Services - HRSA
Need
Health Care Reform 2010
Presidential initiative
Double the number of Community Health Centers
nationally from 1200 to 2400
90 day window to submit grants for a pool of >$1B
3. Web Mapping Requirement
Deploy a web mapping application to assist with
grant process
Non-negotiable - System must be online when grant
period opens
Week 1- significant utilization
Middle Weeks- use, but not at peak
Approaching deadline- last minute use spike
4. Requirements
High availability application
Downtime not an option-Redundancy
High Performance
Subsecond performance for maps
Capacity
Grant applicants will depend on system
$1B+ being applied for
5. Why we went “Cloud”- Perishable App
• Ability to scale to immediately- Day 1
• Ability to deploy 5 load balanced map servers for
peak demand
• Managed services for scaling based on use
• Lower demand = turn servers off
• ESRI Cloud allowed peak use for short period of
time
• On Day 90- system may be turned off.
6. Old Way, Cloud Way
Old Cloudy – Web Mapping
• 5x Servers • 10x Servers
• order, setup, deploy, • Deploy – right click, wait 5
maintain minutes
• 168 Hours x $6/hr = ~$1000
• 5x Licenses
• Peak -$1000/week
• $300k capital
• Non-peak – $250/week
• After 90 days not needed! • $30k for grant period
• Bottom line- allows for scale
up/down with much lower
capital outlay
7. How we did it- Hybrid Solution
ESRI
ArcGIS Server -Cloud Infrastructure
Amazon
EC2
ESRI ArcGIS Server for web mapping
S3/CloudFront
Map tile cache hosting
Hosting.com
vCloud hosting (using VMware)
Adobe ColdFusion
Microsoft SQL Server
13. What we Learned-
Moving tiles into the cloud-
• Trying to upload 45 Million map tiles = #FAIL
• Simply copying that many files takes >1 day
• Load with Amazon Import/Export
Use fastest HD you can find- WD 10k SATA
Took 20 hours to load files
Within 48 hours files are online and ready to use
• Manage with CloudBerry
Change permissions/setup CloudFront
14. Hybrid Choices
Hosting.com Managed Services- Windows 2008 R2
Servers online is not as immediate as Amazon- but
you have a person to call
Machines setup and has 24/7/365 team to monitor
and assist as needed
Both flavors of Cloud are important in the future.
15. Launch Day- Plan for Anything
• You will have DOWNTIME, need to manage it
• Launch Day- 2 events caused unplanned
downtime
Amazon S3 Outage- extremely rare but it happens
Datacenter – Cisco switch failure
16. Health Check your Site
Setup tools to watch all of your servers
Recommend Pingdom- found it when Twitter had an
outage
SAAS Monitoring service
Uses global array of servers to test and notify of
issues
20. If we did it again… we might:
Use even more Amazon S3/CloudFront to cache
repeatable operations
Geometry request for boundaries-
states/counties/HPSA
Gzip/cache speeds load
Cache directly in Amazon- avoid transfer
CloudFront- now has immediate expiration
21. Risks of Cloud
• Amazon is great…but many cannot handle
unknown risk of elastic pricing
• Need team that can administer
• Technical saavy – new paradigm
• Changes rapidly – must keep up
22. Conclusions/Questions
• Deployment was successful, next one will be
easier and better
• Cloud for web mapping cannot be ignored
• Hybrid solutions work very well
• Focus on your business requirements, outsource
all of the rest
• Simple is always best
23. For more information:
Michael Lippmann
mlippmann@blueraster.com
703-875-0914
www.blueraster.com
blog.blueraster.com
@blueraster