Upload
rightscale
View
229
Download
2
Embed Size (px)
Citation preview
april25-26 sanfrancisco
cloud success starts here
Performance: Key Elements to Consider in the CloudCraig Wesely Irwin, VP Channel &
Alliances
# 2# 2
#RightscaleCompute
The Cloud - Fact or Fiction?• The Cloud = “It Just Works”
• All applications behave and scale the same way
• Theoretical analysis works when capacity planning for the cloud
• Texas Style - Bigger is better
• If my web or mobile app crashes, then that’s a good thing.
# 3# 3
#RightscaleCompute
Today’s Situation• Web & Mobile Apps Continue to Crash• Slow Response Times• Why?
# 4# 4
#RightscaleCompute
So you’re now in the cloud, how will
you prepare for the unexpected?...
# 5# 5
#RightscaleCompute
So you’re now in the cloud, so how
do you respond to the expected?
# 6# 6
#RightscaleCompute
The Peak Load Challenge
Load
Rapid Load Ramp-Up
Snowball Effect
Load Fixed Capacity Scalability Application Transparency
Server & Cloud
# 7# 7
#RightscaleCompute
Tips & Suggestions #1 For peak and high load
Have backup “minimalistic start/landing pages“ Small is Fast
#2 Extensive use of Front End Cache systems Optimize the cache solution, consider Varnish Less traffic is less problem, no direct DB access
#3 Implement Scaling & Queuing System Redirect excess traffic using Load Balancer Create Informative ”Wait” pages
Bottom Line: Like a good boy scout Always be prepared : Test the solution before launch
# 8# 8
#RightscaleCompute
Why Run A Performance Load Test ?
• Is the site stable?• When does it crash?• How can I make it faster?• Can my application scale?
# 9# 9
#RightscaleCompute
The Cloud - Fact or Fiction?• The Cloud = “It Just Works”
• All applications behave and scale the same way
• Theoretical analysis works when capacity planning for the cloud
• Texas Style - Bigger is better
• If my web or mobile app crashes, then that’s a good thing.
# 10# 10
#RightscaleCompute
Load
vs.
Capacity
# 11# 11
#RightscaleCompute
vs.
TestEnvironment
ProductionEnvironment
# 12# 12
#RightscaleCompute
Load Test 1 to 1
Throughput
Nr ofusers
Point of collapse
Completefailure
Response time∞
Load Maximum– How many users can we handle?– What is a good result?
Behavior in the ”Danger Zone”– Does the application become
unstable above load maximum?
Problem Analysis– Where are the bottlenecks?– How to fix them?
Nr ofusers
# 13# 13
#RightscaleCompute
Do you have Performance Targets?• Examples
• Stable beyond 100,000 Concurrent users• Able handle 100,000 page views per hour• < 4 second average response time• Top 5 in our industry• Our peak time response time shall be better then
competition’s• 95% of our users shall make a travel selection for
purchase in less then 30 seconds
# 14# 14
#RightscaleCompute
Load Test Findings• #1 Identify the Backend Calls
• Database calls don’t kill your application • Lack of caching does!
• #2 Check the delivery of static Content • Optimize the cache solution, consider Varnish• Consider using a CDN, if needed
• #3 Web Infrastructure• Load Balancer• Server model • Bandwidth• Scaling & Failover
# 15# 15
#RightscaleCompute
Know Your Numbers - Not all Web & Mobile Apps are equal
# 16# 16
#RightscaleCompute
Numbers Don’t Lie – Eliminate Subjectivity• Establish a baseline • My start page | Login | Book a flight• Response Time – Average 3.2 sec• Typical Values – Median 2.5 sec• Standard Deviation 2.8 sec• SLA 99.9%• 95% is better than 11.8 sec
# 17# 17
#RightscaleCompute
Web Performance Monitoring for the Cloud
Browser Scenarios Analytics Response time
Browser
Analytics Trend Complex
Application SLA on
applications
Application
Basic Monitor Alerting Up – down Basic SLA
UP/Down
Inside monitoring
Vital Signs Drill Down Root cause Consolidation
with other system
Correlation1 2 3 4
# 18# 18
#RightscaleCompute
End User Performance Monitoring – What’s the value?
# 19# 19
#RightscaleCompute
Tips & Suggestions• #1 Set Goals
• Uptime • Performance
• #2 Hate the average• Work with the exceptions• Remove the 10 Worst transactions
every month
• #3 Fire drill • Help identifying problems• Correlation of data
# 20# 20
#RightscaleCompute
Cliff Notes• Performance Testing will help you
discover the weakest link in advance…• Saving you time and $$$
• Check your numbers...Know your application• They don’t lie
• Plan for the unexpected...• It happens everyday
april25-26 sanfrancisco
cloud success starts here
Questions?