radial | soasta ir webinar
TRANSCRIPT
© 2016 Radial, Inc.
About the Speakers
TOM CHAVEZSr. Evangelist
SOASTA
ZAK STAMBOREditor, Online
Marketing Internet Retailer
NORM MORRISONSenior Director of Performance
Management Radial,formerly eBay Enterprise
4
The leader in omnichannel commerce technology and operationsHundreds of brands and retailers confidently partner with Radial to profitably exceed retail customer expectations and optimize how, when and where their brand promises are fulfilled.
© 2016 Radial, Inc. 8
Managing PerformanceTechnology
Synthetic Transaction Monitoring
– Single page and multi-step validation of key business transactions
– Scripted and executing continuously
– Provides actionable data for availability and performance – “why did it fail, why was it slow”
Real User Monitoring (End User):
– Measures page load time as users navigate the site
– Data collected via a JavaScript “tag”
– Data is available on browser version/type, ISP, geography, mobile vs. desktop
© 2016 Radial, Inc.
SyntheticMonitoring
Real UserMonitoring
9
Strengths WeaknessesManaging PerformanceTechnology
• Clean Room
• Continuous
• Scripted
• Actionable Detail
• Actual experience
• Coverage
• Coverage
• Sampling rate
• False alarms
• Many variables
• Quantity of data
• Browser support
9
© 2016 Radial, Inc. 10
Finding the Right Response Time Measurement MethodTwo TriesFirst attempt:– Used synthetic monitoring with lots of single page tests– Approach proved to be ineffective because of limited coverage, noise in data (example –
third parties) and lack of visibility/sensitivity to client side rendering– Clean room approach generated unrealistic results– Unable to scale up to entire platform
Second attempt: – All sites instrumented via RUM with mPulse from SOASTA– Alert thresholds set for volume and page load time thresholds– Review trends daily and weekly to spot issues creeping up gradually
© 2016 Radial, Inc. 11
Radial uses SOASTA mPulse for Real User MonitoringSOASTA mPulse Deployment:– ~40 e-commerce webstores on 3 different platforms– Standardized page group naming, conversion metrics, hosting location– 500M beacons / month– 180M beacons on Cyber Monday (peak day 2015)– Reporting and Dashboards for both internal and client stakeholders– Alerting for page load time and beacon volume– SOASTA Data Science Workbench for big data analysis
© 2016 Radial, Inc. 13
SOASTA mPulse Deployment Overview
Deploy Monitor Analyze
– Create new app in mPulse– Enable tag via Tag
Manager or in application code
– Validate data– Normalize Page Group
Names– Configure conversion
metrics
– Dashboards for stakeholders/key users
– Highlight / segment key page groups, geographies, device types and business metrics
– Daily / weekly reporting– Alerts for page load time,
beacon volume and key business metrics like conversion
– Examine 50th, 75th and 95th percentile page load times
– Conversion impact score– Investigate CS / Survey
complaints using resource timing data
© 2016 Radial, Inc. 15
Using RUM for identification, prioritization and remediationProcess Overview
1. Examine data for top pages by volume2. Rank pages by business value3. Frontend (browser) and Backend (server) performance analyzed
separately4. Break out devices types: Desktop, Mobile and Tablet5. Analyze 50th , 75th and 95th percentiles to understand performance
distribution6. Reviewed customer comments on surveys and call center activity for
possible performance investigations7. Deep dive into prioritized pages to identify root cause
© 2016 Radial, Inc. 16
Using RUM with Data Science Workbench to rank by business value
– Top 10 page groups are ranked by the relative conversion impact score
– The two highest ranked pages are chosen for optimization based on this ranking
– Search is added based on survey feedback and client feedback on poor performance
© 2016 Radial, Inc. 17
Using RUM for identification, prioritization and remediationExample: Page Load Time by Percentile
– Top 10 page groups broken out by percentile, frontend and backend separated
– Search has slow median page load time, 75th and 95th are very slow
– Search backend performance is the root cause!
© 2016 Radial, Inc. 18
Using RUM for identification, prioritization and remediationExample: Waterfall Analysis
Waterfall is timeline of resources loading from the network in the browser– HTML from application server– JavaScript, CSS and Images critical to page
display– Ancillary code and images– Third party content– Tracking pixels
© 2016 Radial, Inc. 19
Using RUM for identification, prioritization and remediationExample: Waterfall Analysis (cont.)
Waterfall is timeline of resources loading from the network in the browser– HTML from application server– JavaScript, CSS and Images critical to page
display– Ancillary code and images– Third party content– Tracking pixels
© 2016 Radial, Inc. 20
Using RUM for identification, prioritization and remediationExample: Waterfall Analysis (cont.)
Waterfall is timeline of resources loading from the network in the browser
Product Images are loading slowly for two reasons:
– Image compression problem. 300K vs. 30K
– Four images are loaded in parallel
© 2016 Radial, Inc. 22
Using RUMIdentifying issues with third party integrationsWaterfall is timeline of resources loading from the network in the browser– HTML from application server– JavaScript, CSS and Images critical to page
display– Multivariate Testing integration blocks page
load
Over 25% of total page load is waiting for a single third party integration!
© 2016 Radial, Inc. 23
Sites today are complex
– Every site has some “third party” content– Lots of tracking pixels (even pixels that load other pixels!)– Content is active (and may dynamically load additional content)– Multiple third parties provide content which comes together in the
browser– Performance in the browser is just as important as the system and
network performance
© 2016 Radial, Inc. 24
We solve site complexity with:
– Synthetic Monitoring for functional testing and availability monitoring– mPulse Real User Monitoring and Data Science Workbench from SOASTA
– monitoring and managing response time– deep diagnostic data for managing third parties– ability to analyze performance impact on key business metrics such as
conversion