Upgrade to Pro — share decks privately, control downloads, hide ads and more …

D2-3 Tharmes Siva - Continuous Performance Delivery: Find and Fix the Things That Matter

D2-3 Tharmes Siva - Continuous Performance Delivery: Find and Fix the Things That Matter

Businesses want to ensure that their end users are happy so they can maximize productivity, loyalty, and profits. But poor performing applications cause serious business issues. In today’s DevOps organizations and agile environments, it is essential to ensure application performance management throughout the full application lifecycle from the End User to Infrastructure. To achieve this, modern application performance management faces a challenge:

on one hand, it has to be extremely flexible and agile
on the other hand has to provide all necessary information on all transactions end-to-end
in production as well as development and test environments so that people are talking the same language.

Because DevOps is about speed and iterations, we will share in this presentation a performance management philosophy which will maximize the value of DevOps for high performing applications. It allows you to find and fix the things that really matter instead of hunting ghost performance problems with your DevOps groups.

DevOpsDays Zurich

May 08, 2017
Tweet

More Decks by DevOpsDays Zurich

Other Decks in Technology

Transcript

  1. © 2015 Riverbed Technology. All rights reserved. 2 Application Performance

    Management Solutions for DevOps The Context Validate that code performs well; leverage diagnostics to fix bugs without reproducing them Test app for problems and share diagnostics when anomalies are detected Deploy and monitor app in production; share diagnostics with Dev and QA teams when issues are detected Measure user experience, usage trends and measure business impact of the app
  2. © 2015 Riverbed Technology. All rights reserved. 3 Every millisecond

    is critical Business Impact of Poor Digital Experience 1 sec of eCommerce site page slowdown can cost up to $1.6B in annual sales1 A typical broker loses $4M in revenues per millisec when their trading platform is 5 millisecs behind their competitor2 1https://www.fastcompany.com/1825005/how-one-second-could-cost-amazon-16-billion-sales 2http://www.forbes.com/sites/advisor/2014/04/16/the-brokerage-world-is-changing-who-will-survive/#4207c2526eb9
  3. © 2016 Riverbed Technology. All rights reserved. 4 © 2016

    Riverbed Technology. All rights reserved. 4 APM is NOT about proving your innocence
  4. © 2016 Riverbed Technology. All rights reserved. 5 End User

    Network Web Apps Services Databases Point Solutions Point Solutions Point Solutions Point Solutions Point Solutions Point Solutions Silo’d Performance Management
  5. © 2016 Riverbed Technology. All rights reserved. 7 To fix

    the things that matter you have to find the things that matter and not be led astray find
  6. © 2016 Riverbed Technology. All rights reserved. 8 Improve Collaboration

    and Reduce Silos Across Groups Provide Stakeholders with Meaningful Data New Performance Management Approach Needed for DevOps Provide Granular data through all stages of lifecycle Define Develop Testing Production
  7. © 2016 Riverbed Technology. All rights reserved. 10 Server Delays

    Application Code Instrumentation Network Delays Network Traffic Monitoring End User Experience Real User Monitoring Business Transaction Performance SLA Violations Server / OS Resource Monitoring
  8. © 2016 Riverbed Technology. All rights reserved. 12 © 2016

    Riverbed Technology. All rights reserved.
  9. © 2016 Riverbed Technology. All rights reserved. 13 © 2016

    Riverbed Technology. All rights reserved.
  10. © 2016 Riverbed Technology. All rights reserved. 14 © 2016

    Riverbed Technology. All rights reserved.
  11. © 2016 Riverbed Technology. All rights reserved. 15 Capture the

    right things from the right places in the right way
  12. © 2016 Riverbed Technology. All rights reserved. 16 © 2016

    Riverbed Technology. All rights reserved. 16 H/V/C 2 JVM 2 Host/VM/Container 1 JVM 1 CLR 1 Client Web Service SQL Remote Method Servlet Code Instrumentation TTFB Download Resources Redirect DNS End User Experience SELECT this FROM that WHERE xy=8 AND z>12; D https://wsvc2.comp.com/lookupsvc/GetThat?XY=8&Z=12 C A B https://wsvc1.comp.com/bar.asmx/DoThis?XY=8 https://app.comp.com/foo.jsp?do=this&X=2&Y=4 E F G G G G G https://app.comp.com/foo.jsp?do=this&X=2&Y=4
  13. © 2016 Riverbed Technology. All rights reserved. 17 © 2016

    Riverbed Technology. All rights reserved. 17 URL # TXs Resp Time /page/c.jsp 134 8.63 /dir/a.aspx 792 6.27 /d.do 1,219 4.92 /pages/ b.jsp 876 3.12 /dir/e.aspx 12,41 8 1.07 SQL # Calls Resp Time SELECT x FROM y WHERE … 2,641 2.15 INSERT INTO z VALUES … 376 1.97 SELECT a FROM b WHERE … 11,84 7 0.78 UPDATE c SET … 192 0.63 TOP URLS TOP SQL
  14. © 2016 Riverbed Technology. All rights reserved. 18 © 2016

    Riverbed Technology. All rights reserved. 18 URL # TXs Resp Time /page/c.jsp 134 8.63 /dir/a.aspx 792 6.27 /d.do 1,219 4.92 /pages/ b.jsp 876 3.12 /dir/e.aspx 12,41 8 1.07 SQL # Calls Resp Time SELECT x FROM y WHERE … 2,641 2.15 INSERT INTO z VALUES … 376 1.97 SELECT a FROM b WHERE … 11,84 7 0.78 UPDATE c SET … 192 0.63 TOP URLS TOP SQL URL=/page/*.jsp AND SQL=SELECT x FROM y *
  15. © 2016 Riverbed Technology. All rights reserved. 19 © 2016

    Riverbed Technology. All rights reserved. 19 MICRODATA TOP URLS TOP SQL URL # TXs Resp Time /page/c.jsp 134 8.63 /dir/a.aspx 792 6.27 /d.do 1,219 4.92 /pages/ b.jsp 876 3.12 /dir/e.aspx 12,41 8 1.07 SQL # Calls Resp Time SELECT x FROM y WHERE … 2,641 2.15 INSERT INTO z VALUES … 376 1.97 SELECT a FROM b WHERE … 11,84 7 0.78 UPDATE c SET … 192 0.63 URL=/page/*.jsp AND SQL=SELECT x FROM y * MACRO DATA MACRO DATA
  16. © 2016 Riverbed Technology. All rights reserved. 20 © 2016

    Riverbed Technology. All rights reserved. 20 URL=/page/*.jsp AND SQL=SELECT x FROM y * MICRODATA URL # TXs Resp Time /page/c.jsp 134 8.63 /dir/a.aspx 792 6.27 /d.do 1,219 4.92 /pages/ b.jsp 876 3.12 /dir/e.aspx 12,41 8 1.07 TOP URLS MACRO DATA SQL # Calls Resp Time SELECT x FROM y WHERE … 2,641 2.15 INSERT INTO z VALUES … 376 1.97 SELECT a FROM b WHERE … 11,84 7 0.78 UPDATE c SET … 192 0.63 TOP SQL MACRO DATA COMPLEX ANALYSIS
  17. © 2016 Riverbed Technology. All rights reserved. 23 © 2016

    Riverbed Technology. All rights reserved. 23 TRANSACTION EVERY MATTERS . © 2016 Riverbed Technology. All rights reserved.
  18. © 2016 Riverbed Technology. All rights reserved. 31 Traditional Physical

    Application Map Performance Graph – Logical Relationship/Impact Mapping
  19. © 2016 Riverbed Technology. All rights reserved. 33 Multinational financial

    services company 30+ Tier 1 Applications 1,000s of different types of transactions Account details page frequently extremely slow The problem persisted optimizing the slowest methods and For months they tried analyzing the slowest transactions optimizing the slowest methods analyzing the slowest transactions
  20. © 2016 Riverbed Technology. All rights reserved. 34 optimizing the

    slowest methods analyzing the slowest transactions optimizing the overall slowest methods analyzing all transactions
  21. © 2016 Riverbed Technology. All rights reserved. 35 95% Faster

    7 Million Transactions / Day 2,000 Hours Saved / Day 53ms Method Optimized 1.1sec Response Time Before Optimization 62ms Response Time After Optimization Real-World Benefits
  22. © 2016 Riverbed Technology. All rights reserved. 36 We just

    didn’t realize how bad it was. We knew this could be improved. How long has it been like this? Forever . What else uses this code? EVERYTHING .
  23. © 2016 Riverbed Technology. All rights reserved. 37 Everything. 30+

    Tier 1 Applications 1,000s of different types of transactions All shared that common monitoring method All saw significant improvements in performance Tens of millions of transactions per day
  24. © 2016 Riverbed Technology. All rights reserved. 38 © 2016

    Riverbed Technology. All rights reserved. 38 Big Data APM for DevOps Data Quality Matters: Macrodata vs. Microdata, 1-sec granularity A Big Data approach is needed to solve overarching problems Fix the things that matter : $$$ - not just seconds