Cloud Wars: Performance Benchmarking AWS, GCP and Azure
5 likes1,092 views
The document details ThousandEyes' public cloud performance benchmark report, highlighting the importance of data-driven decision-making in cloud performance analysis. It presents findings on inter-region and inter-availability zone measurements, revealing significant performance variations among cloud providers like AWS, Azure, and GCP. The report concludes with best practice recommendations for optimizing cloud architectures based on empirical data.
Cloud Wars: Performance Benchmarking AWS, GCP and Azure
- 2. @ThousandEyes • Public Cloud Performance Benchmark Report – Overview – Methodology • Benchmark Findings • Summary and Best Practice Recommendations • Q&A Agenda Archana Kesavan Sr. Product Marketing Manager @archana_k7
- 3. @ThousandEyes About ThousandEyes Network Intelligence solution that helps you understand performance from every user to every application over any network User App Enterprise, Endpoint and Cloud Agents Routing End-to-End Performance Data User Experience App Performance Routing Topology Network Topology Network Connectivity Device Performance 18/20 top SaaS companies 5/6 top US banks 50+ Fortune 500 companies 8/10 top global software companies
- 4. @ThousandEyes Understand Experience for Any User and App Lightweight software-based agents easily installed on your own network, in data centers, branch offices & VPCs. Enterprise Agent Endpoint Agent Browser-based plugins installed on end-user laptops and desktops. End User ExperienceInternal Vantage Points Cloud Agent Globally distributed agents installed and managed by ThousandEyes in 150+ POPs around the world. External Vantage Points
- 5. @ThousandEyes HOW DO WE MAKE DECISIONS ABOUT THE CLOUD TODAY?
- 6. @ThousandEyes Cloud Decision Factors Global PresenceGlobal Presence Pricing $ Services Pricing $ Services Performance Cloud Decisions ? Global PresenceGlobal Presence Pricing $ Services Pricing $ Services Performance Cloud Decisions
- 7. @ThousandEyes What’s Been Missing? No network architecture No loss or jitter No predictability analysis Single provider Lack comparisons Limited geos No multi-cloud Snapshots not trends DEPTH TIME BREADTH
- 8. • Industry’s first vendor- neutral, data-based research • Empowering data-driven decisions in the cloud PUBLIC CLOUDPERFORMANCEBENCHMARKREPORT 2018 Edition
- 9. @ThousandEyes Data Methodology and Scope END USER MEASUREMENTS INTER-AZ MEASUREMENTS INTER-REGION MEASUREMENTS MULTI-CLOUD MEASUREMENTS Pre-deployed Active Monitoring Platform Global Vantage Points Path Visualization Bi-directional Network Metrics THOUSANDEYES AGENT THOUSANDEYES AGENT
- 10. @ThousandEyes End User Measurements 27 Global Locations 2515 1555 Regions
- 11. @ThousandEyes Inter-AZ Measurements A B C Region X A B C Region X A B C Region X
- 12. @ThousandEyes Inter-Region Measurements 15 REGIONS25 REGIONS15 REGIONS
- 13. @ThousandEyes Multi-Cloud Measurements 15 REGIONS 25 REGIONS 15 REGIONS
- 14. @ThousandEyes Data-Driven Decision Framework 30 DAYS 10 MIN-INTERVALS 40M Unique Measurements Loss Latency 80M Unique Measurements Jitter 120M Unique Measurements Network Path 160M Unique Data Points 20M Unique Measurements 40M Unique Measurements
- 15. @thousandeyes #cloudstate18 © 2018 ThousandEyes. All Rights Reserved. GOOD BAD UGLY
- 16. @ThousandEyes How Fast Can You Go Within Your Cloud? < 2 ms E X P E C TAT I O N 0.82 ms eu-west-2 sa-east-1 us-east-1 ap-south-1 0.61 ms 1.13 ms 0.92 ms 0.72 ms 1.05 ms centralus 1.05 ms 0.79 ms southamerica-east1 europe-west-2 asia-south-1 asia-southeast1 0.60 ms 0.84 ms 0.92 ms 0.93 ms < 2 ms EXPECTATION Inter-AZ performance is reliable and consistent Indicates robust regional backbone for redundant multi-AZ architectures
- 17. @ThousandEyes ■ 10% faster than Baseline ■ Baseline ■ 10%–30% slower than Baseline ■ 30% slower than Baseline How to Choose Inter-Region Pairs? Region Pair Bidirectional Latencies (ms) from Sydney, Australia (Primary Region) AWS Azure GCP Tokyo 109.59 107.32 104.34 Singapore 174.65 108.36 168.49 Mumbai 228.48 162.15 228.92 Region Pair Bi-directional Latencies (ms) from Sydney, Australia (Primary Region) AWS Azure GCP Tokyo 109.59 107.32 104.34 Singapore 174.65 108.36 168.49 Mumbai 228.48 162.15 228.92
- 18. @ThousandEyes Leverage Data for Inter-Region Decisions Get the report for complete inter-region data www.thousandeyes.com
- 19. @ThousandEyes • Comparable performance in North America and Western Europe • No significant packet loss or outages in a 4-week period Where is End User Performance Strong? HOSTING REGION: VIRGINA, VA BI-DIRECTIONAL LATENCIES HOSTING REGION: UNITED KINGDOM BI-DIRECTIONAL LATENCIES ■AWS ■Azure ■GCP Asia Europe North America Oceania South America USER LOCATION 0 50 100 150 200 250 300 350 400 ms Asia Europe North America Oceania South America USER LOCATION 0 50 100 150 200 250 300 350 400 ms ■AWS ■Azure ■GCP
- 20. @ThousandEyes But, Asia Has A Different Story
- 21. @ThousandEyes GCP is 3X Slower 450 400 350 300 250 200 150 100 50 0 HOSTING REGION: MUMBAI, INDIA BI-DIRECTIONAL LATENCIES ■ AWS ■ Azure ■ GCP USER LOCATION GCP has 3x the network latency ms
- 22. @ThousandEyes Why is GCP 3x Slower to Asia? Users in Europe take a sub-optimal route to India 9 36 1 1 1 Traffic moves through the GCP backbone through the United States
- 23. @ThousandEyes No direct route from Europe to India Why is GCP 3x Slower to Asia?
- 24. @ThousandEyes AWS Mumbai Performance Anomaly Considerably higher latency variation with AWS 450 400 350 300 250 200 150 100 50 0 HOSTING REGION: MUMBAI, INDIA BI-DIRECTIONAL LATENCIES ■ AWS ■ Azure ■ GCPms USER LOCATION
- 25. @ThousandEyes AWS in Asia is Less Predictable Hosting Region: Tokyo Hosting Region: Singapore Hosting Region: Mumbai 50 ms 100 ms 150 ms0 BI-DIRECTIONAL LATENCY VARIATIONS ■ AWS ■ Azure ■ GCP
- 26. @ThousandEyes HOW DO YOU GET PAST PERFORMANCE VARIATIONS?
- 27. @ThousandEyes Is Multi-Cloud a Safe Alternative? Negligible packet loss and jitter AWS, Azure and GCP directly peer with each other Traffic between clouds does not leave their backbones The BIG 3 play nice! Consider Multi-Cloud for best of breed choices
- 28. @ThousandEyes WHY IS AWS SO MUCH LESS PREDICTABLE IN ASIA?
- 29. @ThousandEyes Why AWS Exhibits Less Predictability Singapore DATA CENTER Frankfurt USER Internet HongKong Frankfurt Internet Frankfurt Internet Not all clouds are created equal Architectural and connectivity differences have an impact on performance
- 30. @ThousandEyes 4 6 12 1 1 AWS Loves the Internet Traffic from users across the globe traverse the Internet longer on AWS deployments Traffic from Singapore enters the AWS backbone in Dallas, Texas
- 31. @ThousandEyes Azure and GCP Network Differently 9 12 2 6 1 2 1 1 7 Traffic from Singapore enters the MSFT backbone in Singapore
- 32. @ThousandEyes AWS Address Performance Instability With New Service • AWS Global Accelerator for improved performance • Pay AWS more $$ to ride their internal backbone instead of the Internet • Cloud Trend Alert: Monetization of the provider backbone https://blog.thousandeyes.com/aws-addresses- performance-instability-with-aws-global-accelerator/
- 33. @ThousandEyes Summary of Learnings NOT ALL CLOUDS ARE CREATED EQUAL Architectural and connectivity differences have an impact on performance GEOGRAPHICAL PERFORMANCE VARIES Do not assume uniform performance across the globe and use data to test, validate and optimize INTRA-CLOUD PERFORMANCE IS STELLAR Robust internal backbone supports redundant and distributed application architectures MULTI-CLOUD IS SAFE Symbiotic coexistence between the Big 3 and stable multi-cloud performance AWS has less performance stability in Asia GCP is 3x slower from Europe to India Inter-AZ performance between 0.5 ms – 1.5 ms AWS, Azure and GCP peer directly with each other
- 34. @ThousandEyes Recommendations qUse data, not your gut to guide your cloud investment decisions qFind your “best” and ”worst” region pairs before deciding on your cloud architecture qConsider your organization’s tolerance to Internet exposure and evaluate risk vs $$ qFactor in inter-AZ and inter-region performance. Application performance and user-experience rely on it qTrust, but verify. Avoid assumptions in the cloud
- 35. @ThousandEyes Implement a Cloud Readiness Lifecycle Baseline performance and use data to guide cloud decisions READY Monitor continuously as there is no steady state in the cloud OPERATE Monitor performance to clouds, within clouds and between clouds DEPLOY
- 36. @ThousandEyes Stay In the Loop on Public Cloud Research For more PCPBR information, please go to https://www.thousandeyes.com/research/public-cloud @archana_k7 @thousandeyes Webinars, blogs, reports Email us for a “customized” briefing
- 37. @ThousandEyes Questions? Q & A DOWNLOAD THE REPORT AT www.thousandeyes.com