Analyzing the Barcode Scans of Millions of Users - The Backend Infrastructure at Scandit

Transcript

1. Cassandra User Group Zurich August 14, 2014 Dr. Christof Roduner COO and co-founder christof@scandit.com Analyzing the Barcode Scans of Millions of Users The Backend Infrastructure at Scandit
2. COMPANY SUMMARY
3. 3 AMBITION «Be the leading provider of software- based barcode scanning and data capture solutions for smartphones, tablets and wearables.»
4. 4 COMPANY PROFILE  Scandit AG, founded in 2009, headquartered in Zurich, Switzerland  Founded by ETH Zurich and MIT alumni (former Auto-ID Center research scientists)  Specialists in mobile image processing and cloud computing  Over 12,000 licensees worldwide  Key customers include Ahold, CapitalOne, Bayer, Nasa, Coop, Intuit, Saks 5th Avenue, Tetrapak, Shopkick and GS1  Scandit Inc. with offices in San Francisco and Boston
5. ENTERPRISE TRENDS
6. 6 CONSUMERIZATION OF THE ENTERPRISE
7. 7 BRING YOUR OWN DEVICE (BYOD)
8. 8 SCANDIT
9. PRODUCT OVERVIEW
10. 10 SCANDIT ENABLES SW-BASED AIDC ON SMARTPHONES, TABLETS, WEARABLES Great scan performance Ruggedized Clunky High TCO Outdated HW and SW Only for heavy users * 60% of current customers do not actually require ruggedness. Rugged accessories / cases can provide certified drop, dust, water and fire protection and improved ergonomics ** Not limited to ID but also supporting mobile OCR, image recognition, etc. Great scan** performance Ruggedizable* Excellent UX Low TCO Modern HW and SW Dedicated and BYOD users Cloud Services
11. 11 DEMO VIDEO www.scandit.com/video
12. 12 ENTERPRISE-GRADE BARCODE SCANNING  Proven Technology  Installed on 25+ million devices  More than 12,000 licensees  Works on any device  Scans on 3,000+ Android devices  Even those without autofocus, with low resolution cameras (240p)  Wearables like Google Glass, Vuzix, etc.  Support for Xamarin, Titanium, Phonegap  Cloud-based management platform  “Scanalytics”: scan management and analytics (top products, categories, at- home vs. in-store, etc.)  Device management capabilities  Product information managment
13. 13 SCANALYTICS EXAMPLE
14. 14 SCANALYTICS EXAMPLE
15. 15 SCANALYTICS EXAMPLE
16. CASSANDRA USE
17. 17 OVERVIEW AND DATA FLOW 4DBOEJU $MPVE .PCJMF
    %FWJDF .PCJMF
    "QQ 4DBOEJU
    4%, #BSDPEF 3FDPHOJUJPO &OHJOF %FWJDF
    "DUJWBUJPOT
    4DBOT 6TFS
    %FWJDF
    BOE -JDFOTF .BOBHFNFOU 4DBO
    1FSGPSNBODF
    "OBMZTJT 4DBOBMZUJDT &OHJOF 3FBMUJNF
    4DBOBMZUJDT %BTICPBSE 3FQPSUJOH
    &OHJOF 1SPEVDU *OGPSNBUJPO
    .BOBHFNFOU $BTTBOESB
18. 18 OUR UNIQUE COMPUTER VISION ALGORITHMS 0 50 100 150 200 250 300 0 100 200 300 400 500 600 700 brightness value position in scan line / pixels Brightness values along blurry scan line 0 50 100 150 200 250 300 0 100 200 300 400 500 600 700 brightness value position in scan line / pixels Brightness values along sharp scan line
19. 19 SCAN PERFORMANCE ANALYSIS
20. 20 SCAN PERFORMANCE ANALYSIS
21. 21 GOOGLE GLASS www.scandit.com/barcode-scanner-sdk/wearable-device-support/
22. 22 TECHNOLOGY STACK
23. 23 ARCHITECTURE OVERVIEW %FWJDF
    "DUJWBUJPOT
    4DBOT #BDLFOE
    /PEF  "1*
    4DBO
    1FSGPSNBODF
    "OBMZTJT 4DBOBMZUJDT &OHJOF $BTTBOESB -PBE
    #BMBODJOH 3BCCJU.2 #BDLFOE
    /PEF O "1*
    4DBO
    1FSGPSNBODF
    "OBMZTJT 4DBOBMZUJDT &OHJOF $BTTBOESB 3BCCJU.2 All nodes identical
24. 24 WHY WE CHOSE CASSANDRA  Scalability  High-volume storage (permanently store billions of scans)  High-volume throughput  Support large number of concurrent client requests (millions of mobile devices, each client operation creates lots of writes)  Write-heavy environment  Availability  Low maintenance - even as our customer base grows  Multiple data centers  Data model (wide rows) is a good fit
25. 25 FOUR YEARS OF CASSANDRA USE AT SCANDIT…  We’ve had Cassandra in production use for almost 4 years  Linux machines  Stable and mature – no serious incidents  Upgrades mostly painless (we’ve followed most releases from 0.6.x to 2.0.9)  But a few hiccups…
26. 26 PRODUCTION EXPERIENCE  Main challenge: Ruby community driver  Missing features: load balancing, retry policy  Transparent failover does not work  Rolling restart of cluster will kill your application  No support for new features (e.g., user-defined types)  Switch to JRuby and use Datastax driver…  Repair  Very important to run regularly  Standard «nodetool repair -pr» never worked for us  We found that repair coordinator hangs waiting for snapshots  Workaround: use «-par» option to prevent snapshots (undocumented)
27. 27 PRODUCTION EXPERIENCE  Occasional table corruption  East to fix: remove SSTable file and repair  IP address renumbering  Changing a node’s IP address is not straightforward  We decided to play it safe and used iptables redirection instead  Any experiences or ideas?
28. 28 TIPS FOR PRODUCTION USE  If you have to delete data, always delete entire rows (not columns)  Dreaded “tombstone overwhelming” exception  Aborted queries (or outright crashes pre C* 2.0.x)  Have good log monitoring in place  Cassandra is not “set up and forget”  We use Kibana  OpsCenter is your friend  Helps with debugging
29. THANK YOU! – QUESTIONS? By the way, we are hiring…