Impression of using CloudSearch

2014年05月15日
AWSプロダクトシリーズ｜よくわかるAmazon CloudSearch

Transcript

1. Impression Of using Amazon CloudSearch AWSϓϩμΫτγϦʔζʛΑ͘Θ͔ΔAmazon CloudSearch 2014/5/15
2. Who am I ? • ٢ాɹঊ (Takumi Yoshida @yoshi0309) • http://blog.yoslab.com • Software Developer (Search Solution) • Apache Solr / Elasticsearch / FAST ESP / AWS ! • I love EC2 Route53 and CloudSearch
3. Question: Can I replace my search engine ? Yes, but NOT all…
4. 4 Great Functions 1. Nice UI and easy setup • you can setup domain in 10 minutes! 2. Built-in auto scale out 3. Multi domains / multi schema 4. Lucene and dismax query support • easy to move from Apache Solr.
5. How About Price ? Yes, Reasonable Price
6. *OUFSOFU *OUFSOFU 6TFS "QQMJDBUJPO *UFN
   %BUBCBTF %PNBJO
    *%$
   "84
   
   $MPVE4FBSDI
   %BJMZ
   'FFEJOH
7. My Price Simulation Tital Document Num 900,000 Document Size 3KB Number of Update / Day 300,000 Number of Query / Day 200,000 Data Out / Day 40GB Number of Domain 3 ¥200,000 / month
8. Compare with ASP Search • my price simulation : 200K yen / month • an ASP search : 400K yen / month ~
9. WARNING ! Do NOT feed a document “one by one”
10. Need Consideration Before Replacing
11. User Dictionaries • Fine precision for Japanese tokenization. • There is no user dictionary for define your terms. χίχίେඦՊͰCloudSearchͷ೔ຊޠਫ਼౓Λ୳ͬͯΈΔ http://blog.yoslab.com/entry/2014/04/16/000858
12. Keyword: ͜Ε͘͠ΐΜ Only ʮ͘ʯis highlighted.
13. Scale Out Speed
14. • You can also request increase in the maximum number of search instances or partitions for a search domain. But, this is not Warmup …
15. *OUFSOFU *OUFSOFU 6TFS "QQMJDBUJPO *UFN
    %BUBCBTF %PNBJO
     *%$
    "84
    
    $MPVE4FBSDI
    %BJMZ
    'FFEJOH VPC • CloudSearch is NOT under VPC.
16. VPC • for Feeding Speed (Near Realtime Search!) • for Security of document data • You cannot use Security Group function • Changing Access Policy is consuming time
17. But, by the way … Where is the Crawler ?
18. I’m making it, Now! • Apache ManifoldCF • http://manifoldcf.apache.org/ • you can crawl File System / Web / RSS / Windows Share / Wiki / JDBC / FileNet / Documentum / Dropbox / SharePoint and so on. • Output connector for Amazon CloudSearch may be included in next release - August planed (v1.7)
19. Conclusion
20. Conclusion • Easy setup, Great functions • Reasonable price, but be careful with feeding documents and do indexing • Fine precision for Japanese except new coinage and terminology of your business • Spike access
21. Q & A
22. Thank you for attention !