dicas e truques       para ser um bom             inquilino   ... no CloudFabio Kung
inquilino?
AB            C                   D    multitenancy
IaaS        PaaS       Containers/Zones
considere ...espalhar seus  processos
weightedfair scheduling      [1]
App AP1 P2 P3              20     App BP4 P5 P6 P7              40     App CP8              40
App A           App B          App C1 2 3 0       4 5 6 7  0        8    0      20              40            40          ...
App B      App C4 5 6 7  0   8    0        40       40
App B      App C     App A4 5 6 7  0   8    0   1 2 3 30        40       40         20
App C     App A8    0   1 2 3 30    40         20
App C      App B       App A8    0   4 5 6 7 40   1 2 3 30    40           40         20
App B       App A4 5 6 7 40   1 2 3 30        40         20
App B       App A     App C4 5 6 7 40   1 2 3 30   8   80        40         20       40
App A    App C1 2 3 30   8   80      20       40minimum + bursting
co-scheduling
SMP
VM A      VM B     VM CvCPU vCPU   vCPU   vCPU vCPU
synchronization    costs
VM A      VM B     VM CvCPU vCPU   vCPU   vCPU vCPU
CPU stacking [2]
evite I/O      ... se puder
I/O intensiveapplications need their   own domain [3]
espalhe seus processos http://www.12factor.net/concurrency
efemeridade
eu acho...
PROVE!
blá blá blá ...  monitoração                  ... blá blá blá
logs => event stream
pipeline                                     Archive,                                    dashboards                       ...
Log routers • https://github.com/heroku/logplex • https://github.com/fluent/fluentd      events-to-metrics• https://github...
archives + querieshttps://twitter.com/danicuki/statuses/228170302515982336
Librato
the pager
pipeline                                     Archive,                                    dashboards                       ...
pipeline                               Archive,                              dashboards                                   ...
média
%til
naked                 domainshttps://devcenter.heroku.com/articles/avoiding-naked-domains-dns-arecords
more...http://www.12factor.net
References1. Groves, Taylor, Jeff Knockel and Eric Schulte. “BFS vs. CFS -Scheduler Comparison”. 11 Dec 2009. 2 Aug 2012 <...
Obrigado!@fabiokung
Dicas e truques para ser um bom inquilino no Cloud
Dicas e truques para ser um bom inquilino no Cloud
Dicas e truques para ser um bom inquilino no Cloud
Dicas e truques para ser um bom inquilino no Cloud
Dicas e truques para ser um bom inquilino no Cloud
Dicas e truques para ser um bom inquilino no Cloud
Dicas e truques para ser um bom inquilino no Cloud
Dicas e truques para ser um bom inquilino no Cloud
Dicas e truques para ser um bom inquilino no Cloud
Dicas e truques para ser um bom inquilino no Cloud
Upcoming SlideShare
Loading in...5
×

Dicas e truques para ser um bom inquilino no Cloud

1,714

Published on

Palestra no QConSP 2012, em 04/08/2012.

Published in: Technology, Education
1 Comment
3 Likes
Statistics
Notes
No Downloads
Views
Total Views
1,714
On Slideshare
0
From Embeds
0
Number of Embeds
2
Actions
Shares
0
Downloads
0
Comments
1
Likes
3
Embeds 0
No embeds

No notes for slide
  • \n
  • \n
  • \n
  • unidades computacionais\n
  • \n
  • \n
  • \n
  • shares, IaaS vs. PaaS\n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • fair share only what&amp;#x2019;s IN THE QUEUE\nMinimum perf. + Burstability\n
  • \n
  • SMP applications assume they are running on physical CPUs\n
  • +vCPUs = harder to schedule\n
  • \n
  • any order, locks are held\ncpu stacking\nSMP kernel has concurrency, spinlocks, etc\n
  • make a good use of your CPU time (do not spend it with context switches)\n
  • hard to share I/O resources\nhard to fair share -&gt; async\n
  • I/O is also heavily impacted by this\nisolate I/O components, attached resources\n\n
  • attachable resources\nbe prepared to scale-out\nvariability\n
  • \n
  • \n
  • data oriented discussions\n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • \n
  • internal, nagios\n
  • \n
  • \n
  • \n
  • help drive product decisions\n
  • absolute values are ok, be careful with bucket aggregators\n
  • URL or ALIAS DNS entries\n
  • \n
  • \n
  • \n
  • Transcript of "Dicas e truques para ser um bom inquilino no Cloud"

    1. 1. dicas e truques para ser um bom inquilino ... no CloudFabio Kung
    2. 2. inquilino?
    3. 3. AB C D multitenancy
    4. 4. IaaS PaaS Containers/Zones
    5. 5. considere ...espalhar seus processos
    6. 6. weightedfair scheduling [1]
    7. 7. App AP1 P2 P3 20 App BP4 P5 P6 P7 40 App CP8 40
    8. 8. App A App B App C1 2 3 0 4 5 6 7 0 8 0 20 40 40 fator de “justiça”
    9. 9. App B App C4 5 6 7 0 8 0 40 40
    10. 10. App B App C App A4 5 6 7 0 8 0 1 2 3 30 40 40 20
    11. 11. App C App A8 0 1 2 3 30 40 20
    12. 12. App C App B App A8 0 4 5 6 7 40 1 2 3 30 40 40 20
    13. 13. App B App A4 5 6 7 40 1 2 3 30 40 20
    14. 14. App B App A App C4 5 6 7 40 1 2 3 30 8 80 40 20 40
    15. 15. App A App C1 2 3 30 8 80 20 40minimum + bursting
    16. 16. co-scheduling
    17. 17. SMP
    18. 18. VM A VM B VM CvCPU vCPU vCPU vCPU vCPU
    19. 19. synchronization costs
    20. 20. VM A VM B VM CvCPU vCPU vCPU vCPU vCPU
    21. 21. CPU stacking [2]
    22. 22. evite I/O ... se puder
    23. 23. I/O intensiveapplications need their own domain [3]
    24. 24. espalhe seus processos http://www.12factor.net/concurrency
    25. 25. efemeridade
    26. 26. eu acho...
    27. 27. PROVE!
    28. 28. blá blá blá ... monitoração ... blá blá blá
    29. 29. logs => event stream
    30. 30. pipeline Archive, dashboards incApp id logs en ts t s even LogApp metrics logs router Pager eve nt s logsApp Event nts ide processor inc
    31. 31. Log routers • https://github.com/heroku/logplex • https://github.com/fluent/fluentd events-to-metrics• https://github.com/heroku/pulse
    32. 32. archives + querieshttps://twitter.com/danicuki/statuses/228170302515982336
    33. 33. Librato
    34. 34. the pager
    35. 35. pipeline Archive, dashboards incApp id logs en ts t s even LogApp metrics logs router Pager eve nt s logsApp Event nts ide processor inc
    36. 36. pipeline Archive, dashboards incApp id en ts t s even ExternalApp metrics events Monitor Pager eve nt sApp Event nts ide processor in c
    37. 37. média
    38. 38. %til
    39. 39. naked domainshttps://devcenter.heroku.com/articles/avoiding-naked-domains-dns-arecords
    40. 40. more...http://www.12factor.net
    41. 41. References1. Groves, Taylor, Jeff Knockel and Eric Schulte. “BFS vs. CFS -Scheduler Comparison”. 11 Dec 2009. 2 Aug 2012 <http://slimjim.cs.unm.edu/~eschulte/data/bfs-v-cfs_groves-knockel-schulte.pdf>2. Sukwong, Orathai and Hyong S. Kim. “Is Co-scheduling TooExpensive for SMP VMs?”. EuroSys 2011 12 Apr 2011, Carnegie MellonUniversity. 29 Jul 2012 <http://eurosys2011.cs.uni-salzburg.at/pdf/eurosys2011-sukwong-slides.pdf>3. Ongaro, Diego, Alan L. Cox and Scott Rixner. “Scheduling I/O inVirtual Machine Monitors”. Rice University. 2 Aug 2012 <http://www.cs.rice.edu/CS/Architecture/docs/ongaro-vee08.pdf>4. Pijewski, Bill. “Our ZFS I/O Throttle”. Bill Pijewskis Blog.March 1, 2011. August 3, 2012 <http://dtrace.org/blogs/wdp/2011/03/our-zfs-io-throttle/>
    42. 42. Obrigado!@fabiokung

    ×