Capacidade Planejada

2,970
-1

Published on

Published in: Technology
0 Comments
2 Likes
Statistics
Notes
  • Be the first to comment

No Downloads
Views
Total Views
2,970
On Slideshare
0
From Embeds
0
Number of Embeds
5
Actions
Shares
0
Downloads
27
Comments
0
Likes
2
Embeds 0
No embeds

No notes for slide

Capacidade Planejada

  1. 1. Capacidade Planejada Rodrigo Albani de Campos - camposr@gmail.com - @xinu Wednesday, August 25, 2010
  2. 2. Agenda • Motivações • Capacidade + Velocidade • Universal Scalability Model • Extra: Dimensionamento de pools Wednesday, August 25, 2010
  3. 3. Impacto na imagem do produto Wednesday, August 25, 2010
  4. 4. Impacto na imagem do produto Wednesday, August 25, 2010
  5. 5. SLA Horas / Trimestre Impacto $ 98% 43,2 $142.600.000 99% 21,6 $71.300.000 99,90% 2,16 $7.130.000 99,99% 0,216 $713.000 99,9990% 0,0216 $71.300 99,9999% 0,00216 $7.130 Amazon 2010Q1 Net Sales $7.13 Billion Dollars Wednesday, August 25, 2010
  6. 6. Distinct Query Revenue/ Any Clicks Satisfaction Time to Click Queries/User Refinement User (increase in ms) 50ms 0 0 0 0 0 0 200ms 0 0 0 -0,30% -0,40% 500 500ms 0 -0,60% -1,20% -1,00% -0,90% 1200 1000ms -0,70% -0,90% -2,80% -1,90% -1,60% 1900 2000ms -1,80% -2,10% -4,30% -4,40% -3,80% 3100 $85.000.000 em três meses na Amazon Receita Impacto de lentidão na entrega de conteúdo The User and Business Impact of Server Delays, Additional Bytes, and HTTP Chunking in Web Search - Eric Schurman (Amazon), Jake Brutlag (Google) Wednesday, August 25, 2010
  7. 7. Receita Impacto de lentidão na entrega de conteúdo The User and Business Impact of Server Delays, Additional Bytes, and HTTP Chunking in Web Search - Eric Schurman (Amazon), Jake Brutlag (Google) Wednesday, August 25, 2010
  8. 8. “Fast isn’t a feature, fast is a Requirement” Jesse Robins - OPSCode Wednesday, August 25, 2010
  9. 9. Average # of Average Average # of resources / document size hosts / page page Kb 42,14 8,39 477,26 Composição de páginas na Web - Top Sites Web Metrics: Size and number of resources - Sreeram Ramachandran http://code.google.com/speed/articles/web-metrics.html Wednesday, August 25, 2010
  10. 10. Load Time First Byte Start Complete Requests Bytes In Render Time (s) Mercado Livre 1,972 329 ms 543 ms 2,491 22 74 KB Ebay 1,999 399 ms 1493 ms 3,103 36 237 KB Amazon 4,777 504 ms 1105 ms 6,289 69 454 KB newegg 6,848 328 ms 1211 ms 7,891 138 459 KB Bestbuy 7,508 447 ms 1733 ms 10,41 99 676 KB Submarino 10,436 250 ms 2474 ms 10,436 151 1,125 KB Casas Bahia 15,09 500 ms 4401 ms 15,799 100 732 KB http://www.webpagetest.org/ TEST RESULTS July 2010 Wednesday, August 25, 2010
  11. 11. Capacity and Velocity Wednesday, August 25, 2010
  12. 12. Service time: Tempo de ocupação do recurso (s,ms,μs) Arrival rate: Taxa de chegada de requisições para o recurso (hit/s,qps,etc...) Little’s Law: The long term average number of customers in a stable system L is equal to the long term average arrival rate λ, multiplied by the long term average time a customer spends in the system,W Queuing Theory 101 Wednesday, August 25, 2010
  13. 13. Service  A 0,1 1600 0,09 1400 0,08 1200 0,07 1000 0,06 Service  Time  (s) Frequency 0,05 800 0,04 600 0,03 400 0,02 200 0,01 0 0 0 50 100 150 200 250 300 Hits/s Service  Time Frequency Average Service Time against Arrival Rates Wednesday, August 25, 2010
  14. 14. Service  B 2,5 6000 5000 2 4000 1,5 Service  Time  (s) Frequency 3000 1 2000 0,5 1000 0 0 0 20 40 60 80 100 120 140 160 180 200 Hits/s Service  Time Frequency Average Service Time against Arrival Rates Wednesday, August 25, 2010
  15. 15. Service  B 2,5 6000 5000 2 4000 Frustrated 1,5 Service  Time  (s) Frequency 3000 1 2000 Tolerating 0,5 1000 0 0 0 20 40 60 80 100 120 140 160 180 200 Hits/s Satisfied Service  Time Frequency APDEX - http://www.apdex.org Wednesday, August 25, 2010
  16. 16. Satisfied Tolerating Frustrated APDEX - http://www.apdex.org Wednesday, August 25, 2010
  17. 17. Samples 12000 Satisfied 10000 Tolerating 800 Apdex 86,67% APDEX - http://www.apdex.org Wednesday, August 25, 2010
  18. 18. Forecasting (Sort of...) Wednesday, August 25, 2010
  19. 19. System  C 40 35 30 Measured Throughput Users (N) X(N) 1 2,91 25 Throughput 20 2 5,67 4 10,86 15 8 18,65 10 16 25,91 32 36,68 5 64 37,34 0 0 10 20 30 40 50 60 70 Virtual  Users Measured  X Forecasting (Sort of...) Wednesday, August 25, 2010
  20. 20. System  C 45 40 35 30 25 Throughput 20 15 10 5 0 0 10 20 30 40 50 60 70 80 90 Virtual  Users Measured  X Poly.  (Measured  X) Forecasting (Sort of...) Wednesday, August 25, 2010
  21. 21. System  C 45 40 35 30 25 Throughput 20 15 10 5 0 0 10 20 30 40 50 60 70 80 90 Virtual  Users Measured  X Poly.  (Measured  X) Forecasting (Sort of...) Wednesday, August 25, 2010
  22. 22. System  C 40 35 30 25 Throughput 20 15 10 5 0 0 20 40 60 80 100 120 140 Virtual  Users Measured  X Modeled  X(N) Using the Universal Scalability Model Neil J. Gunther http://www.perfdynamics.com/Test/gcaprules.html#sec:scalability Wednesday, August 25, 2010
  23. 23. System  D 60 50 40 Throughput 30 20 10 0 0 100 200 300 400 500 600 Virtual  Users Measured  X Modeled  X(N) Using the Universal Scalability Model Neil J. Gunther http://www.perfdynamics.com/Test/gcaprules.html#sec:scalability Wednesday, August 25, 2010
  24. 24. Conclusões e Considerações • Velocidade é tão importante quanto disponibilidade • Fast is a requirement • O SLA deve serdo usuário experiência definido considerando a • Não existem bolas de cristal Wednesday, August 25, 2010
  25. 25. In God we trust. Everyone else please show me the data. Wednesday, August 25, 2010
  26. 26. http://capacitricks.wordpress.com/ Wednesday, August 25, 2010
  27. 27. How many servers do we need ? Wednesday, August 25, 2010
  28. 28. Redundância 55% 55% 55% 55% Wednesday, August 25, 2010
  29. 29. Redundância 73,3% X 73,3% 73,3% Wednesday, August 25, 2010
  30. 30. Redundância 110% X 110% X Wednesday, August 25, 2010
  31. 31. How many servers do we need ? Wednesday, August 25, 2010
  32. 32. How many servers do we need ? Wednesday, August 25, 2010
  33. 33. Perguntas ? Wednesday, August 25, 2010
  1. A particular slide catching your eye?

    Clipping is a handy way to collect important slides you want to go back to later.

×