SlideShare a Scribd company logo

Skalowalna architektura na przykładzie soccerway.com

•
•
Spodek 2.0
Spodek 2.0

This document discusses scalable web application architecture. It describes using load balancing and database replication across multiple servers to handle high traffic volumes. It emphasizes using caching at various levels (PHP, databases, CDNs) and preprocessing data to improve performance. A CDN is recommended for distributing static resources globally to reduce latency and failure risk.

1 of 35
Download to read offline
Scalable architecture By Adam Brodziak Global Sports Media b.v.
Abstract  Adam Brodziak An overview of modern web-based application architecture - from hardware infrastructure, through PHP/SQL code, HTML/CSS markup distribution. All of this spiced up by cache, loadbalancing and CDN.
Who is this guy?  Lead developer at Global Sports Media  GSM collects and process sports data  GSM owns soccerway.com portal  Linux user  Interested in frameworks, design patterns  Semantic Web enthousiast  Football (soccer) fan
Topics  The Challenge  Infrastructure  Code  Cache  CDN
Topics  The Challenge  Infrastructure  Code  Cache  CDN
Raw numbers  7 millions visits / month  52 millions pageviews / month  1 billion request / month  6TB of traffic / month  300k users at peak time  Quite a few clients using the same hardware
Not so much, but...  700 leagues  Livescores  Game events  Match statistics  Rankings  Editorials
Traffic growth
The Challenge  Loads of data to process  Scores  Events  Stats  In real-time (livescores)  Growing number of visitors  13K hits/sec at peak-time
10 servers to run it all
Topics  The Challenge  Infrastructure  Code  Cache  CDN
It starts with one
Load balancing
Loadbalancing caveats  Don't relay on the local filesystem  Temporary files, session, logs  Avoid assuming exclusive/single cache  APC, Zend Cache  Use distributed session storage  Memcache, database  Encalsulate above
Separate database server
DB replication
Replicaton caveats  Writes only on master  Reads from slaves  Data consistency  Replication lag  Don't do $master->query('UPDATE session SET logged = 1'); $slave->query('SELECT logged FROM session');
Whole image
Topics  The Challenge  Infrastructure  Code  Cache  CDN
PHP is slow!  Yes, but it does not matter!  Database access is slower  Cache over network is slower  Disk access is slower  HTTP requests are slower  Webservice calls are slower  Discover bottlenecks before blaming PHP
It's about architecture  Heavy tasks in background  CRON, Gearman  Pregenerate stuff  Move some code to SQL  Calculations in queries  Stored procedures  Triggers  C/C++ or Java for heavy computation  Use PHP to glue it together
PHP Frameworks  Hundreds of others  Which one to choose?
Framework? Think again!  Raw performance matters  Support for master-slave replication  Multiple layers of cache  Working with accelerators (HipHop!)  Beware of bottlenecks  i.e. core part of framework is slow  Designed to scale
Topics  The Challenge  Infrastructure  Code  Cache  CDN
Cache is everywhere  CPU: L1, L2  Disk buffer  Linux filesystem  MySQL  PHP (APC)  Smarty  HTTP Proxy  Browser cache
Where to cache?
Memory is cheap  Pre-generate stuff  Store results in memory  APC, memcached  App config in memory  APC with stat=off  Increase RAM for MySQL  Disk is the new tape
Memcached for the rescue!  Dead simple  Key-value  Distributed storage pool  Automatic invalidation after X sec  No garbage collecting invoked  Store arrays, objects, simple values  Easy integration
Topics  The Challenge  Infrastructure  Code  Cache  CDN
Reverse-proxy  First line of cache  Returns content if resource is up-to-date  Works on HTTP level  Can be integrated into existing infrastructure  Can do load balancing  In-memory cache storage  Squid, Nginx, Varnish
Content Delivery Network  Network of servers  Worldwide  Automatic loadbalancing  Fast access (low ping time)  Data redundancy gratis  Ideal for static resources  But not only  Must-have for worldwide websites
CDN as reverse-proxy  HTTP request / response chain  Embraces REST architecture  Requests are distributed  Reduces latency  Lowers traffic volume  Increases availability  i.e. Akamai Edge Suite
CDN at soccerway.com  All of the content is served via CDN  Images, CSS, JS  Generated HTML  JSON for Ajax  90% of traffic via CDN  Origin requests only from Europe  Site online even if servers are down  Can't live without ;)
Thank you for listening Questions?
Interested?  Contact me:  adam@globalsportsmedia.com  www.goldenline.pl/adam-brodziak  www.linkedin.com/in/adambrodziak  We're hiring!  Web developers  Football / sport fans

More Related Content

Skalowalna architektura na przykładzie soccerway.com

  • 1. Scalable architecture By Adam Brodziak Global Sports Media b.v.
  • 2. Abstract  Adam Brodziak An overview of modern web-based application architecture - from hardware infrastructure, through PHP/SQL code, HTML/CSS markup distribution. All of this spiced up by cache, loadbalancing and CDN.
  • 3. Who is this guy?  Lead developer at Global Sports Media  GSM collects and process sports data  GSM owns soccerway.com portal  Linux user  Interested in frameworks, design patterns  Semantic Web enthousiast  Football (soccer) fan
  • 4. Topics  The Challenge  Infrastructure  Code  Cache  CDN
  • 5. Topics  The Challenge  Infrastructure  Code  Cache  CDN
  • 6. Raw numbers  7 millions visits / month  52 millions pageviews / month  1 billion request / month  6TB of traffic / month  300k users at peak time  Quite a few clients using the same hardware
  • 7. Not so much, but...  700 leagues  Livescores  Game events  Match statistics  Rankings  Editorials
  • 9. The Challenge  Loads of data to process  Scores  Events  Stats  In real-time (livescores)  Growing number of visitors  13K hits/sec at peak-time
  • 10. 10 servers to run it all
  • 11. Topics  The Challenge  Infrastructure  Code  Cache  CDN
  • 14. Loadbalancing caveats  Don't relay on the local filesystem  Temporary files, session, logs  Avoid assuming exclusive/single cache  APC, Zend Cache  Use distributed session storage  Memcache, database  Encalsulate above
  • 17. Replicaton caveats  Writes only on master  Reads from slaves  Data consistency  Replication lag  Don't do $master->query('UPDATE session SET logged = 1'); $slave->query('SELECT logged FROM session');
  • 19. Topics  The Challenge  Infrastructure  Code  Cache  CDN
  • 20. PHP is slow!  Yes, but it does not matter!  Database access is slower  Cache over network is slower  Disk access is slower  HTTP requests are slower  Webservice calls are slower  Discover bottlenecks before blaming PHP
  • 21. It's about architecture  Heavy tasks in background  CRON, Gearman  Pregenerate stuff  Move some code to SQL  Calculations in queries  Stored procedures  Triggers  C/C++ or Java for heavy computation  Use PHP to glue it together
  • 22. PHP Frameworks  Hundreds of others  Which one to choose?
  • 23. Framework? Think again!  Raw performance matters  Support for master-slave replication  Multiple layers of cache  Working with accelerators (HipHop!)  Beware of bottlenecks  i.e. core part of framework is slow  Designed to scale
  • 24. Topics  The Challenge  Infrastructure  Code  Cache  CDN
  • 25. Cache is everywhere  CPU: L1, L2  Disk buffer  Linux filesystem  MySQL  PHP (APC)  Smarty  HTTP Proxy  Browser cache
  • 27. Memory is cheap  Pre-generate stuff  Store results in memory  APC, memcached  App config in memory  APC with stat=off  Increase RAM for MySQL  Disk is the new tape
  • 28. Memcached for the rescue!  Dead simple  Key-value  Distributed storage pool  Automatic invalidation after X sec  No garbage collecting invoked  Store arrays, objects, simple values  Easy integration
  • 29. Topics  The Challenge  Infrastructure  Code  Cache  CDN
  • 30. Reverse-proxy  First line of cache  Returns content if resource is up-to-date  Works on HTTP level  Can be integrated into existing infrastructure  Can do load balancing  In-memory cache storage  Squid, Nginx, Varnish
  • 31. Content Delivery Network  Network of servers  Worldwide  Automatic loadbalancing  Fast access (low ping time)  Data redundancy gratis  Ideal for static resources  But not only  Must-have for worldwide websites
  • 32. CDN as reverse-proxy  HTTP request / response chain  Embraces REST architecture  Requests are distributed  Reduces latency  Lowers traffic volume  Increases availability  i.e. Akamai Edge Suite
  • 33. CDN at soccerway.com  All of the content is served via CDN  Images, CSS, JS  Generated HTML  JSON for Ajax  90% of traffic via CDN  Origin requests only from Europe  Site online even if servers are down  Can't live without ;)
  • 34. Thank you for listening Questions?
  • 35. Interested?  Contact me:  adam@globalsportsmedia.com  www.goldenline.pl/adam-brodziak  www.linkedin.com/in/adambrodziak  We're hiring!  Web developers  Football / sport fans