01 zingme practice for building scalable website with php
2 likes351 views
This document discusses best practices for building scalable PHP websites, as demonstrated by ZingMe's architecture. It covers scaling by adding servers horizontally (scaling out) rather than vertically on single servers. Key aspects covered include load balancing with HAProxy, managing user sessions in Memcached to allow scaling, and deploying code changes automatically across servers using SVN and scripts. ZingMe's implementation supports over 1.7 million daily active users on over 300 servers through these scalability techniques.
01 zingme practice for building scalable website with php
- 1. ZingMe Practice For Building Scalable PHP Website By Chau Nguyen Nhat Thanh ZingMe Technical Manager Web Technical - VNG
- 2. Agenda ● About ZingMe ● Scaling PHP application ● Scalability definition ● Scaling up vs scaling out ● Load balancing ● Problem with PHP when using load balancer ● User session management ● Problem with code deployment 2
- 3. Agenda(cont.) ● ZingMe practice in building scalable PHP applications ● Choosing scale out ● Load balancing model ● User session management system ● Code deployment system ● ZingMe more ● Open social API
- 4. About ZingMe ● A social network
- 5. About ZingMe ● 25M registered accounts ● 1.7M daily active users ● 5M monthly active users ● > 300 servers (1.2 K cores) ● 50 DB servers, >40 Memcached servers ● ~ 100 Web servers ● 40 servers for Hadoop farm ● Others (storage, ...)
- 6. About ZingMe ● Load balancing: HA proxy, LVS, hardware (?) ● Web server: Nginx, lighttpd, apache, hardware (?) ● Web caching: squid, varnish, hardware (?) ● Caching: Redis, memcached ● Programming language: PHP, C++,Java, python, bash script ● Searching: Solr, lucene ● DB: MySQL , NOSQL ● Log system: Scriber + Hadoop
- 7. Scaling PHP application ● Scalability definition ● Scaling up vs scaling out ● Load balancing ● User session management ● Code deployment
- 8. Scalability definition “Scalability is a desirable property of a system, a network, or a process, which indicates its ability to either handle growing amounts of work in a graceful manner or to be enlarged” from Wikipedia ”A web application is scalable when is able to manage a growing traffic with additional resources (CPU, RAM) without software changes” by Jan Burkl, Enrico Zimuel (Zend Technologies)
- 9. Scaling up vs scaling out ● Scaling up (vertically) ● Add more resources to single node in a system ● Resources can be software or hardware ● Make system more powerful by increase node powerful ● Has an upper limit ● Hard to implement HA solution ● Cost: expensive
- 10. Scaling up vs scaling out ● Scaling out (horizontally) ● Add more nodes to a system ● Make system more powerful by increase number of nodes in the system ● Maybe has no upper limit ● Easy to implement HA solution ● Cost: cheap
- 11. Scaling up vs scaling out • From slide ”How to scale PHP application” • by Jan Burkl, Enrico Zimuel (Zend Technologies)
- 12. Load balancing ● The need of a load balancer ● Most of big web site choose scale out ● Need a component for distribute load among the nodes ● Such component called load balancer
- 13. Load balancing • from ”Server load balancing architectures” • by Gregor Roth, JavaWorld.com, 10/21/08
- 14. Load balancing ● How load balancer work? ● Has 2 interfaces: Back-end, Front-end ● Front-end receive the request from client ● Back-end forward request to internal node ● Back-end receive the response from node ● Front-end response to client ● Status monitor for monitoring the back-end interface
- 15. Load balancing ● Web application load balancer ● Work in layer 7 ● Can filter HTTP header and body in both request and response ● Can be hardware or software ● F5 BIG-IP, Citrix Netscale, Cisco ● HA Proxy, Nginx, lighttpd
- 16. Problem with PHP when using load balancer ● PHP stores session info into local disk by default ● How to scale session to 2 nodes ? ● Can be scaled if: ● Load balancer has a method for routing the same user requests into the same nodes (1) ● Override the default PHP mechanism for storing session of all nodes in the same storage (2)
- 17. Problem with PHP when using load balancer ● Method (1) ● Load balancer must keep state for connection ● Overhead for routing request ● Not balancing ● Method (2) ● Must change software ● No overhead for routing request ● More balancer than (1)
- 18. PHP session management ● Choose method (2) ● Centralize storage among web server ● Storage can be NFS, DB, memcached... ● All web servers access the same storage ● Override the default behavior of PHP session functions ● Bottleneck can be in storage
- 19. PHP session management ● Call session_set_save_handler( callback $open , callback $close , callback $read , callback $write , callback $destroy , callback $gc ) to override default behavior ● More infos: ● http://php.net/manual/en/function.sessio n-set-save-handler.php
- 20. Problem with code deployment ● Adding new server to system: ● What kind of configuration must be changed? ● Which code will be deployed ? In which server? ● How can we monitor them?
- 21. Problem with PHP code deployment ● What will we change in PHP code ? ● Separate all configuration to configuration file ● Do not hard code any configuration ● Put the code in a central server like SVN ● Use scripts to pull all code to production server ● Build profiler system for monitor the code
- 22. ZingMe practice for building scalable web site ● Choose Scale out ● Use HA proxy for load balancing ● Use Memcached for session storage ● Simple code deployment system base on SVN and bash script
- 23. ZingMe practice for building scalable web site ● Divide into small services ● Use commodity hardwares ● Use open source software ● Server has the same environment ● Add new server < 5mins
- 24. ZingMe load balancing model
- 25. ZingMe load balancing model ● Use HA proxy ● HTTP based proxy ● High availability ● Use epoll ● Use content switching to divide the service into group ● Define load balancing factors based on the power of the back-end server
- 26. ZingMe load balancing model Normal traffic
- 27. ZingMe load balancing model ZingMe traffic
- 28. ZingMe user session management ● Use memcached for high performance ● Override Zend_Session ● 3K hits/s
- 29. Why we choose memcache? ● L1 cache reference 0.5 ns ● Branch mispredict 5 ns ● L2 cache reference 7 ns ● Mutex lock/unlock 25 ns ● Main memory reference 100 ns ● Compress 1K bytes with Zippy 3,000 ns ● Send 2K bytes over 1 Gbps network 20,000 ns ● Read 1 MB sequentially from memory 250,000 ns ● Round trip within same datacenter 500,000 ns ● Disk seek 10,000,000 ns ● Read 1 MB sequentially from disk 20,000,000 ns ● Send packet CA->Netherlands->CA 150,000,000 ns From Jeff Dean - google.com
- 30. ZingMe code deployment ● Developer upload code to SVN ● SO runs scripts to deploy code to production server ● Has staging server for real environment test ● Dev cannot touch to production environment ● Using environment parameter ● Nginx: fastcgi_param ● Apache: setenv
- 31. ZingMe more ● Standard latency for server side code processing < 200ms ● Live demos Profiler system ● Open social API updated infos ● Sandbox for developer ● Will be open soon
- 32. ZingMe more ● Optimize node before scale it up ● Performance C++ > Java > PHP ● Why PHP ? – Easy to implement business – Easy to deploy ● Why C++? – Performance ● Can we mix PHP and C++? – Next session will answer
- 33. Q&A Contact info: Chau Nguyen Nhat Thanh thanhcnn@vinagame.com.vn me.zing.vn/thanhcnn2000