Turbocharging php applications with zend server (workshop)

Eric Ritchie bids you welcome! • Senior Technical Consultant and Trainer at Zend Technologies • Zend Framework and PHP 5.3 ZCE • Eighteen years of system administration experience • Twelve years of PHP (3,4 & 5) and five years Zend Framework development experience • Hobbies: Sampling good wines/whiskies (gifts welcome) © All rights reserved. Zend Technologies, Inc.

Slavey Karadzhov welcomes you! • Senior Technical Consultant and Trainer at Zend Technologies • Zend Framework and PHP 5.3 ZCE • Many years of PHP (3,4 & 5), ZF 1 and galaxy of programming languages. • Hobbies: Travelling, sampling exotic dishes, outdoor sports, hype technologies. (in no particular order) © All rights reserved. Zend Technologies, Inc.

Jan Burkl • Senior Solution Consultant at Zend Technologies • Working with the dark side of the company • Zend Framework and PHP 5.3 ZCE • Moved 10 years ago from Java to PHP. • Hobbies: Soccer, Automation of Life. © All rights reserved. Zend Technologies, Inc.

Agenda • Defining the problem • The sharpest tool in the shed… Code Tracing • Laziness is good (at least for web servers): Using caching to avoid work • If we must work, then procrastinate: Use the Job Queue • Taking the heat off the disk: Ways to reduce disk I/O • The ghost in the machine: Don‘t forget about network performance • Don‘t talk too much: Less is more when sending data © All rights reserved. Zend Technologies, Inc.

What we will not cover (probably) • Database optimisation > caching • Webservice optimisation • Varnish and reverse proxying in general • CDNs (Content Delivery Networks) • Caching technology comparisons • Operating system level optimisations • Javascript performance in general (well, maybe…) • Network Traffic Analysis © All rights reserved. Zend Technologies, Inc.

First job: Bottleneck identification • Many possible tools:  Zend Server Event Monitoring  Profiling  microtime()  Slow query logs • One Swiss army knife:  Zend Server Code Tracing © All rights reserved. Zend Technologies, Inc.

Zend Code Tracing: How to store a trace? • Best way (in most cases)  … Use a web browser  Very quick and easy, but obviously not so good for POST requests  Also, we may not be allowed! © All rights reserved. Zend Technologies, Inc.

Zend Code Tracing: How to store a trace? • The official way  … Use event monitoring  Works for all requests where an event is generated  Great for catching random problems  Custom events allow for „programmatic“ generation of traces © All rights reserved. Zend Technologies, Inc.

Zend Code Tracing: What can go wrong. • Sometimes you fail to store a code trace, because:  The module is disabled  Code tracing is running in standby mode  Lack of permission when using „dump_data“  The triggered event was aggregated  The code trace was never collected © All rights reserved. Zend Technologies, Inc.

Zend Page Cache • Dynamic content is expensive, so don’t regenerate  Low impact, since no code changes may be required  Complete request cached, so a hit is like a static request  Multiple copies of pages can be maintained  Controlled by flexable and comprehensive rules • Sadly, nothing comes for free  Page design can render page caching useless  Highly dynamic content cannot be reasonably cached  Problems with stale content © All rights reserved. Zend Technologies, Inc.

Full Blown • Don‘t ajaxify everything!  Ever profiled 15 MVC runs in 1 request? • MVC really needed for displaying User Name?  Be pragmatic  Lightweight gateway  Reduce Bootstrapping time • Load ressources which are really needed $application->bootstrap(array('session')) ->run();  Remember that you‘re using PHP! © All rights reserved. Zend Technologies, Inc.

JS Architecture • Dojo, ExtJS, jQuery, ...  Is a library enough? • Javascript MVC, Angular.js, EmberJS, ...  TodoMVC: http://addyosmani.github.com/todomvc/  Comparison of JS-Framework • File loading strategy  Reduce amount of requests – CDN?  RequireJS  YUI compress  Rhino © All rights reserved. Zend Technologies, Inc.

Example – Dojo - build.js ./dojo-1.7.3/util/buildscripts/build.sh profile=./profiles/wikizf.profile.js action=release releaseDir=/var/www/wiki-zf-good.session.dev/public cssOptimize=comments layerOptimize=closure mini=true optimize=closure stripConsole=all selectorEngine=acme © All rights reserved. Zend Technologies, Inc.

Example – Dojo - xyz.profile.js dependencies = { layers: [{ name: "dojo.js", dependencies: [ "dojox.data.QueryReadStore", "dijit.form.ComboBox", "dijit.form.Form", "dojo.parser", ] }], prefixes: [ ["dojox", "../dojox" ], ["dijit", "../dijit"] ] }; © All rights reserved. Zend Technologies, Inc.

Data caching • Protects the data layer:  Helps prevent repetitive DB/web service calls  Far easier than scaling the database • But...  Requires code changes  Custom data is also problematic  Risk of delivering stale content  De-caching can bring down the data source © All rights reserved. Zend Technologies, Inc.

Network data cache • Some advantages  Only one cache to update/invalidate  Most effective way of protecting the data source • Many disadvantages...  Single point of failure (or uncertainty when in distributed mode)  Limits scalability  Performance bottleneck  Slower © All rights reserved. Zend Technologies, Inc.

Zend Data Cache: A shared memory cache • Many advantages  Completely scalable (copy/paste architecture)  No single point of failure  Does not require TCP/IP  Very fast • Some disadvantages...  Not the path of least resistance  More work for the data source or the developer  Your colleagues may laugh at you (but they would be wrong) © All rights reserved. Zend Technologies, Inc.

Zend Data Cache: Usage • Inserting into the cache  $res = zend_shm_cache_store($key, $value, $ttl)  $res = zend_disk_cache_store($key, $value, $ttl) • Reading from the cache  $value = zend_shm_cache_fetch($key)  $value = zend_disk_cache_fetch($key) • In all cases the key can contain a namespace to allow grouping of data, e.g. namespace::key © All rights reserved. Zend Technologies, Inc.

Zend Data Cache: Usage • Deleteing from the cache  $res = zend_shm_cache_delete($key)  $res = zend_disk_cache_delete($key) • Wiping the cache  $res = zend_shm_cache_clear($namespace)  $res = zend_disk_cache_clear($namespace) © All rights reserved. Zend Technologies, Inc.

Shared memory caching • Taming the disadvantages  True, we need to work a little, but... • We can use Zend Server‘s Job Queue to perform remote cache maintenance • We can get a list of active servers from the Web API provided by Zend Server • Really, we don‘t have to code much ourselves © All rights reserved. Zend Technologies, Inc.

Shared memory cache on steroids • Don‘t decache, update  Regenerate the data and insert into the cache  At a minimum do this for all components of the index page  Reduces data source load © All rights reserved. Zend Technologies, Inc.

Zend Job Queue Why do now what you can do a little later? Or, genius is being calm on the surface while being calculating in the background. Your application should do just that. Or, your server‘s marketing department. © All rights reserved. Zend Technologies, Inc.

Zend Job Queue • Doesn’t make PHP faster, but makes it look that way  Break slow tasks out of the user workflow  Offload the heavy lifting to a background server.  Delay expensive tasks to off peak hours  Helps to prevent repetitve work reducing overall load • Possible to create jobs which depend on other jobs, run regularly and have set priorites • Hooks into Zend Server‘s Event Monitoring component  Jobs can feedback status information  Problems appear in the central Zend Server event list © All rights reserved. Zend Technologies, Inc.

Zend Job Queue: Querying jobs • Finding out the status of our job using the job id  Find out if the job is still queued, running, completed or failed  When finished we get a copy of the script‘s output © All rights reserved. Zend Technologies, Inc.

Zend Job Queue: Querying jobs • Searching for jobs  Search for an existing queued job of the same type  Useful for avoiding repetitive work  Returns the number of matching jobs along with job details © All rights reserved. Zend Technologies, Inc.

Optimization vs. Complexity • With every optimization level complexity gets higher  Who else drapped into the Caching pitfall? • First: implement functionality • System performing? • If not, check why • Get rid of bottlenecks © All rights reserved. Zend Technologies, Inc.

Reducing disk I/O • Use an opcode cache e.g. Zend Optimizer+  Avoids the need to open PHP files and compile the contents for every request.  Compiled „opcodes“ are instead held in shared memory for later reuse.  Particularly important for framework based projects where tens of files may be needed to answer one request.  Also saves some compiler time, but disk I/O savings are usually far more significant.  Most obvious benefit for scripts with short run times or running on loaded servers. © All rights reserved. Zend Technologies, Inc.

Reducing disk I/O: Zend Optimizer+ • Performs opcode caching • Also adds multiple optimisation passes • Allows for the blacklisting of files/directories • Simple configuration: © All rights reserved. Zend Technologies, Inc.

Reducing disk I/O • Store static files elsewhere • Do not cache to the file system • Must have local storage? ...Use a ramdisk • Reduce PHP/Apache logging to minimal levels  E_COMPILE_ERROR|E_RECOVERABLE_ERROR|E_ERROR|E_CORE_ERROR • Don‘t forget the Zend Server logs!  Search for log_verbosity_level in the configuration © All rights reserved. Zend Technologies, Inc.

File system concerns • EXT3 – The default problem  For a web server ReiserFS can be 20x (or more) faster, really!  Even XFS is 2.5x faster  If you must use EXT3/4 then throw memory at the server • My NFS/Samba server is fast... Honest!  Even „fast“ NFS servers have terrible performance under load  Older implementations cannot use buffer cache  Poor scalability • Distributed file systems  Extremely poor scalability © All rights reserved. Zend Technologies, Inc.

PHP uses your network • PHP communicates with network services like  Databases (ex: MySQL, Oracle)  Caching systems (ex: Memcache, Redis ..)  Job Queue Systems (ex: Zend Job Queue, RabbitMQ )  Session Clustering Daemon (ex: Zend Session Clustering) • If one of these services overloads the network then all other suffer from slowdown  Network congestion  Insufficient bandwidth  High latency © All rights reserved. Zend Technologies, Inc.

Overloaded network, slow application • The caching systems have to be very fast  If they are slow then the complete application can be slow • The session clustering has to be fast  Lots of PHP requests wait to get a read/write lock -> the application looks slow © All rights reserved. Zend Technologies, Inc.

Know your traffic • A bit of Linux/Unix networking knowledge is required • Count the incoming and outgoing traffic  iptables • Inspect the network streams  Tcpdump  Wireshark © All rights reserved. Zend Technologies, Inc.

Traffic counters • Network traffic can be  On different interfaces: • eth0, lo  Incoming/Outgoing  On different ports  Uses different transport protocols © All rights reserved. Zend Technologies, Inc.

Traffic counters (2) • Iptables can help you  Zeroing the counters • iptables -t mangle -L -Z INPUT -v -n • iptables -t mangle -L -Z OUTPUT -v -n  Start measuring • Ex: Memcache outgoing – iptables -t mangle -I OUTPUT -p tcp --dport 11211 • Ex: Memcache incoming via udp – iptables -t mangle -I INPUT -p udp --sport 11211 © All rights reserved. Zend Technologies, Inc.

Turbocharging php applications with zend server (workshop)

More Related Content

What's hot

Similar to Turbocharging php applications with zend server (workshop)

Recently uploaded

Turbocharging php applications with zend server (workshop)