performance and scalability

27
Performance & Scalability Andy Thornton

Upload: mediacurrent

Post on 08-May-2015

8.491 views

Category:

Technology


2 download

DESCRIPTION

Introduction to Drupal Performance tuning.

TRANSCRIPT

Page 1: Performance and Scalability

Performance & Scalability

Andy Thornton

Page 2: Performance and Scalability

This is my apology for making my slideshow very text heavy ....

An opening Apology

However, I added some fun pictures, to hopefully compensate ... 

And yes, these slides will be available for download.

Page 3: Performance and Scalability

Who's this presentation for?Who's this presentation for?

Page 4: Performance and Scalability

Agenda

Capacity Planning and ArchitectureTesting ToolsPressflow & VarnishWhere are the pinch points?PHPCachingMySQLApacheDrupalTweaksHandy ToolsFurther ReadingQuestions?

Page 5: Performance and Scalability

Determine Metrics » Test Plan » Analyse Results » Rinse and repeat

Capacity Planning

“However Beautiful the strategy, you should occasionally look at the results”

- Winston Churchill

Page 6: Performance and Scalability

Testing Tools

The GrinderA java based load testing tool.http://grinder.sourceforge.net/

JMeterJava based tool, designed to load test functional behavior and measure performance.http://jakarta.apache.org/jmeter/

Openloadhttp://www.opendemand.com/openload/

SoastaLeveraging resources from the cloud, Soasta is fast becoming a leader in performance testing. (Acquia Partner)http://www.soasta.com/

RRD ToolOpenSource high performance data logging and graphing system for time series data. http://www.mrtg.org/rrdtool/

XHProfProfiler for PHP providing some valuable metrics and reporting.http://mirror.facebook.net/facebook/xhprof/doc.html

Page 7: Performance and Scalability

Acquia & Soasta

CloudTest Dashboard

Page 8: Performance and Scalability

Drupal Distribution Optimised for performance and scale

• Support for database replication• Support for Squid or Varnish as reverse proxy caches• Optimised for MySQL• Optimised for PHP• Supported by Acquia

Varnish

Varnish stores web pages in memory so the web servers don't have to create the same web page over and over again. The web server only recreates a page when it is changed.

Reporting from Varnish can provide some good data to help with capacity planning.

Page 9: Performance and Scalability

Investigating Issues

Memory Tapped Out?    Apache is generally the culprit, fine tune your config and possibly add more memory.

CPU is maxing out?

    PHP is your prime suspect as it's CPU intensive, look into using an Opcode cache.

Hard Drives are getting Thrashed!

    Take a look at MySQL, enabling the query cache may help. Fine tune your queries.

Page 10: Performance and Scalability

Pain points?

Apache

PHPMySQL

Page 11: Performance and Scalability

Accelerators (Opcode Caching)• Alternative PHP Cache (APC)• eAccelerator• XCache• Nusphere PhpExpress• Zend Accelerator 

Application Profiling

Some applications work well on small scale sites, but get bogged down when moved into production.

• Zend Studio• Komodo• Eclipse (PDT)• Xdebug• Aptanta Studio

PHP parses and compiles all code into an intermediary series of opcodes, or more accurately an intermediary series of opcodes for each request.

Application ProcessesCreating new processes is a slow operation. CGI applications that fork a new  process on every invocation will run substantially slower the more processes are running.

Running PHP in multi-threaded mode can improve response times, but consider disabling unused services, for example:

• telnetd, inetd, atd, ftpd, lpd, sambad• sendmail• portmap• xsf, windows manager, xinit, X

Page 12: Performance and Scalability

Tip 1: If APC is compiled with mmap support, you can use /dev/zero (usually the default)

/dev/zero

apc.mmap_file_mask = /dev/zero

Andy will explain why /dev/zero is handy...

Tip 2: Make sure to give it enough memory.Tip 3: Always measure and adjust as needed.

Sometimes, a cache setup can slow things down!

When Drupal Attacks!

Page 13: Performance and Scalability

Andy's Page of Weird Apache Commands

Display currently loaded modulesapachectl -t -D DUMP_MODULESGood candidates to remove, mod_cgi, mod_dav, mod_ldap*How do I see if that made a difference?ab -n 100 -c 5 http://www.domain.com/test.html

Is the swap file being used?vmstat 1 60Runs vmstat every 1 second 60 timesDecreasing Apache TimeoutsNormally the default is set to 5 minutes (300 seconds), how about reducing it to 20 seconds?TimeOut 20 You can do this in your virtual host or server config.

Page 14: Performance and Scalability

Maxclients Setting in Apache

Whats a good way to find the maxclients number?

    MaxClients ≈ (Ram - size of all other processes) / (size of the apache process)

We are making an educated guess based on dividing the system memory (our physical RAM) by the maximum size of an apache process, with enough wiggle room to have the operating system run smoothly.

To find the size of a running process;

    ps -ylC apache2 --sort:rss

Divide the size by 1024 to get the process size in meg. You can also use pmap, use top to find the pid, then use;

    pmap -x <pid id>

Another handy way of seeing how your memory is doing.

    free -m 

and good old VMSTAT to see if your memory is being paged

    vmstat 5 60

Page 15: Performance and Scalability

Huge List of Apache Tips

mod_expires Allows Drupal to send out http expires headers caching all files in the users browsers (~2 weeks) or until a new page is made available.

This is good for all static files / images / css / javascript.

Drupal is pre-configured to use mod_exp if it's available. Configure it's use in your .htaccess

# Cache files for 2 weeks after accessExpiresActive OnExpiresDefault A1209600

# Don't Cache Dynamic pagesExpiresByType text/html A1

You don't let Apache cache HTML content as  Drupal's content isn't static. This is why Drupal uses it's own cache.

Reduce DNS LookupsYou can tell Apache to not perform a DNS lookup on files by their mime type.

HostnameLookups Off<Files ~"\.(html/cgi)$">    HostnameLookup On</Files>

File NegotiationBe specific when specifying filenames rather than wildcards (when possible).

Instead of:DirectoryIndex index

Use:DirectoryIndex index.cgi index.php

Page 16: Performance and Scalability

More Apache Tips ...

mod_deflate this allows your web server to compress files before being sending them to the users browser

AddOutputFilterByType Deflate text/html text/css

More advanced example:

SetOutputFilter DEFLATE

# Netscape 4.x has some problems...BrowserMatch ^Mozilla/4 gzip-only-text/html

# Netscape 4.06-4.08 have some more problemsBrowserMatch ^Mozilla/4\.0[678] no-gzip

# MSIE masquerades as Netscape, but it is fine# BrowserMatch \bMSIE !no-gzip !gzip-only-text/html

# Don't compress imagesSetEnvIfNoCase Request_URI \\.(?:gif|jpe?g|png)$ no-gzip dont-vary

htaccess If you have access to your VirtualHosts in Apache, move your directives out of htaccess and move them in to the VirtualHost for your website. 

The reason for this is that Apache loads your virtual host once when started but Drupal searches for htaccess files in multiple directories at runtime.

Disable htaccess lookups with:

<Directory />  AllowOveride None</Directory>

Page 17: Performance and Scalability

Webserver

Worker Uses multiple child processes with many threads each. Each thread handles one connection at a time.

PreforkUses multiple child processes each with one connection at a time. On many systems it's comparable to worker (in speed) but it uses more memory. Generally, recommended for Drupal due to it's threading model.

Alternatives to Apache

Nginx (engine-X)Faster than Apache and has more predictable memory usage.  Not as straight forward to setup (rewrite rules for example).

LIGHTHTTPDGood performance. Although there has been discussion on the drupal forums as to its ability to cleanly run Drupal 7. Not for the feint of heart.

Microsoft Web MatrixRuns Drupal under IIS with PHP, good for Microsoft shops.http://www.microsoft.com/web/drupal/

Apache supports pluggable concurrency modules, called Multi Processing Modules. So which is a good fit for my website?

Page 18: Performance and Scalability

Drupal does a lot of work in the database, especially for authenticated users and modules, so how can we get the best out of our database?

Enable Query CacheThis feature is generally disabled by default. To enable it, assign a value to query_cache_size in your mysql configuration file.

[mysqld]query_cache_size =  64M

You can query the setting as it's a variable. 

SHOW VARIABLES LIKE 'have_query_cache';

You may have to do some testing to find the best value to use.

Logging Slow QueriesYou can instruct MySQL to log all queries that take too long to run, for later analysis.

[mysqld]log_slow_queries=/var/log/slow-queries.loglog_query_time=5

Leaving off the query time will default it to 10 seconds.

Analysing Slow QueriesPrepend your query with EXPLAIN and run it for more information, or analyse your query with MAATKIT.

EXPLAIN will show which indices are being used, sometime just indexing the table can be a good fix.

Database Optimisations

Page 19: Performance and Scalability

Table locking can be a good indicator of problems in your database. 

MyISAM and InnoDbPerformance wise, they both stand up well, but whats the difference for Drupal?

Table Locking

MyISAM = Table Level LockingInnoDB   = Row level Locking

How do I know if I need to make a change?Take a look at 

SHOW STATUS LIKE 'Table%';

Table_locks_immediate 1151552Table_locks_waited 15324

How do I change a table type?

ALTER TABLE accesslog TYPE='InnoDB';

Good Candidates for InnoDBWhich tables would benefit from changing to use row level locking.

• Cache tables• Watchdog• Sessions• Accesslog

Tipsmysqlreport can be automated to show any wait times for transactions.

Keep your cron short to prune tables regularly if tables get too big.

You can use the Devel module to identify Query expensive pages in Drupal.

More Database Tips

Page 20: Performance and Scalability

Drupal 6

Caching in Drupal can be enabled through the Performance module in Admin and settings.php. 

Drupal Ships with 6 Cache tables• cache• cache_block• cache_page• cache_filter• cache_menu• cache_form

Developer TipWhen writing your own modules and need to cache data, think about using your own tables. It reduces write contention with Drupal for using the cache tables and doesn't bloat the table with your data.

Most functions also have a $reset parameter which instructs the function to clear down it's internal cache.

Consolidate Javascript & CSSEnabling these options "smushes" javascript and stylesheets into single files and minifies them. 

Drupal Performance Module Options

Normal: Drupal bootstraps in phases, when normal is selected it uses just enough phases to load a page from cache. Keeping db queries to a minimum.

Aggressive: Completely bypasses loading of all modules. Boot and Exit hooks are never called for cached pages. This means less PHP is parsed since no modules are loaded. There's are also fewer database calls. 

Fastpath: Not enabled from the admin panel, this option is enabled from settings.php. The idea is that a call to the file system is faster as there's no ramping up for a database query. This may not scale across load balanced hosting..

Page 21: Performance and Scalability

Drupal 6

Drupal doesn't store session information for the first anonymous visitor. This is so webcrawlers and spiders don't fill your session tables up. However, these tables can get very large. 

Garbage CollectionDefault value for garbage collection is a little over 2 days, you can also increase the frequency of collection with

session.gc_maxlifetime (seconds)session.cache_expire    (minutes)

Note: When you adjust maxlifetime, adjust cache_expire to be the same.

Tip: As Drupal can serve cached pages to anonymous users and anonymous users don't normally use interactive features of Drupal. How about reducing the time they are logged in or log them out when they close their browser.

# 86400 seconds = 24 hourssession.cookie_lifetime, 86400# Logout on browser closesession.cookie_lifetime, 0

Pruning SessionsDrupal controls when session start by turning off PHP's session autostart functionality in htaccess.

php_value session.auto_start 0

The session table is cleared out when PHP's garbage collection runs. The lifetime of a session record is determined by

session.gc.maxlifetime (seconds)

Other settings you can experiment with:

session.cache_expiresession.cache_limitersession.cookie_lifetimesession.save_handlersession.use_only_cookiessession.use_trans_sid

Page 22: Performance and Scalability

Not a great deal has changed with Drupal 7, there's been a lot of code optimisation and some elements of Pressflow have made their way into the Drupal 7 codebase. 

However, the following changes are notable.

Removed

• Caching Mode Option on the performance page• Throttling Block throttling.

New Features

• New Performance Logging• Improved performance on uncached pages• When logged in, performs fewer queries on path alias lookups• Tighter Integration with APC

Whats Changed in Drupal 7

Page 23: Performance and Scalability

Google CSEGoogle's custom search service with some impressive features, including branding support and on-the-fly indexing support. Well documented API, but requires some work to get setup initially.http://www.google.com/cse/

Acquia SearchBuilt upon Lucene and Solr from Apache, hosted service with easy integration with Drupal. All administration tasks are built in to the admin panel. A very powerful alternative to Drupal's built in search.http://acquia.com/products-services/acquia-search 

Alternatives to Search

Apache SolrJava based open source enterprise search platform from the Apache Lucene project.http://lucene.apache.org/solr/

Page 24: Performance and Scalability

Theme OptimisationUse of sprite maps / minified code / Optimised Images / Google JSAPI / Browser CachingExpiring Content / Reduce HTTP requests / Non Blocking Javascript

Throttling and Block CachingEnabling throttling allows you to turn off modules & blocks when the system starts to get sluggish. You can set the threshold in the admin panel, and determine which modules and blocks to turn off from their respective admin pages.

ModulesCacherouter / Boost / Devel / Authcache

Content Delivery NetworksThe capacity sum of strategically placed servers can result in an impressive boost in the number of concurrent users. 

• Akamai Technologies• Amazon CloudFront• BitGravity• CacheFly• CD Networks• Windows Azure

More Tweaks

Page 25: Performance and Scalability

Web Load Testing & Page Audithttp://www.webpagetest.org/

Open Source Testing Toolshttp://www.opensourcetesting.org/performance.php

JMeter http://jakarta.apache.org/jmeter/

Apache Benchhttp://httpd.apache.org/docs/2.0/programs/ab.html

mysqlreporthttp://hackmysql.com/mysqlreport

YSlow (Firebug)http://developer.yahoo.com/yslow/

Performance Articles (Google Code)http://code.google.com/speed/articles/

Devel Module (Drupal) http://drupal.com/project/devel

Ubuntu, Landscapehttp://www.canonical.com/enterprise-services/landscape

SOASTA CloudTesthttp://acquia.com/SOASTA/next-steps

Client (browser based tools)Boomerang / Firebug / Chrome Dev Tools / 

MAATKIThttp://www.maatkit.org/

Handy Tools

Page 26: Performance and Scalability

The Art Of Capacity PlanningWeb-based companies live or die by the ability to scale their infrastructure to accommodate increasing demand. This book is a hands-on and practical guide to planning for such growth. 

O'Reilley Media : Amazon

High Performance MySQLHigh Performance MySQL is the definitive guide to building fast, reliable systems with MySQL. This book covers every aspect of MySQL performance in detail and focuses on robustness, security and data integrity.

O'Reilley Media : Amazon

Even Faster WebsitesSteve Sounders works on the performance team at Google and has written a couple of great books on performance. 

O'Reilley Media : Bio

Further Reading

Page 27: Performance and Scalability

Questions?

[email protected]

Slides will be made available on http://www.drupalatlanta.com/ after the presentation.