PHP Performance on IBM i - Seiden Group slides/PHP on IBM i... · PHP Performance on IBM i ... • Recommends indexes across all queries!17. Alan Seiden Consulting Let your PHP apps

PHP Performance on IBM i

Part 2 Accelerate your application

www.SeidenGroup.com

Let your PHP apps fly on IBM i: High Performance PHP (Part 2)Alan Seiden Consulting

Speaker: Alan Seiden

• Principal at Seiden Group • Mentor CIOs and development teams

• Deliver modern technical solutions

• Host and sponsor of CIO Summit

• Club Seiden: The next generation • Open source advocate, contributor

!2


Seiden Group

!3

• Team of experts who: • Mentor IBM i teams, IT Directors, CIOs

• Consult on projects and process

• Develop applications

• Troubleshootwww.seidengroup.com


Club Seiden

!4

Forum for younger developers with a passion for open source, IBM i and collaboration


Second of three sessions on peformance

Session 1: How fast? Find the bottlenecks! (measure, measure, measure)

Session 2: Accelerate your application * (speed up PHP and the server back end)

Session 3: Ensuring a fast user experience (speed up the front end)

* You are here

!5


Session agenda

• DB2 and PHP performance tips

• Toolkit

• Zend Server configuration for speed

• Reducing IFS accesses

• Go asynchronous

!6


PHP 7

!7


PHP 7 is faster than previous versions

!8

• PHP 7.x comes with Zend Server 9.x• Speed boost• Spurred by competition with Facebook’s “Hack”


Let’s try a simple speed test<?php $profile['start'] = microtime( true );

for ($i=0 ; $i<1000000 ; $i++) { serialize($i); } $profile['end'] = microtime( true ); $profile['duration'] = $profile['end'] - $profile['start']; echo "Took " . $profile['duration']. " seconds ";

Try it from CALL QP2TERM or SSH terminal: cd /usr/local/zendsvr6/bin php -n /www/zendphp7/htdocs/test/speedtest.php cd /usr/local/zendphp7/bin php -n /www/zendphp7/htdocs/test/speedtest.php

!9


PHP and Apache jobs

!10


Number of PHP jobs

• FastCGI PHP “children” § /www/zendphp7/conf/fastcgi.conf § Default: SetEnv=“PHP_FCGI_CHILDREN=10" § Increase from 10 to # of expected concurrent PHP requests § Each “child” needs temp storage, CPU, memory; find a balance

§

!11


Number of Apache HTTP threads

• Apache’s ThreadsPerChild • /www/zendphp7/conf/httpd.conf • Default: ThreadsPerChild 40 • Increase to number of expected HTTP connections

• PHP requests • CSS • Javascript • AJAX requests

!12


Active and idle thread dashboard

• “Real-time server statistics” on HTTP Admin page

!13


DB2 and SQL

!14


Specs

• OS matters: Newer OS releases speed up complex DB2 queries with a smarter query optimizer

• IBM i is built to scale as large as its resources ‣ Example: IBM i’s query optimizer will choose the best

plan it can, given the amount of memory available to it. More memory, a more intelligent plan.

!15


DB2 query optimization

• IBM i has great tools • You can find many articles written by DB2 experts

• I’ll share a couple of favorites ‣ Index Advisor ‣ SQL Plan Cache

• See IBM’s book • IBM i Database Performance and Query Optimization • http://pic.dhe.ibm.com/infocenter/iseries/v7r1m0/topic/rzajq/

rzajq.pdf

!16


Index Advisor

• Now in web-based Navigator and Access Client Solutions • Recommends indexes across all queries

!17


SQL Plan Cache

• Lets you see what queries are REALLY running, who’s running them, and how long they take

!18


Persistent db connections

!19


Persistent connections

•Create a pool of database jobs •Known as a “connection pool” •You will connect quickly because a job is waiting for you •DB2 will...

• Choose a job each time you run a query • Create new jobs to handle high workload

•The word “persistent” may be misleading •No guarantee that a PHP script connects to same job each

time •Cannot rely on maintaining state (QTEMP, library lists)

between requests !20


db2_pconnect() to connect persistently

•resource db2_pconnect ( string $database , string $username , string $password [, array $options ] )

• Persistent is much faster than non-persistent § db2_pconnect can reuse connections, reducing the time needed

to connect (after the first time) to almost zero

• How db2_pconnect() reuses connections § Connections defined by database, username, and password § Tries to reuse an existing connection matching these 3 params § db2_close() on a persistent connection does nothing § db2_pclose() forces the conn to close

!21


Rules for using persistence

• Because connections are shared when defined with an identical database, user, and password, please: § Avoid unpredictable results by also specifying the same $options

for these connections § Promote sharing of jobs by minimizing the number of user

profiles that you connect with • Each user profile creates a new set of database jobs • Each set of jobs consumes system resources when they start • Though jobs don’t do much harm once they are started

!22


Persistence = connection pool

• On IBM i, job initialization is relatively slow § Gives us good things—auditing, logging, security § Solution: a pool of pre-initialized jobs

• Pre-started DB2 jobs save time § Generally run in subsystem QSYSWRK, job name QSQSRVR § These prestart jobs can be configured with CHGPJE command

• In PHP, persistent connections reuse initialized jobs § db2_pconnect() § Dramatic speed boost

!23


db2_pconnect() example with library list

$database = 'MYDB';

$user = 'MYUSER'; $password = 'MYPASS';

// set library list (works the same for connect or pconnect)

$options = array('i5_naming' => DB2_I5_NAMING_ON, 'i5_libl' => 'MYLIB1 MYLIB2'

); $conn = db2_pconnect($database, $user, $password, $options);

if ($conn) {

echo "Connection succeeded."; } else { echo "Connection failed.";

} // MYTABLE will be found, if in library MYLIB1 or MYLIB2

$stmt=db2_exec($conn,"SELECT * FROM MYTABLE");

!24


Rules for using persistence

• Because connections are shared when defined with an identical database, user, and password, please: § Avoid unpredictable results by also specifying the same $options

for these connections § Promote sharing of jobs by minimizing the number of user

profiles that you connect with • Each user profile creates a new set of database jobs • Each set of jobs consumes system resources

• More information: § “DB2 and PHP Best Practices on IBM i” at http://

seidengroup.com/presentations

!25


More about db2_connect, db2_pconnect

•Manual pages § http://www.php.net/manual/en/function.db2-

connect.php § http://www.php.net/manual/en/function.db2-

pconnect.php § http://www.php.net/manual/en/features.persistent-

connections.php

!26


DB2 server prestart job configuration

• Prestart jobs named QSQSRVR run in QSYSWRK • Or, if remote DRDA, QRWTSRVR in QUSRWRK • Configurable pool of jobs CHGPJE SBSD(QSYS/QSYSWRK) PGM(QSYS/QSQSRVR)

STRJOBS(*YES) INLJOBS(xx) THRESHOLD(xx) ADLJOBS(xx) MAXUSE(xx or *NOMAX)

• Defaults are somewhat low ‣ Initial jobs = 5, threshold = 2, adljobs = 2. Maxuse = 200

(*NOMAX may be better)

• More information about QSQSRVR prestart jobs ‣ http://www.mcpressonline.com/tips-techniques/database/techtip-

grab-control-of-the-db2-qsqsrvr-jobs.html

!27


If you increase the prestart jobs...

• Consider increasing FastCGI and Apache settings mentioned in Session 1

‣ FastCGI PHP children ‣ Look in /www/zendphp7/conf/fastcgi.conf ‣ Default: SetEnv=“PHP_FCGI_CHILDREN=10" ‣ Increase from 10 to # of expected concurrent PHP requests ‣ Each “child” needs temp storage, CPU, memory; find a balance

‣ Next, Apache’s ThreadsPerChild ‣ /www/zendphp7/conf/httpd.conf ‣ Default: ThreadsPerChild 40 ‣ Increase to number of expected HTTP connections

!28


PHP Toolkit

!29


Call RPG/COBOL: adapt 5250 logic to the web

!30


Toolkit Architecture

!31

RPG, COBOL, System Resources

XMLSERVICE

Your PHP code

Toolkit

Toolkit Methods

Sends XML


PHP IBM i toolkit is open source

• Ships with Zend Server 5.6+ ‣ Many improvements since first release

• Where to start ‣ https://www.seidengroup.com/toolkit/

• Zend Forum ‣ http://forums.zend.com/viewtopic.php?f=113&t=41648 ‣ “PHP on IBM i”->”Zend Server for IBM i”->”New Toolkit” ‣ Many tips there

!32


Performance “don’t”s

• Don’t use disconnect() unless it’s the end of the day, logout, etc. ‣ Reconnection in stateful mode is slow ‣ Official samples set a bad example here

• Don’t allow inquiry messages to arise, hanging your job ‣ MONMSG in CL is one way to address

• Don’t run commands or programs with warning CPF ‣ Example: ADDLIBLE can return “already exists” error. Prefer CHGLIBL or

other technique ‣ Toolkit will fetch joblog for error reporting, which can be slow ‣ Not XMLSERVICE’s fault…in future we may allow “fast” (no joblog) mode

sometimes for non-critical errors

!33


Performance “do”s

• Don’t use disconnect() unless it’s the end of the day, logout, etc. ‣ Reconnection in stateful mode is slow ‣ Official samples set a bad example here

• Don’t allow inquiry messages to arise, hanging your job ‣ MONMSG in CL is one way to address

• Don’t run commands or programs with warning CPF ‣ Example: ADDLIBLE can return “already exists” error. Prefer CHGLIBL or

other technique ‣ Toolkit will fetch joblog for error reporting, which can be slow ‣ Not XMLSERVICE’s fault…in future we may allow “fast” (no joblog) mode

sometimes for non-critical errors

!34


Share existing DB2 conn with toolkit

Either db2_pconnect or regular db2_connect can be used. Share a connection instead of creating a new one in the toolkit.

require_once('ToolkitService.php');

// connect to db using chosen i5 naming mode

$namingMode = DB2_I5_NAMING_ON; // ON or OFF // connect to db

$db = db2_connect('DBNAME', 'user', 'pass', array('i5_naming' => $namingMode));

// Connect to toolkit, passing db resource and i5 naming mode instead of dbname/user/password.

$conn = ToolkitService::getInstance($db, $namingMode);

!35


Avoid RTV* commands if need speed

• RTVJOBA, RTVUSRPRF, etc. ‣ Handy for diagnostics, and they work fine ‣ Inside the toolkit they use REXX, which makes them slow ‣ Avoid in production environments that need to scale ‣ Alternative: compile RTV* command into a CL, then call the CL from

toolkit

// RTVJOBA works but is relatively slow require_once('ToolkitService.php');

$cmdString = 'RTVJOBA JOB(?) USER(?) NBR(?) CURUSER(?) SYSLIBL(?) CURLIB(?) USRLIBL(?) LANGID(?) CNTRYID(?) CCSID(?N) DFTCCSID(?N)'; // Send the command; get output array of key/value pairs. Example: CURUSER=>FRED, ...

$outputArray = $this->ClCommandWithOutput($cmdString);

!36


Do not use disconnect()

• The disconnect() method ends the toolkit job • When you connect again it will be slow to restart • Better to leave the job running. It will not use much resources

when idle • If you really want the toolkit job to end, use “idle timeout,”

where the job ends after a period of inactivity

require_once('ToolkitService.php'); $conn = ..... connect....

$conn->setOptions(‘internalKey’=>’/tmp/alan123’)); // end job after an hour of inactivity

$conn->setOptions(array(‘idleTimeout’=>3600));

!37


Or use stateless mode

• In stateless mode, no jobs are created besides database jobs • It’s the simplest way to connect to the toolkit • Use a persistent connection (parameter 5 = true) for speed

‣ You can also pass in an existing persistent connection (shown a few slides back)

require_once('ToolkitService.php'); $conn = ToolkitService::getInstance(‘*LOCAL’, ‘ALAN’, ‘ALANPW’, ‘ibm_db2’, true); // true = persistent connection $conn->setOptions(‘stateless’=>true)); // stateless is simple

!38


Or use stateless mode

• In stateless mode, no jobs are created besides database jobs • It’s the simplest way to connect to the toolkit • Use a persistent connection (parameter 5 = true) for speed

‣ You can also pass in an existing persistent connection (shown a few slides back)

require_once('ToolkitService.php'); $conn = ToolkitService::getInstance(‘*LOCAL’, ‘ALAN’, ‘ALANPW’, ‘ibm_db2’, true); // true = persistent connection $conn->setOptions(‘stateless’=>true)); // stateless is simple

!39


Reduce IFS reads/writes

!40


IFS not as robust as DB2

• IBM i reverses the normal dynamic § Database (DB2) is fast; stream file access (IFS) can be slower

• Depends on workload, how many files accessed § Not as important as reducing HTTP access or other network

activity, for example

• Here are several tips...

!41


Tips to reduce IFS accesses

• Turn off or reduce logging to files in production ‣ Large log files can be a culprit on some systems

• Make files cacheable for long periods of time by setting “expires” headers (see Session 3 for more) ‣ https://httpd.apache.org/docs/current/mod/

mod_expires.html

• Use an opcode cache (see Optimizer+ discussion later in presentation)

• more on next slide

!42


More tips to reduce IFS accesses

• In the UNIX world, file access is called “stat”. Any tips for reducing IFS “stats” will be effective on IBM i

• Examples:

• In Apache, turn off checking for .htaccess files if you’re not using them, with “AllowOverride none” ‣ # disable htaccess checks <Directory /> AllowOverride none </Directory>

• Allow symbolic links ‣ Options +FollowSymLinks -SymLinksIfOwnerMatch ‣ Otherwise, Apache will make a separate call on each filename to

ensure it is not a symlink.

!43


Zend Server config for speed

!44


Zend Server for IBM i

• Download ‣ http://www.zend.com/en/products/server/downloads-ibmi

• Zend Server 9.x editions ‣ http://www.zend.com/en/products/server/editions-ibm-i ‣ Basic, Professional, Enterprise ‣ Same download, different license

• Professional and higher include the monitoring/tracing tools we’ll be discussing here

!45

http://www.zend.com/en/products/server/editions


Make the most of OpCache

• OpCache: “opcode cache” that speeds PHP execution ‣ The first time a PHP script runs, OpCache stores the compiled

version in memory, speeding future runs

‣ Works well with its default settings

• What’s configurable? ‣ Amount of memory to allocate to OpCache ‣ Number of scripts to keep in memory ‣ Timestamp-checking interval ‣ How often to check if a script has been changed

• Next slide: details on the timestamp-checking interval

!46


Timestamp checking by OpCache

• Timestamp validation frequency: how often OpCache will check if a script has been changed § Ensures you are not running stale code § Default: 2 seconds

• The default (2 secs) is usually OK • For max speed, disable timestamp validation via

admin GUI § Use in production only § Usually only a small improvement § But I’ve seen BIG improvement on systems with slow IFS access § Caution if you turn off timestamp validation:

• Changes to your app NOT applied till you clear optimizer cache • Good reason to keep timestamp validation on in development env.

!47


Turn off unused PHP extensions

• Extensions provide extra functionality to PHP § Each one uses memory

• Many are enabled by default § Default: ibm_db2 is on, SQL Server is off § Turn off those that you do not use § Do not turn off extensions needed by Zend Server’s GUI

• See my blog for exact instructions § http://www.alanseiden.com/2011/08/08/php-performance-tip-

disable-unused-extensions-with-zend-server/

!48


Cache it

!49


Two types of cache

• Zend Data Cache ‣ Save any type of data as key/value pairs ‣ Available in all editions of Zend Server ‣ Good for saving query results or any data that you want to reuse

• Zend Page Cache ‣ Cache a whole page (URL) ‣ Examples: home page (not

personalized), JSON content for state or country codes (thatdon’t change often)

!50


Zend Data Cache

• Save string or array data ‣ Similar to memcache

• APIs for shared memory and disk storage ‣ “shm” means shared memory. “disk” means disk ‣ zend_shm_cache_fetch(), zend_shm_cache_store() ‣ zend_disk_cache_fetch(), zend_disk_cache_store()

!51


Zend Data Cache

• How to use the APIs ‣ Pass a key name, an object or string, and the “time to live”

number (optional) • boolean zend_shm_cache_store(string [namespace::]key, mixed var [, int ttl])

• Example: cache district number for 300 seconds • $success = zend_shm_cache_store(‘district’, ‘12345’, 300);

‣ Fetch with just the key name • mixed zend_shm_cache_fetch(string [namespace::]key)

• Example: retrieve district number • $district = zend_shm_cache_fetch(‘district’); // false if fails to fetch

• White paper with examples ‣ http://static.zend.com/topics/Zend-Server-Data-Caching-

Whitepaper-0106-T-WP-R1-EN.pdf

!52

http://static.zend.com/topics/Zend-Server-Data-Caching-Whitepaper-0106-T-WP-R1-EN.pdf

http://static.zend.com/topics/Zend-Server-Data-Caching-Whitepaper-0106-T-WP-R1-EN.pdf


Zend Page Cache

• A premium feature of Zend Server • Set up URLs to cache via Zend Server’s GUI • Match exact URL or regular expression

!53


Reduce Session Locking

!54


PHP session files remember info

• Can keep track of user who is logged in, for example ‣ Cookie in browser tells PHP which session file to use ‣ session_start() initiates session in PHP

• Opens and locks the file for updating ‣ Located in /tmp by default ‣ Access data via $_SESSION

array

!55


AJAX requests burden sessions

• With AJAX, one page is > 1 request • Multiple requests (AJAX/JSON) in one page • If PHP sessions used (session_start), all requests from one

browser use same PHP session on back-end

!56

• Numerous PHP requests launched in one page. Note the “waterfall” shape; each waits for the previous to finish


Load test shows this session is slow

!57


View from performance data investigator

!58


DEQW everywhere

!59


Locking found

!60


Solution to AJAX session locking

!61

• As soon as your application finishes writing data to the session, close it • session_write_close() ends the lock

• You can still read $_SESSION array

• Comparison on next slide


Without, and with, session_write_close()

!62

WITHOUT session_write_close()

WITH IT Parallel run, faster! e.g. 834ms instead of 1.84ms


Go asynchronous

!63


Premium features that can help

• Zend Job Queue offloads long-running processes to run asynchronously. Examples: § Sending an email message § Crunching large datasets § Sending data to a web service on another server § Creating PDF files

!64


Job Queue is part of Zend Server

• Comes with Professional Edition and up • Web-based • See job status in Zend Server GUI • From the documentation: “Long-running report generation, order

processing, database cleanup, and pulling of RSS feeds are some examples of jobs that can be executed asynchronously. Zend Server for IBM i incorporates a job queue to provide full support for creating, executing and managing jobs to optimize application performance and reduce server load, minimizing application bottlenecks and improving the end-user experience. “

!65


Special abilities of Zend Job Queue

• Move tasks into a separate execution queue § Off-load to new process, later time, or different server

• Execute certain tasks at a specified time § Schedule, optionally repeat at intervals

• Use complex parameters

• Do it from a PHP API or Zend Server GUI

• Launch faster than PHP-CLI

!66


Creating Jobs

• Use createHttpJob() method

• Pass parameters § Simple: via query string $_GET § Complex: as an associative array of key => value pairs

• Set job options § Name, priority, schedule, etc. § Create deferred or recurring jobs

• Launch right from PHP or use the Zend Server UI

$queue = new ZendJobQueue(); $queue->createHttpJob('http://backend.local/jobs/

somejob.php');

!67


More about Zend Job Queue

Specs: http://files.zend.com/help/Zend-Server/jobqueue-global-api.html Tutorial: http://files.zend.com/help/Zend-Server/working_with_jobs.htm

API signature for createHttpJob: int ZendJobQueue::createHttpJob ( string $url, array $vars, mixed $options)

Example with nested array variables: $queue = new ZendJobQueue(); $queue->createHttpJob( ‘http://yourIBMi.com/send_alert.php’, array(‘from’=> ‘[email protected]’, ‘to’=>array(‘[email protected]’, ‘[email protected]’)), array(‘persistent’=>false) );

!68


Job Queue Management

View job status and manage execution

!69


Other Job Capabilities

l Managing priorities l Job dependencies l Querying for jobs l Checking job status and queue statistics l Suspending and resuming recurring jobs & queues l Passing custom HTTP headers l Management of jobs

l Handling failures and controlling retries l Code tracing, monitor

!70


Job Queue Tip

l If task will run more than 60 seconds, increase the FastCGI request timeout

l /www/zendphp7/conf/fastcgi.conf ‣ Default: RequestTimeout="60"

• Otherwise you’ll get HTTP 408 errors and unpredictable results in the Job Queue GUI

• New in Zend Server 2018: manage long-running CLI jobs via Job Queue: https://www.itjungle.com/2018/09/10/zend-server-2018-brings-php-7-2-to-ibm-i/

!71


That’s part 2. Coming up next...

!72

Session 3: Ensure a fast user experience


Contact and tips

Alan Seiden Seiden Group Ho-Ho-Kus, NJ

!73

[email protected] ● 201-447-2437 ● twitter: @alanseiden

Free newsletter: http://seidengroup.com/tips

CIO Summit information:[email protected]

mailto:[email protected]?subject=

http://seidengroup.com/tips

PHP Performance on IBM i - Seiden Group slides/PHP on IBM i... · PHP Performance on IBM i ... • Recommends indexes across all queries!17. Alan Seiden Consulting Let your PHP apps

Documents