Configuring and tuning IBM HTTP Server (IHS)

This article explains some of the strategies for configuring and tuning IHS.

Introduction

IHS is based on the Apache HTTP Server. In the CLM standard topologies, IHS also functions as reverse proxy server. A reverse proxy server provides a layer of indirection for backend application servers, and can provide load balancing at the http server tier to other application servers.

This article introduces some monitoring and tuning techniques. Generally, tuning is not always necessary as the default IHS settings are designed to handle reasonable load. However symptoms of poor performance include IHS crashing and CPU thrashing.

The IHS directives and what they mean

The tunable IHS directives live in the httpd.conf in the \config directory. Always backup the httpd.conf file before making a change (rename it to httpd.conf.bak for example). After making changes, you will need to restart IHS for changes to take effect. All numeric values must be positive, non-zero integers.

Multi-Processing Module (aka MPM) directives

MaxClients is a key factor that defines how much load IHS can serve. MaxClients is not the same as maximum number of users served. MaxClients is a soft limit to control the simultaneous connections or threads. Changing this value doesn't mean IHS will deny new or additional user requests. Requests over this value will be put in queue until there are available threads.

• MaxClients is recommended to be set lower than 2000 for a single IHS server. Monitor the number of bsy threads (see "section IHS error logs" below) to determine the peak of the simutanous connections based on your typical load, and then add 25% as a secure factor to set the value of MaxClients.

• Note that on Windows there is no MaxClients directive and that ThreadsPerChild is used instead.

ServerLimit, ThreadsPerChild, ThreadsLimit along with MaxClients affect the capability of IHS in handling loads. To tune these, start with ThreadsPerChild and consider platform, plugin, SSL and caching. A higher ThreadsPerChild can benefit the WebSphere plugin, but the trade-off compensates CPU for SSL.

• ThreadsPerChild default is 25.

• ThreadsLimit should equal ThreadsPerChild.

• ServerLimit should be equal to MaxClients divided by ThreadsPerChild rounded to nearest integer. Ideally, change MaxClients so that it equals ServerLimit multiplied by ThreadsPerChild without any fractions. If ServerLimit multiplied by 25 is greater than the amount of available RAM, lower MaxClients and readjust ServerLimit.

StartServers, MaxSpareThreads, MinSpareThreads determine how IHS reacts when load varies.

• MinSpareThreads should not exceed the value of 25 or 10% of MaxClients.

• MaxSpareThreads can be increased in step with MaxClients.

• StartServers is best to leave at default of 1.

MaxRequestsPerChild can remain at the default of 0.

Other considerations

IHS server settings

System resources

On a linux system be sure that there are enough user processes allowed. You may need to increase the max user processes setting using this command:

ulimit -u 65536

Also decrease the stack size. 512 MB is sufficient for CLM apps:

#add this line into the envvar file under folder /conf 
ulimit -s 512

Tools and tips

IHS error logs

Ordinarilly, IHS logging is enabled by default. In test systems, you may wish to disable the access_log to save disk space. To do this comment out the following in the httpd.conf file

CustomLog logs/access_log common

However, for production environment, consider the options for effectively managing log files http://httpd.apache.org/docs/2.2/logs.html.

IHS error logs live in the /logs directory. The frequency they are written to is controlled by the mpmstats module which is automatically enabled in IHS version 8.x. To obtain more accurate statistics, set the ReportInterval to a small value. For instance, set it to 60 (seconds).

Sample output of mpmstats:

[Mon Jun 04 14:32:00 2012] [notice] mpmstats: rdy 88 bsy 187 rd 0 wr 5 ka 160 log 0 dns 0 cls 22

In the above example bsy 187 indicates the busy threads and should be less than the MaxClients value.

Monitoring the log files closely by watching for [alert], [warn] and [error] messages can give you hints on what goes wrong, for example:

[Thu May 10 15:10:34 2012] [alert] (11)Resource temporarily unavailable: apr_thread_create: unable to create worker thread

The above message shows that IHS hits a resource limit and cannot create more threads. If the MPM modules are configured correctly (see section "The IHS directives and what they mean" above), it may be often remedied by reducing MaxClients or increasing the maximum user processes on the system. This link provides handful information for IHS trouble-shooting: http://publib.boulder.ibm.com/httpserv/ihsdiag/errorlog.html

Server status

Runtime monitoring is possible through the IHS server-status page, located at http://your_server_name/server-status

Enable the IHS server-status page by adding an allow directive to the corresponding section of the httpd.conf.

Apache.org makes a sample page available at http://www.apache.org/server-status.

Useful commands on Linux

Watch open IHS processes with this command:

ps -ylC httpd

Sort by process size:

ps -ylC httpd --sort:rss

Sort by parent process:

ps -ylC httpd --sort:ppid

Check an individual process with child threads

pstree -cp <pid>

Related topics: Deployment web home

External links:

• IBM article Using IBM HTTP Server 6.1 and later diagnostic capabilities with WebSphere
• IBM article on tuning IHS: IHS performance tuning
• IBM technote: Tuning IHS to maximize the number of client connections to WebSphere Application Server
• IBM IHS Q&A: IHS Questions and Answers
• IBM Infocenter IHS v8 Infocenter
• Apache.org article: Apache tuning
• IBM developerWorks article: Proxy server versus the HTTP plug-in: Choosing the best WebSphere Application Server workload management option

Additional contributors: None