This is my first attempt at IIS connector tuning for ColdFusion 11 (and 10). Most of my career has been spent developing ColdFusion code and is now focusing more on server related activities. Plus CF 10 and 11 were slow to be implemented by our customers.

It seems that most information out there for connector tuning is based around one sole blog post: http://blogs.coldfusion.com/post.cfm/coldfusion-11-iis-connector-tuning (and CF 10’s version).

My post focuses on a three instance approach, using ColdFusion 11 Enterprise, with individual site connectors (as opposed to “all IIS sites”).

The basic concept as I understand it is to set the connection_pool_size to 500 and monitor the site. Add up this number in each site, using the same instance, and use that value for maxThreads for the AJP connector setting. Then gradually increase that value by 100 under load testing conditions until stable. Then give that number some wiggle room for future growth. After that number is set, then set the max_reuse_connections and connection_pool_timeout.

So lets say, as an example, that I use the “All IIS Sites” option for my connector instead of individual connectors. If I use the recommended connection_pool_size of 500 for each site, I’d use 3000. Then based upon the equation of connection_pool_size / # of sites, I’d set “max_reuse_connections = 500”. Example:

worker.list=instance1
worker.instance1.type=ajp13
worker.instance1.host=localhost
worker.instance1.max_reuse_connections=500
worker.instance1.connection_pool_size=3000
worker.instance1.connection_pool_timeout=60

<Connector port="8013" protocol="AJP/1.3" redirectPort="8446" tomcatAuthentication="false" maxThreads="3000" connectionTimeout ="60000"/>

Now, when I looked up the workers.properties specs for Tomcat I found that max_reuse_connections is not a standard property. I’m assuming this is one of the customizations made by Adobe. Based upon how the value of this property is a division of the number of sites, that this property is per site. Therefore in conclusion, I have up to 500 connections to reuse for each site in my total pool of 3000.

Now, lets say we’re using individual connectors. Each of the six workers.properties would look like this based upon Adobe’s blog:

worker.list=instance1
worker.instance1.type=ajp13
worker.instance1.host=localhost
worker.instance1.max_reuse_connections=83
worker.instance1.connection_pool_size=500
worker.instance1.connection_pool_timeout=60

<Connector port="8013" protocol="AJP/1.3" redirectPort="8446" tomcatAuthentication="false" maxThreads="3000" connectionTimeout ="60000"/>

So, as per Adobe’s blog, connection_pool_size / # of sites rounds down to 83, instead of 500. 6 sites X 500 connection_pool_size = 3000, which is reflected in the instance’s server.xml file.

In the end, the instance still allows for 3,000 connections; 500 coming from each site.

Question: Why am I using the same calculation for max_reuse_connections when combining all sites or connecting them individually? Shouldn’t I be able to use up to the value of each connection_pool_size for each connector? If max_reuse_connections is for each site, shouldn’t that number be the same no matter individual or “All IIS” connector types?

For example:

worker.list=instance1
worker.instance1.type=ajp13
worker.instance1.host=localhost
worker.instance1.max_reuse_connections=500
worker.instance1.connection_pool_size=500
worker.instance1.connection_pool_timeout=60

<Connector port="8013" protocol="AJP/1.3" redirectPort="8446" tomcatAuthentication="false" maxThreads="3000" connectionTimeout ="60000"/>

At one time I was a fan of ColdFusion Builder 2.  However as I used it more and more, it just wasn’t mature enough and caused me frustrations. These where mostly attributed to it being built on top of Eclipse.

For the past few months now, I’ve heard so many people rave how good the new ColdFusion Builder 3 has become and that I should try it out again.

For reference, I’ve been using Sublime Builder 2 and 3 for a little over a year. Also, the project that needs this the most has the LESS source files and a large code base in separate SVN folders. I know, the SVN structure is a little screwy, but it fits the project good.

So for my first tackle: LESS editing and compiling.

Sublime Text

I have the package manager installed.

To get LESS editing working I install the LESS package and I’m done. It works.

To get LESS compiling I installed the Less2Css package, added to the project config my main LESS file and the output directory values.

Now everytime I save a LESS file, it compiles the main LESS files and dumps it into the correct directory.

This took me about 1/2 hour to research, install and test.

ColdFusion Builder 3

I have the Eclipse Marketplace installed.

To get LESS edit working, I installed the Eclipse plugin for LESS located at http://www.normalesup.org/~simonet/soft/ow/eclipse-less.en.html. This was pretty straight forward and seems to work with a simple test or two.

Now is where the nightmare starts.

The default LESS Compiler is “lessc”. However, even though it is “lessc” that we want, it just won’t fire, saying it can’t find the file. I then changed it up to “lessc.cmd” after fumbling around for awhile trying to figure it out. This works, but I can’t get it to save into the specific folder I need it in, which is another project’s CSS folder. Instead, it just wants to save in the same directory. I fiddled around with the settings, but no go. I even tried adding a linked resource to both the LESS source folder from the main project and to the main source file folder from the LESS project. I submitted a issue ticket for help. Their documentation is quite lacking.

I ended up giving up trying to compile to the right directory using the plugin.

I looked at using “Crunch!” that a co-worker uses for the same client project. While this would work, it’s kind of clunky and is run outside of Eclipse. Which means opening another program, using more resources, and having to refresh the CSS folder inside Eclipse after it compiled. I abandoned this quite quickly.

I then installed “WinLess” which watches the directory and compiles upon changes. But that didn’t let me save to a specific directory, so I tossed that.

I then fiddled with creating an External Eclipse Tool. I don’t remember exactly what happened, but got frustrated and abandoned that too.

I finally found a solution that works pretty good. I created a new “Builder”. Now everytime I save a file, a command-line command is run and the LESS file is compiled into the correct folder and automatically refreshes the CSS folder in the Eclipse project.

I found out how to do this at http://stackoverflow.com/questions/7686281/unnoticeable-lesscss-compiler-in-eclipse-pdt under the “Manually installing a Builder” answer.

I run a .CMD file as such

CD C:\Users\[user]\node_modules\.bin
node ..\less\bin\lessc "C:\website\source\less\my1.less" "C:\website\css\my1.css"
node ..\less\bin\lessc "C:\website\source\less\my2.less" "C:\website\css\my2.css"

I am currently working on a branch, which is created from trunk using SVN. I am using TortoiseSVN 1.8.6 as the client.

The directory structure looks like this:

web root
|- (many dirs)
|- tags
||- script.cfm

I noticed that after my last merge from trunk, that the script.cfm file was reverted to old code. I’m pretty sure it was brought down from trunk last week right after it was updated in trunk. I also know it was the same code as trunk because it worked. (it resolved an error)

When I ran a merge against the web root again to see what was going it, I saw it was merely ignoring any sign that something was going on with the “tags” directory.

If I where to try and just merge the “tags” directory or the “script.cfm” file directly, it would give me a tree conflict stating that the current directory was obstructed. Now the whole directory is reverted and nothing had it open. I even ran a clean up.

After some messing around, I noticed another error that made me believe it was trying to handle the directory as an actual tag folder.

To resolve this we renamed the “tags” folder to “cftags” on trunk. But it seems that Tortoise wants nothing of it (crashed), so I’m recreating the branch from trunk and merging in all my changes manually, loosing all my check-in history.

web root
|- (many dirs)
|- cftags
||- script.cfm