Small change, Big difference!

Performance testing can save you a lot of time ;-) and money. While load testing the new site of Selor (recruitment agency/testing agency for the Belgium Government), We, the team, found that a lot of time was being wasted, in seemingly harmless code. It was only a few milliseconds, but in many places.

We came up with few hypotheses.

• the new virus scanner, who caused some problems in the past few weeks.
• the network disc.

After trying the first option, we saw a minor improvement.
But when hacking the application to use local disc, instead of a network drive, we had a factor 3 for normal response times. Not a bad start.

Result 

We asked for a new rolled out system, and made sure we were using local discs, where possible, using the network disc only for our most critical parts.

We can now handle 5-to-7 times more traffic then before, without degrading performance.
The customer and the team were thrilled with excitement. Being fast is cool!

Challenges?

The application is a modern SOA application, meaning a lot of remoting is going on. Without AppDynamics, we would have lot more hypotheses and many wiled gauze chases. But now we could see, where the time was being spend. => Don’t guess, but measure change!

(not a real example!)

Where to start?

Why no try the free lite version from AppDynamics, for Java and .NET.

Performance Should Be Free

Performance Should Be FREE! Version 2!

Today I hope to add another free tool to your utility belt. To help you resolve your bottlenecks in production faster than before.

AppDynamics Lite version 2. This free profiler/monitor uses your “Business Logic” to visualize the performance problems.  Discover why your search- of checkout-“transaction” is slow.

Once you installed AppDynamics will automatically start detecting your business transactions. If something spikes your interest, you can right click the transaction and ask it to collect samples aka snapshot. These snapshots contain stack traces with method execution times. For example.

com.sample.domain.Cart.checkout(300 ms)

Also external calls, like webservices and SQL statements are collected.

New in version 2, is the trending of MBeans through JMX. This might not get you excited, but you can also add alerting, so receive a mail if a monitored value goes above certain base line.

Now you know if something is going wrong and you can take snapshots (manually) while it’s happening.

Pro’s

•          Easy to install
•           Easy to use
•           Easy to understand
•      Overhead < 2%

Con’s

•           It’s limited: Max data retention of 2 hours
•           Only 30 transactions
•           It’s not perfect. Use it as your first line, but also consider using other tools like JVisualVM & MAT.

Upgrade? Why?

-          Less work! Snapshots are taken atomically.

-          Jumping from tier to tier. Business transaction can run over multiple tiers, now you can follow them and find out, which tier is causing the problem.

…

Advice

Try it! It will help you find what you are looking about 80 % of the time.

When your application matures it’s about 90%. (Personal opinion)

Watch the movie!

Install:

http://www.youtube.com/watch?v=xSenA-zEKR0

New version:

http://www.youtube.com/watch?v=uZkHkVlOohM

Finding the root cause! Fast! With AppDynamics!

The thing I love most about Java are the awesome tools! The plugins for IDE's, the profilers (ex. YourKit) , the memory analyzers (ex. eclipse MAT), the javaagents (ex. JRebel)... But my absolute favorite of all  are the APM-tools(Application Performance Management). This really helps me in my day to day life.

In this series of tutorials we will cover a lot of functionality what a great APM-tool can do for you. I will be using my favorite APM on the market AppDynamics.

Why? It looks awesome and it does the job well. I can use in production without me becoming the bottleneck (< 2% overhead)!

Tutorial number one:

Finding the root cause! Fast!

Step 1: The application dashboard!

1: Here we are on our dashboard. We see our different tiers and nodes of our application.

2: There seems to be a problem with our response times. Better look into it!

3: Here we see the snapshots taken of the incident.

Step 2: A closer look

We have a lot snapshots of slow transactions. No stalls yet! Click on the camera...SMILE.

Step 3: Snapshot list

We see a list of all snapshots taken. We select a snapshot too see what's happening.

Step 4: Insight in the transaction

Node 2 tier 1 seems to be having some difficulty. We better drill down to find out what's going on.

Step 5: Drill Down

Root cause found: A Thread.sleep. It was only a test. Glad no one ever uses Thread.sleep in production...

Things you should know:

AppDynamics takes snapshots automatically when transactions are running slow.

So you don't have to worry about watching your application all of the time.

How does AppDynamics know my transaction is running slow?

Self learning baselines. Meaning you don't have to configure a thing to detect abnormal deviations. Also it becomes more accurate over time!

P.S.  In case you do not agree with the defaults, you can always configure them manually.

*Lite version(FREE): You will have to start a diagnostics session manually to take snapshots. (right click on the transaction => start diagnostic session).

That's all for today.

Next tutorial we'll talk about Business Transactions.

What are they and how can they help us?

How to define them or NOT! (auto-discovery)

Please look forward to my next blog and please give feedback!