Tip 6: Load test your interactive activity

You’ve got my full attention in this section, I really enjoy load testing.  Whether this is using OATS or other software, it’s a bit of a passion of mine.

Here is my recipe for success for interactive load testing.  You have your batch results above, so you are pretty confident with database size and hopefully application (logic layer).  We are now going to test the interactive performance of JD Edwards and how the user is going to experience things.

The first question you need to be honest about, is the peak capacity of users that you are going to test.  If Server Manager tells you 150 users are on, how many people would you load test with?  I can tell you – A LOT LESS!  I would test 40 in that scenario with a wait time of 5 – 8 seconds.  Let me show you why:

Figure 8: Standard ERP Analytics screen showing current activity, both throughput and location

My interactive report says there are 56 users logged into JDE and active in the last 5 minutes.  This is an interactive dashboard that Fusion5 ERP Analytics customers have access to.  You can also see the pages per minute and pages per second.  We are peaking at about 150 a minute in that snapshot, but I can find the peaks over the last 2 months if needed.

Figure 9: Server Manager’s view of the connected world is generally artificially high

Yet Server Manager is trying to tell me that I have 288 users.

Even with my classic double up the AIS users – we have 144 logged in, but only 58 active in the last minute.

What I’m trying to say here is don’t stress your system with too many users. Tuning for 3 x the worst scenario possible is actually going to slow you down.

Figure 10: ERP Analytics screen showing time of day, page views and performance. Reiterating the fact that a busy server is a fast server!

The graph above is unequivocal in showing that performance is better when pages are busy.  Do not have too many idle web servers because you have catered for 3x the users – your users are actually going to experience worse performance.  This is MORE dramatic at the users drop off.  I see around 20% performance improvement when a JAS server is loaded and cached up nicely.

Now that you can determine the number of users you need for load testing, you can execute this with the software or services that you have access to.  At Fusion5 we use OATS and can assist with any load testing you need.  We also validate and continually measure interactive performance using ERP Analytics, which can produce all of the graphs that you see above.

Anecdotally, good performance from JD Edwards is when pages load in about 1.1 seconds. 

Figure 11: Another view of performance over time, but separating download time and server response time. The page load is generally in direct correlation to server response time.

We measure and record exactly what the end user experiences.  We can also report on the network traverse time and the server response time.  These are all critical values when determining what you need to fix.  We can run this reporting on different users or geographies too, so you can compare performance in a single city or around the world.