What good performance looks like–Good to Great

Lots of clients at the moment are getting rid of their proprietary CPU architecture.  This comes in the form of RISC type implementations and moving to commodity x86 architecture.  There are a lot of advantages in this, but the primary seems to be the strategic of enabling an easier cloud migration when the time is right. 

I’m assisting with a number of very large platform migrations at the moment – moving from AS/400 to cloud or commodity.  Generally if people are moving off a 400 today, they have been on that platform for a long time..  As I doubt that ANYONE would buy JDE at the moment and get an AS/400 to run it on.  In fact, I doubt that has occurred in the last 8 years (am I wrong – tell me).

So, we are generally migrating 10-20 years of history and customisation to another platform.  It’s rarely JDE that is the problem in this type of migration, it’s all of the stuff that sits on the side of JDE.  The integrations, CL, RPG and custom SQL statements and triggers that make a migration tricky.

There is one more thing that makes this tricky – PERFORMANCE!  Never underestimate the amazing ability that an AS/400 has to process inefficient code well!  It is awesome at masking bad code by monstoring the job with great I/O, reactive (and somewhat invisible tuning) and very quick CPUs.

I quite often need to spend a lot of time tuning the workload (especially custom code) to get the new platforms to behave like the old – and to be honest, sometimes it will not happen…  A massive tablescan based UBE might just take longer on two-tier architecture and single tier AS/400 – but it’s the exception not the rule.

In general large SQL will run faster on new hardware – but it’s the transfer and processing of large datasets that can be problematic.

Look at the graph below.  This shows a client that has recently done a platform migration to Oracle database appliance (X7HA).  This is really smashing the workload, processing Trillions of I/O’s in the first week – yes Trillions!!! 

You can see a pretty cool and consistent graph below of page load times in JDE vs. activity.  The fusion5 ERP analytics suite allows insights like this.  We can see that the interactive performance actually improves when the site gets loaded up.  Makes sense to me.  Better cache is loaded and the users get a better experience.  What does interest me is that 10am when the users are at their most, we have page response time of about .45 seconds – which is amazing (I know, I have over 40 clients to compare with).

It’s really cool to be able to give clients these REAL insights into performance of their new platform and give them unequivocal empirical evidence that they’ve done the right thing and that their users are getting an exceptional interactive experience from the new hardware.

We are also able to drill down into some very detailed numbers on where performance problems might be – slowest screens, apps, regions, servers or users.