How to Make Things Faster

Chapter 54. Yeti

In conference presentations, sometimes I ask, “How many of you have ever experienced a hardware upgrade where the system wasn’t as much faster after the upgrade as you had hoped?” Usually about 90% of my audience members raise their hand. Then I ask, “OK, how many of you have ever experienced a hardware upgrade where the system became objectively slower after the upgrade?” Usually only about 5% of people in the room raise their hand, and those hand-raises are usually timid. Even the people who’ve seen it aren’t sure whether they believe it. It’s the natural self-doubt inspired by a Yeti sighting.

But it’s true; it happens. Upgrades to faster hardware can make important programs go slower. And I can explain why. I can even tell you when to expect it to happen. There’s nothing supernatural about it.

When Jeff and I visited our Payroll client in Dallas, I had never actually seen evidence for an upgrade making an application slower. But that day we saw an irrefutable case of an upgrade making a high-priority program objectively slower. And we had the data. We had a big ol’ Yeti in a net. 

Remember, this was a client whose PYUGEN process was running so slowly that they couldn’t print paychecks on time.  After suffering for months, they finally bit the bullet and upgraded their CPUs, hoping it would fix PYUGEN. But, to their horror, PYUGEN was objectively slower after the upgrade.

When we arrived, we traced a PYUGEN execution for half an hour, resulting in this profile: ...