Mechanical Sympathy in Compute
Is there an equivalent to mechanical sympathy in the world of compute?
Not too long ago I, as I was putting the wheel bolts back in having swapped to winter wheels onto the family wagon, I started wondering if there’s an analogy to mechanical sympathy in the world of compute.
As anyone that is adept at a bit of spannering will tell you, when you put the wheel bolts back in, you do them up by hand first. Now why the heck would you do that when you have a lovely impact wrench that zipped them off in 0.3 seconds flat sat next to you? Just put them back in with the impact right? Wrong. Because mechanical sympathy.
If wheel bolt is cross threaded when you put it back in by hand, you’ll feel it. You correct the mistake and move on. No harm done. Zip them back in with an impact wrench though and you’ve got a cross threaded bolt torqued down to a billion Nm. Game over. If you make this mistake, you’ll probably only ever make it once.
This got me thinking, are there parallels in the world of HPC, or even just computing more generally, where we do things slower, more deliberately, because experience teaches us that its better?
Avoiding denial of service attacks to the rest of your enterprise infrastructure springs to mind. Hands up who’s pushed a change that resulted in every compute node attempting to connect to some poor single instance service running on the end of an ancient 100Mbs network connection 😆 Go on, admit it, you’ve been there.
I’m sure at a micro level, these are things that good software developers think about all time. Question is, what happens when you get an AI to write the code for you instead? When you’re using the equivalent of an impact wrench to churn out code more quickly? What happens then?
Anyway, it’s Friday, fancy sharing your examples of mechanical sympathy in compute? Or better still examples of when you DDoS’ed something with your HPC cluster?
