Are you suggesting that maybe users (including you and me) should have to put up with painfully slow, stuttering software? I don’t see why his claim requires any justification - to my mind, it should be just as self evident as the fact that inflicting physical pain upon others should be avoided.
> […] modern software developeres tend to aim to be "fast enough" in the way that an structural engineer would choose "strong enough."
But they don’t aim to make software fast enough. My experience last week: Windows 10 file Explorer took ~2.5 seconds to open. Close it and re-open it: same thing. Open it, right mouse click another folder to open a new Explorer window: same thing. Fresh install of Windows 10 on a top of the line workstation-type laptop with 64gb of RAM.
Not all, but a frustratingly large proportion of modern software is dog shit slow.
The reason doesn’t need to spelled out in the article: if the developers of these slow apps cared, they could figure it out. From my experience, it usually looks something like this:
Some developer has an emotional attachment to Protocol Buffers, and will stop at nothing to see its adoption within the org. But their software is pretty heavily invested in JSON. So they rewrite the software to read the existing JSON files from disk (or REST web service response bodies, whatever), and reserialize them to protobuf in memory. Tada! Now we’re using protobufs, great. Of course, nothing meaningful was actually achieved here - they already had a perfectly fine, ready to use, deserialized, in-memory data structure before they added protobuf to the mix. Oh, and that plain struct in memory was faster than traversing a protobuf: the former had small substructures laid out in the same allocation as the parent, whereas substructures in protobuf involves multiple allocations and chasing pointers. Next step: realize that REST is lame, and gRPC is hip. But it’ll be practically impossible to rewrite everything from REST to gRPC, so they do the only reasonable thing: create proxies that sit between the client and server that translate REST requests/responses to/from gRPC! Now that we have that in place, we can add an additional proxy to the mix: Envoy. Envoy is a super popular layer 7 proxy, so it’s gotta be good. What functionality will be used? Any load balancing? RBAC policies? TLS termination? Nope. None of it. But because Envoy is “good”, adding it to the stack with no justification must also be intrinsically “good”, right? Right!
(Edit: do you see how long winded and boring this example is? This is precisely why the author shouldn’t expound on the “why” - anyone involved in the development of needlessly slow software (who isn’t blind to the problem because they are part of it) can recount similar craziness. Adding this to their blog would make for boring reading, and distract from the aim of their article.)
Buzzword/resume driven development, unwarranted layers of indirection (for no gain), absolution of responsibility via appeal to authority (if the top 10 software companies created and/or use some software, then surely we can blindly use that software too and enjoy the same success, despite not giving any consideration to whether it’s even remotely the right tool for the problem at hand), cargo culting, etc.
The reason software is painfully slow usually boils down to lack of critical, rational thought: either out of laziness and/or deferral of responsibility, or because of some emotional attachment to some type of software component.