It’s statements like the above that make “software engineering” sound like an oxymoron. Let’s all ignore that the gold standard for life-critical systems is secure RTOS, most of them microkernels. No, it’s the sharded MangoDB clusters sustaining containerized Baboontoo images from the Crocker registry that are the apex of engineering.
Sort of weird that all of the proof that microkernels aren’t slow are from more than 20 years ago. Probably part of this is that linux has sucked a lot of the air out of the OS research room, but still a little odd.
It’s strange that he didn’t address any of the “Exterminate All Operating System Abstractions” work either. In general, abstraction is a tool for trading freedom for ease of use / productivity. It could be that the abstractions are so much more productive that you see better results with the same amount of effort, but you can almost certainly do better with something custom, given unlimited time.
I think the strongest argument for abstraction like what microkernels provide is that in a real world scenario, programmer time is one of the most expensive commodities, so it’s probably worth trading some freedom for time.
There’s steadily performance claims for microkernels many of which are in commercial production. OKL4 claimed to be in a billion phones with baseband isolation being a common use. Here’s a report on the L4 family:
Blackberry built the Playbook on QNX microkernel. It outperformed the iPad at the time in demos with two games playing simultaneously with seemingly no sluggishness from multitasking. That came from QNX’s design to reduce how much threads can interfere with each other.
This title reminds me of the Penn and Teller clip that would go with the sentiment: https://youtu.be/lb52oT1kA0Q?t=2m42s (note it is from a ‘Bullshit!’ episode on religion)
Worth reading for this gem alone:
Sort of weird that all of the proof that microkernels aren’t slow are from more than 20 years ago. Probably part of this is that linux has sucked a lot of the air out of the OS research room, but still a little odd.
It’s strange that he didn’t address any of the “Exterminate All Operating System Abstractions” work either. In general, abstraction is a tool for trading freedom for ease of use / productivity. It could be that the abstractions are so much more productive that you see better results with the same amount of effort, but you can almost certainly do better with something custom, given unlimited time.
I think the strongest argument for abstraction like what microkernels provide is that in a real world scenario, programmer time is one of the most expensive commodities, so it’s probably worth trading some freedom for time.
There’s steadily performance claims for microkernels many of which are in commercial production. OKL4 claimed to be in a billion phones with baseband isolation being a common use. Here’s a report on the L4 family:
https://www.researchgate.net/publication/262172784_From_L3_to_seL4_What_have_we_learnt_in_20_years_of_L4_microkernels
Blackberry built the Playbook on QNX microkernel. It outperformed the iPad at the time in demos with two games playing simultaneously with seemingly no sluggishness from multitasking. That came from QNX’s design to reduce how much threads can interfere with each other.
This title reminds me of the Penn and Teller clip that would go with the sentiment: https://youtu.be/lb52oT1kA0Q?t=2m42s (note it is from a ‘Bullshit!’ episode on religion)