Andrew Tanenbaum - MINIX 3: A Reliable and Secure Operating System - Codemotion Rome 2015

[@gblargg]

Recovering from software failures seems tricky. First off, hardware failures are generally due to physical degradation, whereas software failures are due to design defects. If you have a defect in software, how do you remove that from the system as you can do with hardware? You can't just have multiple copies of the same software, as the defect will be present in all and trigger under the exact same conditions as it first did. It seems like the software analog to physical fault-tolerance is having multiple pieces of software, each with a different design. Even then the protocol and commonality they communicate with could have a defect. Wish I'd come across this great talk sooner. Holding software to a higher standard is refreshing and not done anywhere near enough.

[@soyitiel]

Hey quick question, if anybody knows: if there's no disk driver at the starting point, how does the reencarnation server load the disk driver from disk to memory? Is there some sort of primitive disk driver embedded in the server? I'm sure I'll know the answer if I read the source code, but, since I'm here, I thought I'd ask

[@lolisamurai]

exactly my thoughts about bloatware and the need for lightweight software that works better

[@2DReanimation]

Fascinating! If only it could be done with no performance sacrifices! That's the dream. Then we could have a single operating system for the smallest to the biggest hardware. A single core that all OS developers can help improve. A micro-kernel that's perhaps not completely compatible with Linux, but with just recompiling Linux software for it, it would be entirely compatible. That'd be amazing. Of course, being 100% compatible with NetBSD will be great. But again, if it loses performance, it won't be worth it for most regular usage.

[@nextlifeonearth]

Look up the L4 kernel. It's a performance optimised microkernel design, significantly faster than MINIX.

[@2DReanimation]

Yes I found it shortly after. And it's amazing it's reached such high performance. I also learned that it can run Linux on top of it with little performance impact!

[@qwqqq2416]

@@nextlifeonearth Hello, I wonder if MINIX is dead, at all? And should I get this book: Operating Systems Design and Implementation The MINIX Book by Albert S. Woodhull and Andrew S. Tanenbaum. Or should I opt for something else? Thank you.

[@nextlifeonearth]

@@qwqqq2416 Haven't you heard? Minix is the most popular operating system, because of Intel. (It runs IME) Jokes aside, Minix isn't really meant as an OS you should "use". It's intended as a learning tool and is spectacular in that sense because of its simplicity. Simply to get the fundamentals of OS design in general, that book may well be a central part of your education. It shouldn't be your only source, and no singular source should be. Besides that there are other open source projects you can learn from. Like I took a liking to Redox-OS because of its emphasis on memory safety. Also a microkernel and a small code base (and it has a gui/toolchain) Then for more "professional" code you could follow kernel.org's mailing lists on driver development if you want to know how that works. I'd say, widen your scope before you're going to focus on one thing. Experiment a lot, that how you'll learn most.

[@qwqqq2416]

@@nextlifeonearth I think my question should have been "Is it alive, at all", but pardon my slip of thoughts. Thank you for the spirit in your answer, I really appreciate it. Do you recommend first reading the Dinosaurs book (Silberschatz and others) or Tanenbum's and Bos's? The former is in it's 10th edition, so I might favor it over the latter, but what you think?

[@AXDOOMER]

37:15 X11 - "The user kind of thing"

[@markteague8889]

AXDOOMER It’s obvious that Dr. Tanenbaum doesn’t think in terms of the user space on a routine basis. I’m sure he feels that user applications is something left to others working at that level of the system. He enjoys contemplating the symmetry, beauty and fault tolerance of the low level algorithms that manage system resources.

[@gblargg]

Minor quibble: 5:53 80% redundancy on a CD-ROM? I don't think that's correct. Doing a little research, 288 bytes of error-correction per 2336-byte sector. Each byte takes 14+3 bits on disc, but this isn't for error correction rather to maintain timing and avoid ambiguity in decoding, much like serial transmission must have start and stop bits for synchronization. Even if you mistreated all this as redundancy, that's about 59%.

[@MFSA8988]

So, more than half is redundancy.... I dont think anyone in that audience is going to care if he said 80% or 59%...

[@gblargg]

@@MFSA8988 I agree but some of us tuck away bits of information we receive from various sources, and it sounded outlandish. CD-ROMs aren't that unreliable.

[@dima357]

Well on my Win 10 I do not remember when it failed last time with blue screen of death, probably much more than a year ago. My 2 Android phones also work well, no need to restart them. So I am perfectly happy with hybrid/monolithic kernels. I am sure Minix enabled backdoor in my Intel PC works super reliable, so Big Brother can always spy on me.

[@herauthon]

Minix into space - that be a goal - send minix into space

[@herauthon]

Gui - is that important ?

[@herauthon]

LLVM memory table store needs a second store point - to secure pwrloss events ? e.g. sdcard, flashmem.. something big enough.

[@herauthon]

or not..

[@fabricobjects-llc3581]

Just imagine if Intel used Ubuntu inside all of their chips instead of Minix.

[@nextlifeonearth]

The chip would be twice as big and people would actually know what was running on their PC, because of GPL. I like the idea of the latter part.

[@АлександрПетров-г8о1н]

Чё никто субтитры не создаст?

[@Ensue85A]

What do you think of the Haiku OS project?

[@hrnekbezucha]

Calling serial and pwm exotic features made me laugh. If you want to be of any use in the embedded world, controlling the hardware peripherals is essential.

[@edino1981]

34:05 I don't know anything about making an OS but i can tell as well that BSD source code looks much better than Linux source.

[@dennishergaarden8521]

you know :D

[@mahkhi7154]

Linux is More Modular. In the Direction of Minix. However, Windows is More Reliable than Linux. Crashes Less Often.

[@adaai2384]

Job control stopped being relevant the day X Windows was released.

[@duhuanpeng1398]

no easy to understand... could you speak slower...:|

[@Frank_UK_was_taken]

Here I am watching this at 2x speed

[@mahkhi7154]

I ASSURE YOU KERNEL KESARIA (WHAT YOU KNOW AS: CTRL-ALT-DELETE), is Just as Reliable a MINIX. In the Past: Hackers Could Takeover: Semi-Kernel Windows, and Then Overwrite (CTRL-ALT-DELETE). Now Semi-Kernel Windows: Doesn't Know, Where CTRL-ALT-DELETE is, in the Memory Address Space. A Programmer Mistake, Wouldn't Kill CTRL-ALT-DELETE. A Hacker Would. Linux: A Programmer Mistake Does and Occasionally Does Freeze its: CTRL-ALT-DELETE. Because its Fatter. I ASSURE YOU KERNEL KESARIA (WHAT YOU KNOW AS: CTRL-ALT-DELETE), is Just as Reliable a MINIX.