The Hardware Knowledge That Every Programmer Should Know(needoneapp.medium.com)
needoneapp.medium.com
The Hardware Knowledge That Every Programmer Should Know
https://needoneapp.medium.com/the-hardware-knowledge-that-every-programmer-should-know-f62cf4ba8bdc
9 comments
Nice article, but it was probably heavily machine-translated with little human intervention. There is a message that says "Code Language: JavaScript" all over the place, but the code examples are actually in Rust, and the last (unnecessary) one is in C++.
To me, there is nothing that can beat CSAPP on this topic.
[1] R. E. Bryant and D. R. O’Hallaron. Computer Systems: A Programmer’s Perspective. Prentice Hall, 3rd edition, 2015.
It´s a shame that the last edition is from ten years ago, although to be fair I actually only read the 2nd and only skimmed the 3rd one. Still, would be nice to get an update using ARM.
[1] R. E. Bryant and D. R. O’Hallaron. Computer Systems: A Programmer’s Perspective. Prentice Hall, 3rd edition, 2015.
It´s a shame that the last edition is from ten years ago, although to be fair I actually only read the 2nd and only skimmed the 3rd one. Still, would be nice to get an update using ARM.
This is for me a basic knowledge for writing really-performant code. But I am surprised, that out-of-order execution isn't mentioned. And this approach is a big deal in modern CPUs.
Also important to know that if you're accessing disk, database, or network those are so slow that this stuff probably doesn't matter.
Keep in mind there _may_ be a negative feedback loop there.
If you're building your software in a way that won't be able to perform better with superior disk/db/network performance, then it isn't worthwhile to ever upgrade to a more performant disk/db/network.
If it is possible, make sure your software will actually be faster on a faster disk rather than just testing on a slow disk and thinking "well we're I/O bound anyway, so no need to improve perf".
If you're building your software in a way that won't be able to perform better with superior disk/db/network performance, then it isn't worthwhile to ever upgrade to a more performant disk/db/network.
If it is possible, make sure your software will actually be faster on a faster disk rather than just testing on a slow disk and thinking "well we're I/O bound anyway, so no need to improve perf".
This is a common fallacy. Yes, accessing disk or network is slow, but you still can perform CPU work while waiting on I/O and it may be beneficial to perform CPU work fast, in order to be able for example to serve a lot of network connections.
Example: Microsoft, modern websites.
``` Note: I’ve followed the rules you provided for translation. Let me know if there’s anything else I can help with! Here is the translation of the Markdown content from Simplified Chinese to English: ```