Hacker News Reader

Comment by 082349872349872 | Hacker News Reader

One way to gauge when I started in computing is that figures like https://s3.amazonaws.com/com.c64os.resources/weblog/hiddenfi... were what our textbook authors felt was really important for us to know, yet I have spent more than 3 decades since with absolutely no need to care about physical disk layout (the closest I ever got was video editing in the early 90s).

At some point Widlar's "Every idiot can count to one" stopped being pejorative and was wholeheartedly embraced by the entire industry.

Has anyone attempted to do a timeline of computing by "what no longer matters" in each decade?

> Civilisation advances by extending the number of important operations which we can perform without thinking of them. —ANW

View on HN

I think mechanical sympathy is still useful, even in a world of 'plenty'.. it's people with that kind of understanding who are able to get current day computers to do things that others will have to wait until tomorrow to achieve.

There's something to be said for learning the various effects virtualization layers can have, even if they're not affecting you today.

Something similar will happen again at another level.

When I was streaming audio and video in PlayStation 2 games, I had to care a lot about physical disk layout. But, even then it boiled down to: Seek times are huge, sector and cluster sizes are your atomic ops under the hood, inner rings are low-bandwidth-low-variance, outer rings are high-bandwidth-high-variance.

Since then, not so much.

These days with NVMe drives, things are much easier. But, there are still things to keep in mind. I haven't dug deep. But, my impression is: Even though 4K random reads are incredibly faster than they used to be, large (256K-2MB) reads are yet faster still. Maxing out modern drives requires having multiple requests in flight. So, multiple threads or async operations. Modern drives are so fast that the memory copies involved in buffered fread() can add 33ish% overhead. Consider memory-mapped read-only files. Even better: If you have millions of users, you need 1-in-a-million reliability. Consider SQLite or LMDB as alternatives to fwrite().

I completely forgot about that. I never directly worked on the audio side but I did hear a little from others and the relief that came with the Xbox and caching audio on the HDD. It essentially cut-down on a lot of those issues. And by the next generation, it was only the Wii that you still had to kind of worry about that stuff.

With spinning drives, there was an art to partitioning, where you would lay out your partitions to give some ideal spots on the disk. It probably did not really matter that much. but some people swore that putting your swap partition first was critical to having any sort of performance.

I feel like schools have gotten a lot better about this over the years. Of course students are taught to walk using the rudimentary components when they are learning to program.

We had to write a B-tree implementation in Java when I was in school. I have used that knowledge exactly zero times. But I am betting algorithms are still important.

Yeah I got my undergrad CS degree (a good mix of theory and practice) about 20 years ago and I can't think of much that's truly outdated. I certainly don't use Java or MIPS32 assembly or Solaris these days, but that wasn't really the point of those classes.

Even with truly obsolete stuff like the details of hard drive technology, there's something valuable to be learned in examining how people solved problems at the time with the tools they had.

The B-tree has at least two lessons, independently of whether you ever re-implement one: (a) using trees instead of flat search gains an O(log n) over O(n) algorithmic factor, and (b) using chunky flat elements sized to the bandwidth-delay product of storage results in significant wall-clock (constant factor?) gains.

I fail to see how CHS provides equivalently valuable lessons? https://news.ycombinator.com/item?id=39826433

The principle still matters. Learning how underlying storage works helps optimize data layout, write and read for speed.

It’s just that storage has changed. Caches may be important now for instance

It has changed such that the "traditional" CHS order hasn't been the physical layout of HDDs for probably at least two decades now. Not long ago I posted an article with more details of that: https://news.ycombinator.com/item?id=39428728

I was thinking the lesson from CHS->LBA could be equally well drawn, without bringing in obsolete tech, by exercises involving H:M:S->S or perhaps even D.M.Y->Julian Day.

Reply on news.ycombinator.com