Sloppy's Journal: I hate Unix schedulers

One of the things that annoyed the hell out of me when I made the "big switch" around 2000-2002 from AmigaOS to Linux, is the dynamic scheduling. I'm pretty sure I've bitched in my /. journal about this before, but I'm too lazy to go back and look.

Hey, when I "nice" a time-consuming process, I fucking expect it to not slow my computer down, no matter how CPU-intense it is. That's how it was on AmigaOS: I could run as many tasks as I wanted, and as long as I gave them a priority lower (or was it higher, damn I don't remember the specifics) than 0, it had absolutely no impact on the responsiveness of the computer, and anything that I ran at a normal priority, ran just as fast as it would if I hadn't been running those other tasks at all. That's the joy of an absolute scheduler: it starves the low-priority tasks, and as a user that's what I want.

But all the so-called "modern" systems after the 1980s, from OS/2 to Windows to Linux (and now Mac OS as of version 10) totally fuck this up.

My Mac here at work runs a long job every morning, that I have niced. When it's running, the whole damn machine feels sluggish and -- seriously -- I can out-type the speed at which my fucking keystrokes are appearing in this fucking web browser's textarea. It is so utterly ridiculous that a 1.5GHz machine can't run as fast as 50 MHz Amiga.

Niced processes should starve if there's anything better to do. Absolutely starve. That is a good thing, not a bad thing.

But can Unix have this? Nooooo, because something (I don't know what) might deadlock (at least according to Linus, when the topic comes up in the context of Linux). Well, get your locks sorted out, Unixheads, so that maybe someday Unix can run as fast as an Amiga that has a tenth of the processing power.

BTW

[Sloppy]

No need to tell me, "The Amiga is dead, get over it." I know. I don't need the Amiga to come back, I just want its nonsucky aspects to come back.

Re:

[loucura!]

If you don't want a process to run, why execute it? If you're willing to starve it, why don't you move it to a period of time when you're not interacting with the machine?

Re:

[Sloppy]

If you don't want a process to run, why execute it?

I want it to run, but only when my web browser is idle. There's an abundance of CPU time available in between each one of my keystrokes, and I'm happy for niced processes to use the CPU during that time. But the very instant I press a key, niced processes should be completely pre-empted until the interactive process (web browser) finishes processing the keystroke (echoing it into the textarea) or otherwise blocks for some reason.

why don't you move it t

Realtine patches

[jawtheshark]

Have you tried the realtime kernel patches for Linux? That's what you're describing: you want a realtime-os. In that case, get the correct kernel. At least with Linux, you can.

Or try QNX ;-)

Re:

[Cyberdyne]

Have you tried the realtime kernel patches for Linux? That's what you're describing: you want a realtime-os. In that case, get the correct kernel. At least with Linux, you can.

Even without the realtime patches, schedtool -B on a recent kernel should give exactly the results Sloppy is after: 'run this process only if you have nothing better to do (so it's OK if it never gets done)', as opposed to nice's 'second class mail (it'll get there, but not as quickly)' meaning.

Meanwhile, my monster Opteron server

Re:

[jawtheshark]

Yes, I understand that and it clearly does and the question is: Why? Why can't it be fixed? My jobs runs at 4h00 every night, I don't notice them. Stupid workaround, yes, I know.

Opteron server spent the entire day fscking, after a long power outage last night defeated the UPS.[...] 95 million files

You think that NTFS would have fared better? ;-)

Re:

[Sloppy]

Even without the realtime patches, schedtool -B on a recent kernel should give exactly the results Sloppy is after: 'run this process only if you have nothing better to do (so it's OK if it never gets done)', as opposed to nice's 'second class mail (it'll get there, but not as quickly)' meaning.

Ooh neat, I'll check into that. Thanks. :-)

Re:

[Sloppy]

That's what you're describing: you want a realtime-os. In that case, get the correct kernel. At least with Linux, you can.

I haven't tried that with Linux, but my Linux machines (i.e. home computers) don't have to work nearly as hard as the computer I use at .. well .. work. And at work, I run MacOS X; no RT patch available (as far as I know!).

I'm surprised that what I want is a special case, though. Maybe (and I'm not really sure about this) an absolute scheduler isn't the best thing for a traditional

Re:

[jawtheshark]

Do you know Tet? Talk to Tet [slashdot.org].... He was an Amiga user, converted to Linux. He will most certainly knows what to do.