| |||||||||||||
Fast clocks on slow hardwareFast clocks on slow hardwarePosted Jun 23, 2005 14:24 UTC (Thu) by utoddl (subscriber, #1232)Parent article: What to merge for 2.6.13? I'd like to get an idea of how much apparent difference the proposed timer changes would make to users on extremely old/slow machines. I used to run 2.2 and early 2.4 kernels on my kids' 200MHz 586, and they seemed pretty peppy (the kernels, not the kids). I just installed Fedora Core 4 on it (Hey, they wanted the games), and... it... is... -- heck, I can't type slow enough to express how slow it is. Slug races are more engaging. (I doubled the RAM last night on a lark; no difference. And I'm not talking about the GUI bloat in KDE and GNOME and the accompanying slugishness, I'm just talking about getting to runlevel 3.) I know the new kernels do a lot of stuff, and on newer hardware with lots of resources to manage they are plenty fast enough. But the old boxes really make the cumulative effects of spent cycles stand out noticeably, even if those cycles are doing neat things. This hardware predates 1KHz clocks by eons. So, how much difference would you expect a 250kHz (or even slower) clock to make on a system like this? When I get around to trying it out, I'll let you know what I find. In the mean time, what benchmarks would you suggest to help quantify these differences?
(Log in to post comments)
Fast clocks on slow hardware Posted Jun 23, 2005 14:28 UTC (Thu) by utoddl (subscriber, #1232) [Link] That should be "250Hz", not "250KHz". A 250KHz clock would leave it a smoking pile of slag.
Fast clocks on slow hardware Posted Jun 24, 2005 18:51 UTC (Fri) by pflugstad (subscriber, #224) [Link] I run a 2.6 kernel on a P5/133 MHz laptop - Debian distro (slimmed down) - with 96MB of RAM. At runlevel 3, it's plenty fast for just about everything I do. I even run fvwm and that works fine for a GUI.
I do compile a custom kernel to remove a lot of the stuff I don't need and disable things like HIGHMEM (or whatever it is) that impact things. I'm also very careful to turn off as much stuff as I can.
Heck, suspend even works on this thing.
So it's very possible - it may not work "out of the box" but it runs just fine on slow systems.
Fast clocks on slow hardware Posted Jun 23, 2005 16:01 UTC (Thu) by a9db0 (subscriber, #2181) [Link] You're not the only one with that concern. I run a P90 as a firewall machine, and a P233 as a file server. Heck, my web server is only a PII400. Those minor performance hits on fast machines quickly get magnified on slower boxes.
Fast clocks on slow hardware Posted Jun 23, 2005 19:29 UTC (Thu) by larryr (guest, #4030) [Link] I would be surprised if going from 100Hz to 1000Hz made the system more than 10% slower as far as total throughput. The main area where the change should be noticeable is improvement in interactive response. The scheduling in 2.6 is significantly different from 2.4, independent of the time slice value. Larry
Fast clocks on slow hardware Posted Jun 24, 2005 2:22 UTC (Fri) by giraffedata (subscriber, #1954) [Link] In fact, the only time I've ever heard anyone complain about the clock tick overhead is on systems where lots of Linux systems share the same CPU and are designed for most of those Linux systems to be idle most the time (the clock ticks even on an idle system).In this case, I'd bet money the myriad differences in the kernel between 2.4 and 2.6 all taken together have nothing to do with the slowdown. A whole lot of other software has changed here too. Remember that when CPUs were about 1000 times slower than this 200 MHz thing, Unix systems were already performing reasonably at 100 Hz. So processing 10 times as many ticks ought to be OK.
Fast clocks on slow hardware Posted Jun 24, 2005 4:26 UTC (Fri) by set (guest, #4788) [Link] hmmm. I dont see why a 2.6 kernel would be inherently slower than a2.4 or 2.2 kernel on old hardware. (And Ive run 2.6 kernels on much older, slower boxes than you describe;) (other than in extreme low memory situations...) Definitely, I could see userspace bloat causing a problem, though. (not just desktops, but all the other crap that might be running in a modern distro). I would also not be running a vender kernel on old hardware; its probably not optimised for it, and who knows what all theyve configured it for. Well, at least not a major vendor; maybe one of the distros designed to be small and lean... My current desktop is a pIIx2@400, and it feels better with a 2.6 vs. a 2.4 kernel. Ive also just tested Ubuntu and fc3 on a k6@333, and the only thing slow was gnome;)
Fast clocks on slow hardware Posted Jun 24, 2005 13:53 UTC (Fri) by evgeny (guest, #774) [Link] As others have put it, it's hardly the clock rate that made the difference.
Recently, I've installed Ubuntu on an old 266MHz PII PC. The thing was so dreadfully slow that even the pointer motion had an observable delay (sure enough, the default animated pointer theme didn't help). Then I recompiled the kernel enabling only the needed features/drivers - and it literally flies since then! I highly recommend you try recompiling the kernel; I do it for all systems I own/administer (and of course for Gentoo and like this is the normal setup procedure). The stock kernels of modern binary distros have so much mess enabled by default that I sometimes wonder how these work at all ;-)
Fast clocks on slow hardware Posted Jun 30, 2005 18:09 UTC (Thu) by xorbe (subscriber, #3165) [Link] As a long time Mandrake user (recently switched back when x86_64 mdk arrived), there is definitely something wrong with FC3 I can tell you. It. Is. So. Slow. Clicking, saving, typing, argggh! I have no idea what the underlying issue is there.
Fast clocks on slow hardware Posted Jul 3, 2005 4:12 UTC (Sun) by dmag (subscriber, #17775) [Link] It's probably the CPU type your kernel is compiled for. Nowadays, most distros optimize for recent CPUs. This code will run on your Pentium, but much slower. Having a slow CPU running a non-optimized program is a double whammy. A kernel recompile (telling it exactly which CPU you have) will give your 'slow' CPU a speed boost.I've seen some distos targeted at older computers, but I don't think they go far enough. Here's what I would do:
|
|||||||||||||