On 13 Dec 2011, Attila Kinali said: > On Thu, 08 Dec 2011 19:59:05 +0000 > Nix <[email protected]> wrote: >> I see no increase in temperature *at all* under heavy CPU load. I find >> this very surprising: 120 wakeups per second (from the timer interrupt >> and the entropy key daemon) shouldn't be stopping the thing from going >> to sleep and cooling down, should it? (Not that 53C really worries me.) > > 120 IRQs/s is nothing. Dont forget that a standard Linux system is
Sorry, that was confusing phrasing on my part. "Heavy CPU load" wasn't the base load mentioned there: it was a giant 'git gc': 100% CPU for an hour plus. :) I was surprised that the base load of 120 wakeups/sec was apparently still high enough that the thing wasn't able to get any cooler than it was when running the giant 'git gc'. (Maybe lm-sensors is lying to me, but I'd not expect a badly calibrated sensor on the Soekris, it's not exactly an off-the-shelf white box.) > configured to 250Hz timer interrupts these days. Systems that need to > be responsive are even configured to 1kHz timer interrrupts. And as > the interrupt service routine is very small, it's still plenty of time > to sleep. > > In short: Even with 120 IRQ/s your system is still basically idle. It's idle (long-term load average 0.05), but it's doing a lot of wakeups, even if it's not doing much work in each one. Some CPUs require long enough to go to sleep again that hundreds of wakeups/sec really would raise their temperature: maybe that would explain why it doesn't get any hotter when I run it at 100% for ages. >> Running an HDD without a fan in the mix to give some airflow seems risky >> to me, no matter the case. > > Why? It's a low power notebook HD. These are made to be run in > enviroments where there is no airflow at all. Just a little bit of > conduction trough the housing. Most notebooks have fans these days, but you're right, I suspect it's the CPU that needs it. I've run several PCs in horrible tower cases with no airflow over the HDD at all, and they never went wrong. I'm just being reflexively paranoid without cause, I guess. >> > This has been tested and reproduced with Linux kernels 2.6.38.x, 2.6.39.1 >> > and 3.0.1 (vanilla, no patches). >> >> FWIW, I've been using a Soekris net5501 for two and a half years now (no >> HDD or wifi). Linux kernels from 2.6.30 up to 3.2rc. No crashes not >> attributable to user error, not one. It's sufficiently reliable that I >> just bought a second one in case lightning strikes the first one and >> cuts me off from the Internet :) > > That's exactly what i'm hitting at: If you run the bare system, you > have no problems. If you choose to include some of the accessories > (like HD mount or wifi cards) sold toghether with the net5501 > (which suggests that these accessories are supported), you will > run into problems. You may be right. I'm probably going to get a lan1741 fairly soon, as my Soekris has no spare Ethernet jacks and the demand on them just keeps going up. We'll see if it retains stability after that. (Hell, given my coordination, we'll see if I can manage to plug the card in without snapping the board in two! :P ) -- NULL && (void) _______________________________________________ Soekris-tech mailing list [email protected] http://lists.soekris.com/mailman/listinfo/soekris-tech
