social.dark-alexandr.net

sss friendica
i am searching for silent 24 sas 3.5/2.5hdd 4u rack with lowest possible power consumption (my current rack eating 100w/h without any disks at all)
i need just box with sas expander, without raid or anything.

any suggestions ?
Iron Bug friendica (via ActivityPub)
rack eating 100w/h? this sounds impossible. something must be wrong with it.
I dislike different "smart" boxes, I use old good USB storage. they turn off if not in use.
sss friendica
@Iron Bug unfortunately i need something with efficient cooling, in consumer grade small racks disks is overheating (
but i do not need anything smart, just sas expander with good cooling
Iron Bug friendica (via ActivityPub)
but without load it should not cool anything and consume any power.
sss friendica
@Iron Bug i think coolers by itself should not eat this much anyway, even at full speed.
Iron Bug friendica (via ActivityPub)
I think so too. so this is too suspicious you said it eats 100 watts per hour. this is crazy.
sss friendica
@Iron Bug but this is it.
i am in progress of optimization of power consumption, so i am measuring each device.
Iron Bug friendica (via ActivityPub)
this is a good point. look for hdparm, powertop.
sss friendica
@Iron Bug unfortunately all internal measurement software is unreliable, i am using external watt-metters, as for software tuning for power consumption - already done, maybe a bit stricter separation between cold and hot data may be implemented, but overall everything is optimized already, so for now to optimize it further i need to change some hardware, as i found hdd rack are eating too much for example )
Iron Bug friendica (via ActivityPub)
these are not measurement software, but power control software.
sss friendica
@Iron Bug ah, i remembered now, powertop have some pci tunables exposed, but in my case all pci devices needed at full speed, no unused or idle devices, so nothing to tune here, as for hdparm - sas controllers using scsi commands, so nothing can be done via hdparm, sdparm and/or sg3_utils is used for this tasks.
but anyway, thank you for help attempt.
Iron Bug friendica (via ActivityPub)
if you need it all at max speed then you barely can cut down power consumption at any meaningful scale. it would be too little economy.
sss friendica
@Iron Bug not really, what i am trying to do now, is replacing power-hungry hardware with some low power equivalent, as for servers, they are already optimized, devices which is not needed just physically disconnected.
sss friendica
@Iron Bug but i will recheck powertop, just in case.
sss friendica
@Iron Bug
powertop
modprobe cpufreq_stats failedFailed to mount debugfs!
exiting...


or not...
Iron Bug friendica (via ActivityPub)
hmmm... in my case it works. it failed to work with some newest kernels like a half of a year ago, but then they fixed this.
Iron Bug friendica (via ActivityPub)
but in my case the kernel is not the newest, I don't use the most fresh versions in working PC or servers. I only use them for testing purpose.
sss friendica
@Iron Bug i have intentionally disabled debugfs, and a lot of other unused features and drivers, also some basic tuning is done in kernel build config, including power consumption tuning.
Iron Bug friendica (via ActivityPub)
extended FS drivers are not bad, especially of they're used not all the time and loaded only for certain tasks.
sss friendica
@Iron Bug fs driver itself introducing some small overhead on various subsystems.
Iron Bug friendica (via ActivityPub)
but you don't need it all the time, you need it only for the time of using the utility. and the 'small overhead' you're telling about is microscopic. I cannot even imagine, maybe it's like 0.00000000001% of cpu time. so this is not measurable in any way.
it may be important only in embedded systems with very limited memory resources.
sss friendica
@Iron Bug first, unlike most modules debugfs cannot be built as module, so it is loaded all the time, second, switching debugfs on/off involving rebuilding of most kernel code, which suggesting what something like #ifdef DEBUGS is used in a lot of subsystems, and even it will add just few if () (unneeded avoidable) calls - it's undesired.
sss friendica
@Iron Bug also i disagree with you judgement like:
 maybe it's like 0.00000000001% of cpu time. so this is not measurable in any way.
it may be important only in embedded systems with very limited memory resources.

this leads to creation java-like shit
Iron Bug friendica (via ActivityPub)
this is especially true for FS. FS is slow interface that is referred quite rarely. all the data transfer goes via DMA when the data is ready and it works with disk controller. FS is not that actively used thing, if considering kernel routines.
Iron Bug friendica (via ActivityPub)
nope. this is completely practical thing, I'm a C programmer, not java or other shit coder. and I measured different things thousands of times. you cannot save time on system calls. this is not the level that can be measured, especially in kernel that works at high priority.
sss friendica
@Iron Bug i believe what everything which is unused and can be turned off, should be turned off, in addition to reducing load, it will also introduce simplicity, which often leads to more simple search of problems.
Iron Bug friendica (via ActivityPub)
yes, on server you can remove video card, if exists. and plug off all unused external devices. but you won't get any serious power consumption saving if you server does not work in standby mode.
Iron Bug friendica (via ActivityPub)
I don't know how you build it, but in my case powertop doesn't demand any debugfs. and really I cannot imagine why it could demand it. it's an FS used for debugging. maybe you enabled debug in kernel or something.
sss friendica
@Iron Bug i have looked at pwoertop in past, long time ago, and i think most (if not all) tuning provided by powertop is already done (where applicable), i see no sense in setting ethernet or sas controller in low power profile.
sss friendica
@Iron Bug i am also see no sense in locking cpu on lowest possible frequency and in lowest possible p/c state
Iron Bug friendica (via ActivityPub)
it's not about 'locking on lowest frequency' but the possibility to low down the power mode to levels C3 and below, when a processor is not used.
Iron Bug friendica (via ActivityPub)
not all processors support it but if they do, why not?
sss friendica
@Iron Bug i am using shedutil frequency scaling governor and also p-state driver, so processor does switching in low power mode then idle