you have scheduler (elevator) maybe checking this code is nice, but itÂs device based (linux) check /sys/block/xxxx/ to make some tweaks i donÂt know the real implementation (must check source code) but i think mdadm is over linux device layer (it donÂt controll access to disk), i think itÂs async, and linux kernel make thinks more sync... check you raid level source code somethink like: raid6.c itÂs small i think itÂs async does you know where linux kernel can get information (inside /sys or /proc filesystem) to know if you disk is weak? maybe a check and wait before send read command can be implemented... 2011/1/31 Piergiorgio Sartor <piergiorgio.sartor@xxxxxxxx>: > On Mon, Jan 31, 2011 at 08:42:25PM -0200, Roberto Spadim wrote: >> the better probability here is: all disks must be waked up >> since you can have acces of 1gb but starting at a position that all >> disks must be used >> donÂt try used small PSU > > No, that's not an option, HDDs must have staggered > spin up, which is, to certain extent, supported by > some SATA HDD manufactures (not all, it seems). > >> HP Proliant ML310G5 start all hardware on power up (a lot o Watts) and >> after slow down thinks... why? check if PSU is ok, if not, donÂt start >> server. thatÂs a good PSU system. > > As I wrote, efficiency is paramount here. > >> ok if you want to test, i think the worst scenario is all disks beeing >> waked up, i think linux use async (many threads) commands to send >> write/read, maybe you will have a small time between wake up (maybe >> just some microseconds) > > That's interesting information. I suspected something > like async access could have happened. > Is this confirmed? Are the requests from md to the > device layer asynchronous? > How about libata? Does this serialize wake ups? > > Thanks, > > bye, > > pg > >> 2011/1/31 Piergiorgio Sartor <piergiorgio.sartor@xxxxxxxx>: >> > On Mon, Jan 31, 2011 at 07:09:24PM -0200, Roberto Spadim wrote: >> >> you psu must be dimensioned to work with everythink at full work load >> >> (itæ a real production NAS right?! not a test) >> >> your SAS/IDE/SATA controller and HDD manual should be checked >> >> how hdd wake up? one command (read/write) over sata/sas/ide channel wake it up? >> >> on linux raid we have a read algorithm and a write algorithm >> >> if a raid1 write occur all disks will wake up >> >> if a raid1 (raid0 or another) read occur only the disk will wake up >> >> >> >> but check you SATA/IDE/SATA controller, how it wake up your disk, and >> >> how you hdd wake up >> > >> > Hi, thanks for the answer, unfortunately I was >> > hoping to have made myself clear enough. >> > >> > First of all, it is a RAID-6, so let's say that's >> > already decided by requirements. With SATA HDDs. >> > >> > Second, the question was exactly about how the HDDs >> > are waked up. This is a SW issue, trying with normal >> > setups, i.e. a couple of disks, it is possible to >> > send them to sleep (hdparm -y /dev/hdX) and the wake >> > them up by a simple access. >> > I had no opportunity to check this with a RAID-5/6, >> > so I was asking if anyone knows. >> > >> > Finally, in order to be power efficient, the PSU, >> > assuming something like an 80 Plus Gold, should work >> > at not less than 20% of the nominal power, otherwise >> > (according to some reviews), the efficiency drops far >> > below the 80%~90% declared by the 80 Plus standard >> > (which is measured at 20%, 50% and 100% of the maximum >> > specified power). >> > It seem it gets easily around 40%~50%. >> > So, the PSU must be somehow under dimensioned for the >> > spin up of 10 HDDs, which seem to require a possible >> > 30W*10=300W (some nasty HDDs seem to require 30W, in >> > this situation) only for the storage. >> > >> > If the HDDs spin up one after the other, then the peak >> > consumption is only 30W, which might allow a lower >> > power PSU, in contrast with the requirement to provide >> > 300W alone for the spin up. >> > >> > So, back to the original question, if a 10 HDDs RAID-6 >> > is in standby, how do the single HDD will be waked up, >> > in case of access? Of course, a quite larger access, >> > i.e. some GiB of data. >> > >> > Thanks again, >> > >> > bye, >> > >> > -- >> > >> > piergiorgio >> > -- >> > To unsubscribe from this list: send the line "unsubscribe linux-raid" in >> > the body of a message to majordomo@xxxxxxxxxxxxxxx >> > More majordomo info at Âhttp://vger.kernel.org/majordomo-info.html >> > >> >> >> >> -- >> Roberto Spadim >> Spadim Technology / SPAEmpresarial > > -- > > piergiorgio > -- > To unsubscribe from this list: send the line "unsubscribe linux-raid" in > the body of a message to majordomo@xxxxxxxxxxxxxxx > More majordomo info at Âhttp://vger.kernel.org/majordomo-info.html > -- Roberto Spadim Spadim Technology / SPAEmpresarial -- To unsubscribe from this list: send the line "unsubscribe linux-raid" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
