let´s think about a raid like: md1 = raid1, /dev/sda (ssd), /dev/sdb (ssd) md2 = raid1, /dev/sdc (hdd), /dev/sdd (hdd) md3 = raid0, /dev/md1 (ssd raid), /dev/md2 (hdd raid) (a raid10 like) we could make md3 write/read round robin only, and md2/md1 diferent read/write optimizations 2011/1/18 Roberto Spadim <roberto@xxxxxxxxxxxxx>: > yes i found it in my /sys filesystem, a rotational information 0 for > hd 1 for ssd > > i write a long time ago a more interesting algorithm but complex... a > minimal time algorithm, it should have information about head position > time, read time (per bit, per byte, per units....) and calculate the > time to make a read in each disk considering that it could be reading > (time to stop read current requestion) and after this get the smallest > time -> the best read performace > > if we use only ssd disk today implementation isn´t good, if we use hdd > maybe a good (if we don´t use 7200rpm + 10000rpm + 15000rpm disks), if > we use a mixed ssd+hdd it will not work very good too... this should > be a per disk optimization (minimal time to read) a round robin is a > good feature (for ssd only) but a mixed array should allow minimal > time algorithm > > any idea how to implement a round robin and a algorithm (per raid > device) selection using sysfs? > there´s a patch but i didn´t found information about how to patch the > kernel with it > can anyone help me? > thanks > > 2011/1/18 Mathias Burén <mathias.buren@xxxxxxxxx>: >> 2011/1/18 Keld Jørn Simonsen <keld@xxxxxxxxxx>: >>> On Tue, Jan 18, 2011 at 12:00:49PM -0200, Roberto Spadim wrote: >>>> like this patch (a long time ago) >>>> http://www.spinics.net/lists/raid/msg30003.html >>>> >>>> >>>> 2011/1/18 Roberto Spadim <roberto@xxxxxxxxxxxxx>: >>>> > hi guys, could we implement a load_balance read algorithm for SSD? >>>> > nearest head isn't as fast as round robin for ssd. >>>> > i'm talking about raid1 (raid10 too) >>>> > what's my problem? >>>> > as i can see, raid0 is faster than raid1 >>>> > for example: >>>> > two disks raid0 is faster than >>>> > two disks raid1. >>>> > >>>> > why? >>>> > nearest head >>>> > instead of a balanced read algorithm (like raid0) the nearest head >>>> > make raid1 use only one disk for searchs where we could use two disks >>>> > (like raid0) >>>> > >>>> > could we implement a round robin for ssd? and make raid1 as fast as >>>> > raid0 for ssd? >>>> > i didn't tested the raid10 algorithm yet. >>>> > thanks a lot. >>> >>> This should only be in use for SSDs. For disks it would be a waste of IO >>> bandwidth. How do we detect whether it is a SSSD. >>> Another way to accomplish an improvement os to use the offset layout of >>> raid10. >>> >>> best regards >>> Keld >>> -- >>> 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 >>> >> >> Hi, >> >> There is a way to check if the device is an SSD or not; the rotational >> queue flag in sysfs. See >> http://amailbox.org/mailarchive/git-commits-head/2009/1/30/4859834/thread >> . >> >> // Mathias >> -- >> 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 >> > > > > -- > Roberto Spadim > Spadim Technology / SPAEmpresarial > -- 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
