" the cylinder of the drive " You cannot know the value of it, and it is less important than stripe size.
I can give you a example, a 2 disk raid-0 system with stripe size=128k. If we do 128k size random access with 128k-aligned offset by multiple thread, we can easily get 200 IO per second. ( read 128k at 128k*n1, 128k*n2, ... 128k*nk) But if we add 64k to each offset, the IO throughput will fall to 100 IO per second. ( read 128k at 128k*n1+64k, 128k*n2+64k, ... 128k*nk+64k) Unfortunately, the default LVM PV header size is 192k, it makes every LV on it misaligned on devices with stripe size more than 64k. With the prior example array, we must call pvcreate with "--metadatasize=192kb" to adjust the PV header size. Regards Kirby Zhou -----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of solarflow99 Sent: Tuesday, August 24, 2010 9:22 PM To: Red Hat Enterprise Linux 5 (Tikanga) discussion mailing-list Subject: Re: [rhelv5-list] How can I align the boundary of partition / LVM / FS within the kickstart script? thats what I thought then, just the partition needs to be aligned to the cylinder of the drive. And for LVM, I think this would just mean the PV, i'm not sure the the LV would be offset differently to cause any problem. On Tue, Aug 24, 2010 at 5:34 AM, Kirby Zhou <[email protected]> wrote: > High level application may consider the stripe size of the underlying > device. > For example, a allocation pool should not let its allocation unit across the > stripe size boundary. > > If partition is not aligned with stripe size, the application would be > confused. > > Regards, > Kirby Zhou > from SOHU-RD +86-10-6272-8261 > > > -----Original Message----- > From: [email protected] [mailto:[email protected]] > On Behalf Of solarflow99 > Sent: Tuesday, August 24, 2010 6:21 PM > To: Red Hat Enterprise Linux 5 (Tikanga) discussion mailing-list > Subject: Re: [rhelv5-list] How can I align the boundary of partition / LVM / > FS within the kickstart script? > > but if the partition is aligned properly with the sector, then > shouldn't it work well? for example, if the stripe size is 16k and > the sector size is 4k, it is always a multiple that divides evenly, I > think this would hold true for both hardware and software raid. > > > > On Tue, Aug 24, 2010 at 2:41 AM, Kirby Zhou <[email protected]> wrote: >> solarflow99, >> >> Stripe size is always larger than sector size. For example, a classical > RAID >> adapter such as DELL Perc6/I should have a stripe size from 16kb to 256kb. >> >> So, if a partition is not aligned with stripe size, a significant >> performance regression of aligned IO should be detected. For example, XFS > is >> sensitive of stripe size, you can do man 'mkfs.xfs', then find "sunit" and >> "swidth" for additional information. >> >> Regards, >> Kirby Zhou >> from SOHU-RD +86-10-6272-8261 >> >> >> -----Original Message----- >> From: [email protected] > [mailto:[email protected]] >> On Behalf Of solarflow99 >> Sent: Tuesday, August 24, 2010 1:08 AM >> To: Red Hat Enterprise Linux 5 (Tikanga) discussion mailing-list >> Subject: Re: [rhelv5-list] How can I align the boundary of partition / LVM > / >> FS within the kickstart script? >> >> On Sat, Aug 21, 2010 at 9:21 PM, Kirby Zhou <[email protected]> wrote: >>> How can I align the boundary of partition / LVM / FS within the kickstart >>> script? >>> >>> There are 3 case which we need align with: >>> >>> 1. The new 4k sector-size hard disk. >>> 2. The strip size of a RAID array. >>> 3. The block size of SSD erasing. >> >> Are you sure about #2? If the partition is aligned with the drive's >> sector size, then the raid stripe size would already be aligned right? >> >> P.S. It's also interesting to mention that anaconda sets a different >> stripe size depending on how big the drive/partition is. I think >> thats hard coded in too. >> >> _______________________________________________ >> rhelv5-list mailing list >> [email protected] >> https://www.redhat.com/mailman/listinfo/rhelv5-list >> >> >> _______________________________________________ >> rhelv5-list mailing list >> [email protected] >> https://www.redhat.com/mailman/listinfo/rhelv5-list >> > > _______________________________________________ > rhelv5-list mailing list > [email protected] > https://www.redhat.com/mailman/listinfo/rhelv5-list > > > _______________________________________________ > rhelv5-list mailing list > [email protected] > https://www.redhat.com/mailman/listinfo/rhelv5-list > _______________________________________________ rhelv5-list mailing list [email protected] https://www.redhat.com/mailman/listinfo/rhelv5-list _______________________________________________ rhelv5-list mailing list [email protected] https://www.redhat.com/mailman/listinfo/rhelv5-list
