Thank you for explanation. I though that it is possible to write only extents since I don't care about the data. As I understand correctly currently there is no possibility to do that? On Tue, Apr 11, 2017 at 12:23 AM, Darrick J. Wong <darrick.wong@xxxxxxxxxx> wrote: > On Mon, Apr 10, 2017 at 08:12:19AM +0200, Arkadiusz wrote: >> I know that the flag means "0010000 Unwritten preallocated extent" but >> is there any way to allocate extents fast without filling whole file >> with data? > > Nope. I wonder if fallocate(ZERO_RANGE) could be taught to issue a > zeroing discard before ensuring that all the extents are marked as > written? > > (Ask the list, see what people say.) > > --D > >> >> On Fri, Apr 7, 2017 at 6:29 PM, Darrick J. Wong <darrick.wong@xxxxxxxxxx> wrote: >> > On Fri, Apr 07, 2017 at 02:52:01PM +0200, Arkadiusz wrote: >> >> Dear maintainers, >> >> I have a question about large file fragmentation. >> >> >> >> When I create large file (20G) on XFS by filling it with zeros. For example: >> >> dd if=/dev/zero of=lun bs=512 count=41873408 >> >> >> >> The bitmap looks as follows: >> >> >> >> xfs_bmap -vp lun >> >> lun: >> >> EXT: FILE-OFFSET BLOCK-RANGE AG AG-OFFSET >> >> TOTAL FLAGS >> >> 0: [0..8388479]: 112..8388591 0 (112..8388591) >> >> 8388480 00000 >> >> 1: [8388480..16777087]: 10485824..18874431 1 (64..8388671) >> >> 8388608 00000 >> >> 2: [16777088..25165695]: 20992064..29380671 2 (20544..8409151) >> >> 8388608 00000 >> >> 3: [25165696..35651383]: 31457344..41943031 3 (64..10485751) >> >> 10485688 00000 >> >> 4: [35651384..37748551]: 8388592..10485759 0 (8388592..10485759) >> >> 2097168 00000 >> >> 5: [37748552..39845631]: 18874432..20971511 1 (8388672..10485751) >> >> 2097080 00000 >> >> 6: [39845632..41873407]: 29380672..31408447 2 (8409152..10436927) >> >> 2027776 00000 >> >> >> >> When I use this file for iSCSI file I/O and then I fill whole volume >> >> from the initiator side with random data. The bitmap doesn't change. >> >> >> >> However, when I create file of the same size using fallocate: >> >> fallocate -l 21439184896 lun >> >> the bitmap at the beginning looks as follows: >> >> xfs_bmap -vp lun >> >> lun: >> >> EXT: FILE-OFFSET BLOCK-RANGE AG AG-OFFSET >> >> TOTAL FLAGS >> >> 0: [0..10485647]: 112..10485759 0 (112..10485759) >> >> 10485648 10000 >> >> 1: [10485648..10485655]: 104..111 0 (104..111) >> >> 8 10000 >> >> 2: [10485656..20971351]: 10485824..20971519 1 (64..10485759) >> >> 10485696 10000 >> >> 3: [20971352..31436559]: 20992064..31457271 2 (20544..10485751) >> >> 10465208 10000 >> >> 4: [31436560..41873407]: 31457344..41894191 3 (64..10436911) >> >> 10436848 10000 >> >> >> >> but when I write some data from the initiator side the extents count grows: >> >> xfs_bmap -vp lun >> >> lun: >> >> EXT: FILE-OFFSET BLOCK-RANGE AG AG-OFFSET >> >> TOTAL FLAGS >> >> 0: [0..7]: 112..119 0 (112..119) >> >> 8 00000 >> >> 1: [8..2047]: 120..2159 0 (120..2159) >> >> 2040 10000 >> > >> > Have a look at "xfs_bmap -vvp lun" for the flags output. ;) >> > >> > --D >> > >> >> 2: [2048..365807]: 2160..365919 0 (2160..365919) >> >> 363760 00000 >> >> 3: [365808..406727]: 365920..406839 0 (365920..406839) >> >> 40920 10000 >> >> 4: [406728..506487]: 406840..506599 0 (406840..506599) >> >> 99760 00000 >> >> 5: [506488..514071]: 506600..514183 0 (506600..514183) >> >> 7584 10000 >> >> 6: [514072..514079]: 514184..514191 0 (514184..514191) >> >> 8 00000 >> >> 7: [514080..524495]: 514192..524607 0 (514192..524607) >> >> 10416 10000 >> >> 8: [524496..524503]: 524608..524615 0 (524608..524615) >> >> 8 00000 >> >> 9: [524504..524543]: 524616..524655 0 (524616..524655) >> >> 40 10000 >> >> 10: [524544..524551]: 524656..524663 0 (524656..524663) >> >> 8 00000 >> >> 11: [524552..529559]: 524664..529671 0 (524664..529671) >> >> 5008 10000 >> >> 12: [529560..529567]: 529672..529679 0 (529672..529679) >> >> 8 00000 >> >> 13: [529568..550255]: 529680..550367 0 (529680..550367) >> >> 20688 10000 >> >> 14: [550256..550263]: 550368..550375 0 (550368..550375) >> >> 8 00000 >> >> 15: [550264..552295]: 550376..552407 0 (550376..552407) >> >> 2032 10000 >> >> 16: [552296..552303]: 552408..552415 0 (552408..552415) >> >> 8 00000 >> >> 17: [552304..554279]: 552416..554391 0 (552416..554391) >> >> 1976 10000 >> >> 18: [554280..554287]: 554392..554399 0 (554392..554399) >> >> 8 00000 >> >> 19: [554288..570623]: 554400..570735 0 (554400..570735) >> >> 16336 10000 >> >> 20: [570624..570631]: 570736..570743 0 (570736..570743) >> >> 8 00000 >> >> 21: [570632..1330943]: 570744..1331055 0 (570744..1331055) >> >> 760312 10000 >> >> 22: [1330944..1330951]: 1331056..1331063 0 (1331056..1331063) >> >> 8 00000 >> >> 23: [1330952..6173167]: 1331064..6173279 0 (1331064..6173279) >> >> 4842216 10000 >> >> 24: [6173168..6232727]: 6173280..6232839 0 (6173280..6232839) >> >> 59560 00000 >> >> 25: [6232728..6249039]: 6232840..6249151 0 (6232840..6249151) >> >> 16312 10000 >> >> 26: [6249040..6249111]: 6249152..6249223 0 (6249152..6249223) >> >> 72 00000 >> >> 27: [6249112..6251087]: 6249224..6251199 0 (6249224..6251199) >> >> 1976 10000 >> >> 28: [6251088..6251127]: 6251200..6251239 0 (6251200..6251239) >> >> 40 00000 >> >> 29: [6251128..6251631]: 6251240..6251743 0 (6251240..6251743) >> >> 504 10000 >> >> 30: [6251632..6296063]: 6251744..6296175 0 (6251744..6296175) >> >> 44432 00000 >> >> 31: [6296064..10485647]: 6296176..10485759 0 (6296176..10485759) >> >> 4189584 10000 >> >> 32: [10485648..10485655]: 104..111 0 (104..111) >> >> 8 10000 >> >> 33: [10485656..20971351]: 10485824..20971519 1 (64..10485759) >> >> 10485696 10000 >> >> 34: [20971352..31436559]: 20992064..31457271 2 (20544..10485751) >> >> 10465208 10000 >> >> 35: [31436560..41869303]: 31457344..41890087 3 (64..10432807) >> >> 10432744 10000 >> >> 36: [41869304..41869311]: 41890088..41890095 3 (10432808..10432815) >> >> 8 00000 >> >> 37: [41869312..41873407]: 41890096..41894191 3 (10432816..10436911) >> >> 4096 10000 >> >> >> >> I thought that fallocate allocates extents and fragmentation shouldn't increase. >> >> >> >> Is there any way to create large files resistant to fragmentation that >> >> doesn't need to fill whole file with zeros? >> >> >> >> Thanks. >> >> -- >> >> To unsubscribe from this list: send the line "unsubscribe linux-xfs" in >> >> the body of a message to majordomo@xxxxxxxxxxxxxxx >> >> More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe linux-xfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
