On Fri, 11 Sep 2009 12:22:03 +0200
Joachim Schipper <joac...@joachimschipper.nl> wrote:
> On Wed, Sep 09, 2009 at 09:41:05PM +0200, Robert wrote:
> > Hi,
> >
> > I'm investigating slow file transfer speeds between two SATA disks
> > connected to an Intel ICH10R (ahci).
> > Some files copy at a good speed of about ~80MB/s other are as slow
> > as only ~4MB/s.
> > I suspect that those files might be heavily fragmented and the
> > transfer suffers from bad random read performance.
> >
> > How can I check which blocks a file occupies?
> > Is there some tool i am missing to get the info from the inode?
>
> I really wouldn't know, but typically ffs isn't really susceptible to
> fragmentation unless it's almost completely full (and the default
> settings leave at least 5% of disk space for root; you'll also see
> "switching from TIME to SPACE" or something similar in the dmesg,
> indicating that the filesystem is nearly full and will now try to
> conserve space instead of speeding aceess, which does mean it'll
> fragment badly.)
>
> What happens when you create a new file, say using dd, and try to read
> that back? Is your file system nearly full? Did you use sufficiently
> large files to even out the effects of read and write caches, either
> on disk or in the OS? Can you post a dmesg and the results of dd on
> seemingly-similar files? Have you confirmed that it indeed depends on
> the file and not something else (i.e. are "fast" files always fast,
> and "slow" files always slow)?
>
> Joachim
Yes, i had taken that all into account before posting my mail.
The "slow files" are fragmented internally. I confirmed that with
'istat' from sleuthkit. The blocks are not continuous.
They are the result of bad application behavior that wrote the
files.
A copy away and back dance naturally "repairs" the file.
After having the way the app in question writes the data fixed, i don't
see the slowdown anymore.
One can generate such slow files for example by using a bittorent
client that does not preallocate the whole file.
Fwiw i included my test below.
- Robert
# dmesg | fgrep sd1
sd1 at scsibus2 targ 1 lun 0: <, , 0> SCSI0 0/direct removable
sd1: 3824MB, 512 bytes/sec, 7831552 sec total
# fdisk /dev/rsd1c
Disk: /dev/rsd1c geometry: 487/255/63 [7831552 Sectors]
Offset: 0 Signature: 0xAA55
Starting Ending LBA Info:
#: id C H S - C H S [ start: size ]
-------------------------------------------------------------------------------
0: 00 0 0 0 - 0 0 0 [ 0: 0 ] unused
1: 00 0 0 0 - 0 0 0 [ 0: 0 ] unused
2: 00 0 0 0 - 0 0 0 [ 0: 0 ] unused
*3: A6 0 1 1 - 486 254 63 [ 63: 7823592 ] OpenBSD
# disklabel sd1
# /dev/rsd1c:
type: SCSI
disk: SCSI disk
label:
flags:
bytes/sector: 512
sectors/track: 63
tracks/cylinder: 255
sectors/cylinder: 16065
cylinders: 487
total sectors: 7831552
rpm: 3600
interleave: 1
boundstart: 63
boundend: 7823655
drivedata: 0
16 partitions:
# size offset fstype [fsize bsize cpg]
a: 2104452 63 4.2BSD 2048 16384 1
c: 7831552 0 unused
# mount /dev/sd1a /mnt
# cd /mnt
# ls -li test/slowfile
25985 -rw-r--r-- 1 1001 1001 223700743 Sep 11 15:44 test/slowfile
# cp test/slowfile copy
# ls -li copy
3 -rw-r--r-- 1 root 1001 223700743 Sep 11 15:56 copy
# dd if=copy of=/ramdisk/copy bs=1m
213+1 records in
213+1 records out
223700743 bytes transferred in 8.516 secs (26267752 bytes/sec)
# dd if=test/slowfile of=/ramdisk/slowfile bs=1m
213+1 records in
213+1 records out
223700743 bytes transferred in 13.738 secs (16282791 bytes/sec)
# cd
# umount /dev/sd1a
# dd if=/dev/rsd1a of=/tmp/dump.dd bs=1m
# istat -i raw -f ufs1 /tmp/dump.dd 25985 # the slow file
inode: 25985
Allocated
Group: 1
uid / gid: 1001 / 1001
mode: rrw-r--r--
size: 223700743
num of links: 1
Inode Times:
Accessed: Fri Sep 11 16:33:50 2009
File Modified: Fri Sep 11 15:44:01 2009
Inode Modified: Fri Sep 11 15:44:01 2009
Direct Blocks:
116680 116681 116682 116683 116684 116685 116686 116687
[snip]
120552 120553 120554 120555 120556 120557 120558 120559
116536 116537 116538 116539 116540 116541 116542 116543
[snip]
117400 117401 117402 117403 117404 117405 117406 117407
120600 120601 120602 120603 120604 120605 120606 120607
[snip]
120760 120761 120762 120763 120764 120765 120766 120767
119464 119465 119466 119467 119468 119469 119470 119471
[snip]
131056 131057 131058 131059 131060 131061 131062 131063
131064 131065 131066 131067 131068 131069 131070 131071
130720 130721 130722 130723 130724 130725 130726 130727
126640 126641 126642 126643 126644 126645 126646 126647
[snip]
# istat -i raw -f ufs1 /tmp/dump.dd 3 # the copy
inode: 3
Allocated
Group: 0
uid / gid: 0 / 1001
mode: rrw-r--r--
size: 223700743
num of links: 1
Inode Times:
Accessed: Fri Sep 11 16:32:31 2009
File Modified: Fri Sep 11 15:56:24 2009
Inode Modified: Fri Sep 11 15:56:24 2009
Direct Blocks:
1664 1665 1666 1667 1668 1669 1670 1671
1672 1673 1674 1675 1676 1677 1678 1679
1680 1681 1682 1683 1684 1685 1686 1687
1688 1689 1690 1691 1692 1693 1694 1695
[snip]
1888 1889 1890 1891 1892 1893 1894 1895
1896 1897 1898 1899 1900 1901 1902 1903
1904 1905 1906 1907 1908 1909 1910 1911
1912 1913 1914 1915 1916 1917 1918 1919
1920 1921 1922 1923 1924 1925 1926 1927
1928 1929 1930 1931 1932 1933 1934 1935
1936 1937 1938 1939 1940 1941 1942 1943
[snip]
