xfs
[Top] [All Lists]

Re: Question: why does dbench take so much longer to run on XFSthen ext2

To: Philip Chiang <pchiang@xxxxxxxxxxxxxxxxx>
Subject: Re: Question: why does dbench take so much longer to run on XFSthen ext2 file system
From: Steve Lord <lord@xxxxxxx>
Date: 17 Jan 2002 10:27:50 -0600
Cc: linux-xfs@xxxxxxxxxxx
In-reply-to: <000801c19ee4$d012ce30$1701a8c0@xxxxxxxxxxxxxxxxxxxxxxxxxxx>
References: <006001c19ee0$aa218170$1701a8c0@xxxxxxxxxxxxxxxxxxxxxxxxxxx> <1011222311.3602.10.camel@xxxxxxxxxxxxxxxxxxxxxx> <000801c19ee4$d012ce30$1701a8c0@xxxxxxxxxxxxxxxxxxxxxxxxxxx>
Sender: owner-linux-xfs@xxxxxxxxxxx
I ran a test here, dbench 150 on this hardware:

Main memory size: 128 Mbytes
2 GenuineIntel Pentium III (Katmai) processors
2 16550A serial ports
1 vga+ graphics device
2 IDE devices:
  /dev/hda: 20005650 sectors (10243 MB) w/512KiB Cache, CHS=1245/255/63,
UDMA(33)
  /dev/hdb: ATAPI 11X CD-ROM drive, 128kB Cache, UDMA(33)
PCI bus devices:
    Host bridge: Intel Corp. 440BX/ZX - 82443BX/ZX Host bridge (rev 3).
    PCI bridge: Intel Corp. 440BX/ZX - 82443BX/ZX AGP bridge (rev 3).
    ISA bridge: Intel Corp. 82371AB PIIX4 ISA (rev 2).
    IDE interface: Intel Corp. 82371AB PIIX4 IDE (rev 1).
    USB Controller: Intel Corp. 82371AB PIIX4 USB (rev 1).
    Bridge: Intel Corp. 82371AB PIIX4 ACPI (rev 2).
    SCSI storage controller: Adaptec 7896 (#2) (rev 0).
    SCSI storage controller: Adaptec 7896 (rev 0).
    Ethernet controller: Intel Corp. 82557 [Ethernet Pro 100] (rev 8).
    VGA compatible controller: ATI Technologies Inc 3D Rage Pro AGP
1X/2X (rev 92).


Note 2 cpus not 1, plus I was running on a scsi partition, not ide.
For completeness here is my scsi setup:

SCSI subsystem driver Revision: 1.00
scsi0 : Adaptec AIC7XXX EISA/VLB/PCI SCSI HBA DRIVER, Rev 6.2.4
        <Adaptec aic7896/97 Ultra2 SCSI adapter>
        aic7896/97: Ultra2 Wide Channel A, SCSI Id=7, 32/253 SCBs

scsi1 : Adaptec AIC7XXX EISA/VLB/PCI SCSI HBA DRIVER, Rev 6.2.4
        <Adaptec aic7896/97 Ultra2 SCSI adapter>
        aic7896/97: Ultra2 Wide Channel B, SCSI Id=7, 32/253 SCBs

  Vendor: SEAGATE   Model: ST39175LW         Rev: 0001
  Type:   Direct-Access                      ANSI SCSI revision: 02
scsi0:A:1:0: Tagged Queuing enabled.  Depth 253
Attached scsi disk sda at scsi0, channel 0, id 1, lun 0
(scsi0:A:1): 80.000MB/s transfers (40.000MHz, offset 15, 16bit)
SCSI device sda: 17783240 512-byte hdwr sectors (9105 MB)
 sda: sda1 sda2 sda3 sda4 < sda5 sda6 >

This is not a very fast disk.

No quotas or posix acl code in the kernel, using 2.4.17 from the cvs
tree yesterday.

I did a clean mkfs on the xfs filesystem

mkfs -t xfs -f -l size=16384b /dev/sda4
mount -t xfs -o logbufs=8,osyncisdsync /dev/sda4 /xfs

dbench 150

......
Throughput 5.61891 MB/sec (NB=7.02364 MB/sec  56.1891 MBit/sec)

real    58m44.971s
user    3m29.210s
sys     6m59.870s

I then remade the same filesystem using ext2
 mkfs -t ext2 /dev/sda5 
mke2fs 1.19, 13-Jul-2000 for EXT2 FS 0.5b, 95/08/09
Filesystem label=
OS type: Linux
Block size=4096 (log=2)
Fragment size=4096 (log=2)
490560 inodes, 979957 blocks
48997 blocks (5.00%) reserved for the super user
First data block=0
30 block groups
32768 blocks per group, 32768 fragments per group
16352 inodes per group
Superblock backups stored on blocks: 
        32768, 98304, 163840, 229376, 294912, 819200, 884736

Throughput 4.05627 MB/sec (NB=5.07034 MB/sec  40.5627 MBit/sec)

real    81m22.488s
user    3m24.930s
sys     6m38.850s

So a little less system time, but 22 and a half minutes more elapsed time
for ext2.

Possibly the dual cpu is the reason here, not immediately sure why though.
Possibly different behavior between ide and scsi, but it is a very large
difference.

Note that osyncisdsync on the xfs mount options gives xfs the same behavior
on O_SYNC I/O as ext2. aim9 numbers show a significant difference on some
benchmarks with and without this mount option.


My leaning is towards a problem with your build of xfs, or the parameters
used. You did not by any chance turn on the debug parameters in xfs did
you? Having a CONFIG_XFS_DEBUG=y in your .config options will turn xfs into
a dog with 2400 ASSERTS and a lot of other checking addded to the filesystem.

Steve


-- 

Steve Lord                                      voice: +1-651-683-3511
Principal Engineer, Filesystem Software         email: lord@xxxxxxx


<Prev in Thread] Current Thread [Next in Thread>