dd test slowly between LV and file

[stan lin]

Hi all
i have a SAN issue , and try to figure it out.
Test1:
rp7410:/dbexp> time dd if=20080313.bck of=/dev/null bs=1024 count=100000

100000+0 records in
100000+0 records out

real 1.5
user 0.1
sys 1.0

Test2:
rp7410:/dbexp> time dd if=/dev/vg09/lvexp of=/dev/null bs=1024 count=100000

100000+0 records in
100000+0 records out

real 26.7
user 0.1
sys 3.3

Why the real(time) response are so different ?
I don't know that it normal?

[Bill Costigan]

I would use something like glance to see what is really happening.

In the first case you are reading from a file (I assume) and are therefore using the filesystem and any caching it might provide. (I also don't know what the disk(s) are like that contain the file.

In the second case you are reading the logical volume directly. (Have you tested the read from the raw device /dev/vg09/rlvexp?)

I would like to see what the physical vs logical I/Os look like. I'm guessing that the filesystem is actually doing fewer and bigger physical I/Os than is the direct lvol access.

[TTr]

I ran a similar test and the numbers I got were more consistent
FILE
real: 31.0
user: 0.0
sys: 3.1

BLOCK (LV)
real: 15.6
user: 0.0
sys: 4.4

RAW (RLV)
real: 9.0
user: 0.0
sys: 0.1

So my question is as well, is the 20080313.bck on the lvexp volume? If not then you have a disk discrepancy. If it is something else affected your test.

[stan lin]

dear all

in first case , "20080313.bck" is a file.

second case , "/dev/vg09/lvexp" is a Logic volumn .

actully, 20080313.bck file is on the lvexp volume.

i just test and compare those 2 file type why have so different performance.

because my storage performance on Logic volumn is very slowly than file , and i try to find some thing wrong with storage .