1 |
On 06/11/2014 11:49 AM, Stefan G. Weichinger wrote: |
2 |
> Am 11.06.2014 12:41, schrieb thegeezer: |
3 |
> |
4 |
>>> everything around 380 MB/s ... only ~350 MB/s for |
5 |
>>> /dev/vg01/winserver_disk0 (which still is nice) |
6 |
>> |
7 |
>> OK here is the clue. |
8 |
>> if the LVs are also showing such fast speed, then please can you show |
9 |
>> your command that you are trying to run that is so slow ? |
10 |
> I originally noticed that virt-backup was slow so I looked into it and |
11 |
> found some dd-command. |
12 |
> |
13 |
> My tests right now are like this: |
14 |
> |
15 |
> |
16 |
> |
17 |
> booze ~ # dd if=/dev/vg01/winserver_disk0 bs=1M of=/dev/null |
18 |
> ^C25+0 Datensätze ein |
19 |
> 24+0 Datensätze aus |
20 |
> 25165824 Bytes (25 MB) kopiert, 13,8039 s, 1,8 MB/s |
21 |
> |
22 |
> booze ~ # dd if=/dev/vg01/winserver_disk0 bs=4M of=/dev/null |
23 |
> ^C6+0 Datensätze ein |
24 |
> 5+0 Datensätze aus |
25 |
> 20971520 Bytes (21 MB) kopiert, 12,5837 s, 1,7 MB/s |
26 |
> |
27 |
> booze ~ # dd if=/dev/vg01/winserver_disk0 of=/dev/null |
28 |
> ^C55009+0 Datensätze ein |
29 |
> 55008+0 Datensätze aus |
30 |
> 28164096 Bytes (28 MB) kopiert, 12,611 s, 2,2 MB/s |
31 |
> |
32 |
> So no "copy from-to same disk here" ... should be just plain reading, right? |
33 |
> |
34 |
> virt-backup does some ionice-stuff as well, but as you see, my |
35 |
> test-commands don't. |
36 |
> |
37 |
> # cat /sys/block/sdc/queue/scheduler |
38 |
> [noop] deadline cfq |
39 |
> |
40 |
> -> noop scheduler to let the controller do its own scheduling |
41 |
> |
42 |
> |
43 |
> |
44 |
> thanks, Stefan |
45 |
> |
46 |
> |
47 |
> |
48 |
yeah this is very very odd. |
49 |
firstly there should not be such discrepancy between hdparm -t and dd if= |
50 |
secondly you would imagine that the first dd would be cached and so |
51 |
would be faster the second time round |
52 |
please check for the turbo boost disable, i'll have a closer look at the |
53 |
cgroups |