libvirt users - Jun 2015 - preallocation=full Vs preallocation=metadata

Hi All

I was reading through this:

http://kashyapc.com/2011/12/02/little-more-disk-io-perf-improvement-with-fallocateing-a-qcow2-disk/

I was basically searching for pointers on improving disk I/O.

I wanted to know the purpose of preallocation=full & 
preallocation=metadata , What is the difference between them ? and which 
one would yield a better disk I/O speed ?

Appreciate your responses in this regard.

Thanks
Jatin

On Tue, Jun 02, 2015 at 09:14:03AM +0530, Jatin Davey
wrote:
> Hi All
> 
> I was reading through this:
> 
>
http://kashyapc.com/2011/12/02/little-more-disk-io-perf-improvement-with-fallocateing-a-qcow2-disk/
In my testing, the above method gives near-raw performance as it
preallocated all the space ahead of time.

The above test can now be done in a single command -- see option (3)
below.
> I was basically searching for pointers on improving disk I/O.
> 
> I wanted to know the purpose of preallocation=full &
preallocation=metadata >
>
> , What is the difference between them ? and which one would yield a better
> disk I/O speed ?
There are three options that modern `qemu-img` supports:

(1) 'preallocation=metadata': allocates qcow2 metadata, and it's
still
    a sparse image.

    $ qemu-img create -f qcow2 -o preallocation=metadata test1-metadata.qcow2 1G
    Formatting 'test1-metadata.qcow2', fmt=qcow2 size=1073741824
encryption=off cluster_size=65536 preallocation='metadata'
lazy_refcounts=off refcount_bits=16
    328K -rw-r--r--. 1 root root 1.1G Jun  3 03:20 copy-test1-metadata.qcow2

(2) 'preallocation=full': allocates zeroes and makes a non-sparse image.

    $ qemu-img create -f qcow2 -o preallocation=full test2-full.qcow2 1G
    Formatting 'test2-full.qcow2', fmt=qcow2 size=1073741824
encryption=off cluster_size=65536 preallocation='full'
lazy_refcounts=off refcount_bits=16
    $ ls -lash test2-full.qcow2 
    1.1G -rw-r--r--. 1 root root 1.1G Jun  3 03:31 test2-full.qcow2

(3) 'preallocation=falloc': which uses posix_fallocate() to
"allocate
    blocks and marking them as uninitialized", and is relatively faster
    than writing out zeroes to a file:

    $ qemu-img create -f qcow2 -o preallocation=falloc test3-falloc.qcow2 1G
    Formatting 'test3-falloc.qcow2', fmt=qcow2 size=1073741824
encryption=off cluster_size=65536 preallocation='falloc'
lazy_refcounts=off refcount_bits=16
    $ ls -lash test3-falloc.qcow2 
    1.1G -rw-r--r--. 1 root root 1.1G Jun  3 03:32 test3-falloc.qcow2

You can test and compare between (2) and (3) what works best for you.

-- 
/kashyap

Thanks Kashyap.

I will test with option 2 & 3 and use whichever gives better results.

Thanks
Jatin

On 6/3/2015 1:08 PM, Kashyap Chamarthy wrote:
> On Tue, Jun 02, 2015 at 09:14:03AM +0530, Jatin Davey wrote:
>> Hi All
>>
>> I was reading through this:
>>
>>
http://kashyapc.com/2011/12/02/little-more-disk-io-perf-improvement-with-fallocateing-a-qcow2-disk/
> In my testing, the above method gives near-raw performance as it
> preallocated all the space ahead of time.
>
> The above test can now be done in a single command -- see option (3)
> below.
>
>> I was basically searching for pointers on improving disk I/O.
>>
>> I wanted to know the purpose of preallocation=full &
preallocation=metadata >
>>
>> , What is the difference between them ? and which one would yield a
better
>> disk I/O speed ?
> There are three options that modern `qemu-img` supports:
>
> (1) 'preallocation=metadata': allocates qcow2 metadata, and
it's still
>      a sparse image.
>
>      $ qemu-img create -f qcow2 -o preallocation=metadata
test1-metadata.qcow2 1G
>      Formatting 'test1-metadata.qcow2', fmt=qcow2 size=1073741824
encryption=off cluster_size=65536 preallocation='metadata'
lazy_refcounts=off refcount_bits=16
>      328K -rw-r--r--. 1 root root 1.1G Jun  3 03:20
copy-test1-metadata.qcow2
>
> (2) 'preallocation=full': allocates zeroes and makes a non-sparse
image.
>
>      $ qemu-img create -f qcow2 -o preallocation=full test2-full.qcow2 1G
>      Formatting 'test2-full.qcow2', fmt=qcow2 size=1073741824
encryption=off cluster_size=65536 preallocation='full'
lazy_refcounts=off refcount_bits=16
>      $ ls -lash test2-full.qcow2
>      1.1G -rw-r--r--. 1 root root 1.1G Jun  3 03:31 test2-full.qcow2
>
> (3) 'preallocation=falloc': which uses posix_fallocate() to
"allocate
>      blocks and marking them as uninitialized", and is relatively
faster
>      than writing out zeroes to a file:
>
>      $ qemu-img create -f qcow2 -o preallocation=falloc test3-falloc.qcow2
1G
>      Formatting 'test3-falloc.qcow2', fmt=qcow2 size=1073741824
encryption=off cluster_size=65536 preallocation='falloc'
lazy_refcounts=off refcount_bits=16
>      $ ls -lash test3-falloc.qcow2
>      1.1G -rw-r--r--. 1 root root 1.1G Jun  3 03:32 test3-falloc.qcow2
>
> You can test and compare between (2) and (3) what works best for you.
>