[linux-lvm] howto speed-up thinpool-devices?

[linux-lvm] howto speed-up thinpool-devices?

[Oliver Rath]

Hi list,

Im using thinpool-devices (and snapshots, too). I know, thinpool-devices
arent the fastest one, but maybe there could make more use of read (and
maybe write) caching. In my test-system are 32GB of ram, but there is a
max use of cache about <1GB, not more, by heavily use of these devices.

Is it possible to speed up the reading with growing up the cache use?
The /sys/block/dm-X/.. directories have lots of parameters, but Im not
experienced for that. Maybe Im looking@the wrong place and modifying
kernel parameters (scheduler etc.) are the better way?

Tfh!
Oliver

Re: [linux-lvm] howto speed-up thinpool-devices?

[Marian Csontos]

On 08/24/2013 10:42 AM, Oliver Rath wrote:
> Hi list,
>
> Im using thinpool-devices (and snapshots, too). I know, thinpool-devices
> arent the fastest one, but maybe there could make more use of read (and
> maybe write) caching. In my test-system are 32GB of ram, but there is a
> max use of cache about <1GB, not more, by heavily use of these devices.

Buffers may be more relevant metric as the I/O among snapshots is 
"shared" on block device level.

I am using thin snapshots rather heavily (in VMs) and buffers are using 
13 GB of total 32 GB RAM on the machine (and it is not running@full 
load now)

Are you using FS on top of the thin devices?

-- Martian

>
> Is it possible to speed up the reading with growing up the cache use?
> The /sys/block/dm-X/.. directories have lots of parameters, but Im not
> experienced for that. Maybe Im looking at the wrong place and modifying
> kernel parameters (scheduler etc.) are the better way?
>
> Tfh!
> Oliver
>
> _______________________________________________
> linux-lvm mailing list
> linux-lvm@redhat.com
> https://www.redhat.com/mailman/listinfo/linux-lvm
> read the LVM HOW-TO at http://tldp.org/HOWTO/LVM-HOWTO/
>

Re: [linux-lvm] howto speed-up thinpool-devices?

[Zdenek Kabelac]

Dne 24.8.2013 10:42, Oliver Rath napsal(a):
> Hi list,
>
> Im using thinpool-devices (and snapshots, too). I know, thinpool-devices
> arent the fastest one, but maybe there could make more use of read (and
> maybe write) caching. In my test-system are 32GB of ram, but there is a
> max use of cache about <1GB, not more, by heavily use of these devices.
>
> Is it possible to speed up the reading with growing up the cache use?
> The /sys/block/dm-X/.. directories have lots of parameters, but Im not
> experienced for that. Maybe Im looking at the wrong place and modifying
> kernel parameters (scheduler etc.) are the better way?
>


You  could look at lot of Howto text around various kernel settings i.e.:

http://en.wikipedia.org/wiki/Swappiness
https://www.kernel.org/doc/Documentation/sysctl/vm.txt

Check with Google for more hints.

Zdenek

Re: [linux-lvm] buffers shared between different devs? (was: howto speed-up thinpool-devices?)

[Linda A. Walsh]

On 9/3/2013 3:06 AM, Marian Csontos wrote:
> Buffers may be more relevant metric as the I/O among snapshots is 
> "shared" on block device level.
----
I am not going to say this isn't true, BUT, some parts of the kernel seem to
report different block numbers for files on the same device  when you mount
the same device (or possible subtrees of a same device) at alternate mount
points via "bind".

I have my "/usr/share" physically located on my "/home" partition under
"/home/share".  In moving files from /usr/share/<subdir> to
/home/share/<subdir2>, the coreutils "mv" cmd, actually does a file copy
-- vs. when a file is detected as being on the same device, it usually
does a simple rename.

If 'mv' is seeing different device id's for a file, it's *less probable*,
but possible that such misinformation could exist at lower levels in the
kernel (?).  Do you know whether or not the kernel "knows" they are on the
same device, or is it possible the buffer area of the kernel has the same
bug as the upper level that is reporting different devices to "mv"?