[linux-lvm] Configuring extent size

[linux-lvm] Configuring extent size

[hansbkk]

There is a lot of apparently authoritative but conflicting information
on this topic out on the interwebs. I suspect the problem is that
changes from LVM1 to LVM2 weren't explicitly highlighted.

For example, I believe this is no longer true, but it still figures
prominently in returns from google:

> a 4 MB extent size imposes a logical volume size limitation of 256 Gigabytes

I believe the below is now true, and that this is the reason the above
is no longer relevant:

> with LVM2, there's no limit on the maximum numbers of extents per PV/LV.

IOW, the FAQ's statement:

> For 32-bit CPUs on 2.6 kernels, the maximum LV size is 16TB.

is a hard upper limit, and is not related to extent size - please confirm.


And is there a maximum size for PV or VG and would this be related to
extent size?


Also, the man page state that the *default* extent size (if none is
specified) is 4MB, but the HowTo states it's 32MB. Which is true?

Or does this vary by vgcreate version, or per-distro configuration?


Finally

> having a large number of extents will slow down the tools

leads me to:

Would there be any negative impact of a very large extent size?

I'm thinking of using 1GB (4096*256=1048576). This would make extents
human-friendly when used directly as units in allocating space. Even
10 GB would only be 2% of the smallest drive I'm using, so wouldn't
hamper flexibility, but IMO this seems a bit "out there".

Any and all feedback welcome.

Re: [linux-lvm] Configuring extent size

[Stuart D Gathman]

On 12/06/2010 08:33 PM, hansbkk@gmail.com wrote:
> Also, the man page state that the *default* extent size (if none is
> specified) is 4MB, but the HowTo states it's 32MB. Which is true?
>
> Or does this vary by vgcreate version, or per-distro configuration?
Based on experience, this varies by distro, so it must be a system
configuration option.
> having a large number of extents will slow down the tools

I believe only a large number segments (contiguous groups of extents)
slows things down.  I could be wrong. 

> leads me to:
>
> Would there be any negative impact of a very large extent size?
The only impact is unused space due to more granular allocation.

Re: [linux-lvm] Configuring extent size

[hansbkk]

I've restated my understanding of your response - please let me know
if you meant something different.

>> having a large number of extents will slow down the tools
>
> I believe only a large number segments (contiguous groups of extents)
> slows things down. �I could be wrong.


The total number of extents isn't the factor that could slow down the
LVM tools, it's the total number of segments (contiguous groups of
extents) that matters.

So for example, setting up fine-grained LVM striping and leaving empty
extents in between large numbers of occupied stripes would have more
of an impact than the total number of extents.

>> leads me to:
>>
>> Would there be any negative impact of a very large extent size?
> The only impact is unused space due to more granular allocation.

So if I had 10GB extents, but needed a bunch of separate LVs that each
only needed <1GB, that would lead to wasted space. . .

Re: [linux-lvm] Configuring extent size

[hansbkk]

While I'm at it, I've also come across statements that the performance
of certain applications (Oracle usually being the one mentioned) can
be sensitive to LVM extent size.

I'm hoping this is not actually the case, and such ideas are coming
from those who may have tried to use LVM to handle their striping.

Which leads to another question, OT enough to start a new thread. . .

On Tue, Dec 7, 2010 at 1:29 PM,  <hansbkk@gmail.com> wrote:
> I've restated my understanding of your response - please let me know
> if you meant something different.
>
>>> having a large number of extents will slow down the tools
>>
>> I believe only a large number segments (contiguous groups of extents)
>> slows things down. �I could be wrong.
>
>
> The total number of extents isn't the factor that could slow down the
> LVM tools, it's the total number of segments (contiguous groups of
> extents) that matters.
>
> So for example, setting up fine-grained LVM striping and leaving empty
> extents in between large numbers of occupied stripes would have more
> of an impact than the total number of extents.
>
>>> leads me to:
>>>
>>> Would there be any negative impact of a very large extent size?
>> The only impact is unused space due to more granular allocation.
>
> So if I had 10GB extents, but needed a bunch of separate LVs that each
> only needed <1GB, that would lead to wasted space. . .
>

Re: [linux-lvm] Configuring extent size

[hansbkk]

Sorry to appear to be talking to myself here, but thought I'd try to
get confirmations/answers on questions outstanding from my OP:

> a 4 MB extent size imposes a logical volume size limitation of 256 Gigabytes
No longer true?

Because this is now true?
> with LVM2, there's no limit on the maximum numbers of extents per PV/LV.

And therefore
> For 32-bit CPUs on 2.6 kernels, the maximum LV size is 16TB.

is a hard upper limit, and is not related to extent size?


I have in the meantime found this:

Max number PVs: 2**32 PVs

but not a statement about any maximum size for PV (nor more importantly for VGs)

I'm guessing the question is academic, and would in any case not be
related to extent size anyway.



On Tue, Dec 7, 2010 at 1:47 PM,  <hansbkk@gmail.com> wrote:
> While I'm at it, I've also come across statements that the performance
> of certain applications (Oracle usually being the one mentioned) can
> be sensitive to LVM extent size.

I'm hoping this is not actually the case, and such ideas are coming
from those who may are using LVM to handle their performance striping
rather than the underlying RAID.


All this is pretty much just one noob's speculation here, some
confirmation would be greatly appreciated.