[linux-lvm] Best Practices deploying LVM

[linux-lvm] Best Practices deploying LVM

[Abraham Pérez]

[-- Attachment #1: Type: text/plain, Size: 584 bytes --]

Hi everyone,

I'm looking for some whitepapper or similar document to find any kind of
best practices using LVM.

For example, usually i make only one volume group in each server and inside
it I create different logical volumes for different purposes, but talking
with one colleague, we discover that he makes one volume group for each
logical volume (more or less)... so my final question is: in performance
terms, what configuration is more efficient and why?

Please excuse my writen english, but I don't use it very often.

Thanks for your attention,
Abraham Pérez

[-- Attachment #2: Type: text/html, Size: 611 bytes --]

Re: [linux-lvm] Best Practices deploying LVM

[Chris Cox]

On Fri, 2009-10-30 at 00:45 +0100, Abraham Pérez wrote:
> Hi everyone,
> 
> I'm looking for some whitepapper or similar document to find any kind
> of best practices using LVM.
> 
> For example, usually i make only one volume group in each server and
> inside it I create different logical volumes for different purposes,
> but talking with one colleague, we discover that he makes one volume
> group for each logical volume (more or less)... so my final question
> is: in performance terms, what configuration is more efficient and
> why?

Define "efficient"... anyhow, the idea of a volume group is
to present a pool of storage.  So making a volume group per logical
volume sort of defeats the intended purpose.

So... a volume group has more to do with defining a class of
storage pool.  Ideally physical volumes inside of a single volume
group have some kind of relationship (e.g. maybe they are all
of the same speed or type or raid, etc).  It's common for low
end systems to have just one volume group.  However, you COULD
segment storage pool types by function.  Thus you could have
a System volume group for the OS volumes and a Data volume group
for user data (just an example).

There is a lot of ways to look at it.

Personally, just because of pvmove, I see physical volumes
as whole disks and not a myriad of piece meal partitions.
That way you get the potential ability to move data off
of a physical volume in a volume group to free areas on
different drives in that same group and remove a drive
from the volume group and from the system (e.g. think of upgrading
the size of drives you use inside of a volume group... you could
use this to upgrade the drives without having to necessarily
take anything down).

So my best practices...

1. Use multiple volume groups to define classes of storage.

2. Use full disk physical volumes over partitions.

3. I like unique naming of volume groups just so that if
a drive lands else where, and it presents itself to the system,
it will not collide with existing volume group names.
(but there is good reason for not doing this... and really
only a factor if you have a tendency to throw drives into
different computers all of the time)

> 
> Please excuse my writen english, but I don't use it very often.

I use english all of the time.... it doesn't seem to help.

Re: [linux-lvm] Best Practices deploying LVM

[Ray Morris]

    I don't know about a whitepaper, but I can address
your example.

> he makes one volume group for each logical volume (more or less)

    If each one has one volume, that's not exactly a volume
GROUP, is it?  If groups and volumes are basically synomous,
he gives up all the benfits of groups.  In fact, he gives
up most of the benefits of logical volumes, since each PV
has to be in one group, and each VG is one LV, you're left
with one LV per PV - might as well just use partitions
directly.
--
Ray Morris
support@bettercgi.com



On 10/29/2009 06:45:39 PM, Abraham P�rez wrote:
> Hi everyone,
> 
> I'm looking for some whitepapper or similar document to find any kind  
> of
> best practices using LVM.
> 
> For example, usually i make only one volume group in each server and  
> inside
> it I create different logical volumes for different purposes, but  
> talking
> with one colleague, we discover that he makes one volume group for  
> each
> logical volume (more or less)... so my final question is: in  
> performance
> terms, what configuration is more efficient and why?
> 
> Please excuse my writen english, but I don't use it very often.
> 
> Thanks for your attention,
> Abraham P�rez
> 

------quoted attachment------
> _______________________________________________
> 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] Best Practices deploying LVM

[malahal]

Ray Morris [support@bettercgi.com] wrote:
>     I don't know about a whitepaper, but I can address
> your example.
> 
> > he makes one volume group for each logical volume (more or less)
> 
>     If each one has one volume, that's not exactly a volume
> GROUP, is it?  If groups and volumes are basically synomous,
> he gives up all the benfits of groups.  In fact, he gives
> up most of the benefits of logical volumes, since each PV
> has to be in one group, and each VG is one LV, you're left
> with one LV per PV - might as well just use partitions
> directly.

I agree, you lose some flexibility but it has some advantage compared to
plain partitions without LVM. E.g. he can make a file system larger than
any disk with multiple disks in the above LVM (one LV per VG)
configuration.  There are other advantages. I am not sure the reason for
making only one LV per VG though!

Thanks, Malahal.

Re: [linux-lvm] Best Practices deploying LVM

[Abraham Pérez]

[-- Attachment #1: Type: text/plain, Size: 2440 bytes --]

Thanks for the instant answers!

Well... I'll try to explain myself better. I'm working in a client who have
a lot of servers running under VMware. This client have a lot of problems
with the storage, because they never have enough space so when they have to
allocate disk in servers, they add small virtual hard disks with, for
example, 5 or 10GB.

Then for the OS installation, we follow the basic schema based on disk
partitions (/dev/sda1 pointing to / with ext3, /dev/sda2 pointing to /home
and so on) and for the applications data, we use VG and LV pointing to /opt

The client have some applications who need a lot of mountpoints, so my
colleague adds 1-3 LV per VG (aproximated) and I only create only one VG and
inside it, different LVs.  With this infrastructure, we have to discard
different kinds of hard disk because they're exactly the same... and we have
that doubt: what schema is better and why, discarding concept things like a
volume group was designed to be a group, because we're looking for good
reasons based in performance of future actions, it's not important... or am
I mistaken???

I don't know if I explained myself very well, so thanks all anyway!

Regards,
Abraham Pérez

2009/10/30 <malahal@us.ibm.com>

> Ray Morris [support@bettercgi.com] wrote:
> >     I don't know about a whitepaper, but I can address
> > your example.
> >
> > > he makes one volume group for each logical volume (more or less)
> >
> >     If each one has one volume, that's not exactly a volume
> > GROUP, is it?  If groups and volumes are basically synomous,
> > he gives up all the benfits of groups.  In fact, he gives
> > up most of the benefits of logical volumes, since each PV
> > has to be in one group, and each VG is one LV, you're left
> > with one LV per PV - might as well just use partitions
> > directly.
>
> I agree, you lose some flexibility but it has some advantage compared to
> plain partitions without LVM. E.g. he can make a file system larger than
> any disk with multiple disks in the above LVM (one LV per VG)
> configuration.  There are other advantages. I am not sure the reason for
> making only one LV per VG though!
>
> Thanks, Malahal.
>
> _______________________________________________
> 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/
>

[-- Attachment #2: Type: text/html, Size: 3135 bytes --]

Re: [linux-lvm] Best Practices deploying LVM

[Ray Morris]

    With your way, you can add a new 2GB disk and use
half of it for /home, half for /opt, if you wish.
You can also leave some of it unused and expand
any LV in the future as needed.  That's one important
reason why most people use voluem groups as groups -
contianing several volumes.  Does your colleague know
of ANY advantage to creating a bunch of different
groups?   If not, your way wins - it has advantages
over his way, and his way has no advantage.

> we have to discard different kinds of hard disk
> because they're exactly the same

   I have no idea what this is supposed to mean.
Different kinds of hard disk are exactly the same?
If this is supposed to mean "we are not able to
use different types of drives for different
partitions", I can understand that.  However,
for what purpose would you use different types
of disks?  Perhaps he wants a fast disk for one
partition, and a large, cheap disk for another
partition?  If you use two cheap disks in a
striped LV it's going to be faster AND cheaper
than the "fast drive" option.  MAYBE if you were
going to use a super fast RAID array of SSD drives
for some small amount of data, but not if we're
talking standard magnetic hard drives.


> a lot of servers running under VMware. This client
> have a lot of problems with the storage, because they
> never have enough space so when they have to allocate
> disk in servers, they add small virtual hard disks
> with, for example, 5 or 10GB.

lvextend.  Ours resize automatically on the fly, using
a cron job that checks to see if any virtual servers
need more space.

> discarding concept things like a volume group was designed
> to be a group, because we're looking for good reasons

   "Concept reasons", like using the tool designed for the
job, may be the very best reasons because that one reason
actually covers the hundred reasons that don't come to mind
immediately.  You don't know what issues you'll run into
next week or next month, but you can bet you won't be the
first one - other people will have had the same issues,
and will use the standard tools for the standard purposes
to solve that problem.  Better for you if you can use
the same solutions.  Also, there are certain features
and optimizations you don't know about, but you'll gain
from those grouping features if you use groups as groups.
No one knows about the features and optimizations that
will be added next year, but if you use the tools the
way they were designed you'll benefit from future
enhancements that allow you to better use them for their
purposes.
--
Ray Morris
support@bettercgi.com

Strongbox - The next generation in site security:
http://www.bettercgi.com/strongbox/

Throttlebox - Intelligent Bandwidth Control
http://www.bettercgi.com/throttlebox/

Strongbox / Throttlebox affiliate program:
http://www.bettercgi.com/affiliates/user/register.php


On 10/30/2009 03:52:43 AM, Abraham P�rez wrote:
> Thanks for the instant answers!
> 
> Well... I'll try to explain myself better. I'm working in a client  
> who have
> a lot of servers running under VMware. This client have a lot of  
> problems
> with the storage, because they never have enough space so when they  
> have to
> allocate disk in servers, they add small virtual hard disks with, for
> example, 5 or 10GB.
> 
> Then for the OS installation, we follow the basic schema based on disk
> partitions (/dev/sda1 pointing to / with ext3, /dev/sda2 pointing to  
> /home
> and so on) and for the applications data, we use VG and LV pointing  
> to /opt
> 
> The client have some applications who need a lot of mountpoints, so my
> colleague adds 1-3 LV per VG (aproximated) and I only create only one  
> VG and
> inside it, different LVs.  With this infrastructure, we have to  
> discard
> different kinds of hard disk because they're exactly the same... and  
> we have
> that doubt: what schema is better and why, discarding concept things  
> like a
> volume group was designed to be a group, because we're looking for  
> good
> reasons based in performance of future actions, it's not important...  
> or am
> I mistaken???
> 
> I don't know if I explained myself very well, so thanks all anyway!
> 
> Regards,
> Abraham P�rez
> 
> 2009/10/30 <malahal@us.ibm.com>
> 
> > Ray Morris [support@bettercgi.com] wrote:
> > >     I don't know about a whitepaper, but I can address
> > > your example.
> > >
> > > > he makes one volume group for each logical volume (more or less)
> > >
> > >     If each one has one volume, that's not exactly a volume
> > > GROUP, is it?  If groups and volumes are basically synomous,
> > > he gives up all the benfits of groups.  In fact, he gives
> > > up most of the benefits of logical volumes, since each PV
> > > has to be in one group, and each VG is one LV, you're left
> > > with one LV per PV - might as well just use partitions
> > > directly.
> >
> > I agree, you lose some flexibility but it has some advantage  
> compared to
> > plain partitions without LVM. E.g. he can make a file system larger  
> than
> > any disk with multiple disks in the above LVM (one LV per VG)
> > configuration.  There are other advantages. I am not sure the  
> reason for
> > making only one LV per VG though!
> >
> > Thanks, Malahal.
> >
> > _______________________________________________
> > 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/
> >
> 

------quoted attachment------
> _______________________________________________
> 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] Best Practices deploying LVM

[Abraham Pérez]

[-- Attachment #1: Type: text/plain, Size: 6465 bytes --]

Well, finally my english results a barrier for explain myself (lol).

Anyway, now I have another point of view, so thank you all very much for
your assistance!!!

2009/10/30 Ray Morris <support@bettercgi.com>

>   With your way, you can add a new 2GB disk and use
> half of it for /home, half for /opt, if you wish.
> You can also leave some of it unused and expand
> any LV in the future as needed.  That's one important
> reason why most people use voluem groups as groups -
> contianing several volumes.  Does your colleague know
> of ANY advantage to creating a bunch of different
> groups?   If not, your way wins - it has advantages
> over his way, and his way has no advantage.
>
>
>  we have to discard different kinds of hard disk
>> because they're exactly the same
>>
>
>  I have no idea what this is supposed to mean.
> Different kinds of hard disk are exactly the same?
> If this is supposed to mean "we are not able to
> use different types of drives for different
> partitions", I can understand that.  However,
> for what purpose would you use different types
> of disks?  Perhaps he wants a fast disk for one
> partition, and a large, cheap disk for another
> partition?  If you use two cheap disks in a
> striped LV it's going to be faster AND cheaper
> than the "fast drive" option.  MAYBE if you were
> going to use a super fast RAID array of SSD drives
> for some small amount of data, but not if we're
> talking standard magnetic hard drives.
>
>
>
>  a lot of servers running under VMware. This client
>> have a lot of problems with the storage, because they
>> never have enough space so when they have to allocate
>> disk in servers, they add small virtual hard disks
>> with, for example, 5 or 10GB.
>>
>
> lvextend.  Ours resize automatically on the fly, using
> a cron job that checks to see if any virtual servers
> need more space.
>
>
>  discarding concept things like a volume group was designed
>> to be a group, because we're looking for good reasons
>>
>
>  "Concept reasons", like using the tool designed for the
> job, may be the very best reasons because that one reason
> actually covers the hundred reasons that don't come to mind
> immediately.  You don't know what issues you'll run into
> next week or next month, but you can bet you won't be the
> first one - other people will have had the same issues,
> and will use the standard tools for the standard purposes
> to solve that problem.  Better for you if you can use
> the same solutions.  Also, there are certain features
> and optimizations you don't know about, but you'll gain
> from those grouping features if you use groups as groups.
> No one knows about the features and optimizations that
> will be added next year, but if you use the tools the
> way they were designed you'll benefit from future
> enhancements that allow you to better use them for their
> purposes.
> --
>
> Ray Morris
> support@bettercgi.com
>
> Strongbox - The next generation in site security:
> http://www.bettercgi.com/strongbox/
>
> Throttlebox - Intelligent Bandwidth Control
> http://www.bettercgi.com/throttlebox/
>
> Strongbox / Throttlebox affiliate program:
> http://www.bettercgi.com/affiliates/user/register.php
>
>
>
> On 10/30/2009 03:52:43 AM, Abraham Pérez wrote:
>
>> Thanks for the instant answers!
>>
>> Well... I'll try to explain myself better. I'm working in a client who
>> have
>> a lot of servers running under VMware. This client have a lot of problems
>> with the storage, because they never have enough space so when they have
>> to
>> allocate disk in servers, they add small virtual hard disks with, for
>> example, 5 or 10GB.
>>
>> Then for the OS installation, we follow the basic schema based on disk
>> partitions (/dev/sda1 pointing to / with ext3, /dev/sda2 pointing to /home
>> and so on) and for the applications data, we use VG and LV pointing to
>> /opt
>>
>> The client have some applications who need a lot of mountpoints, so my
>> colleague adds 1-3 LV per VG (aproximated) and I only create only one VG
>> and
>> inside it, different LVs.  With this infrastructure, we have to discard
>> different kinds of hard disk because they're exactly the same... and we
>> have
>> that doubt: what schema is better and why, discarding concept things like
>> a
>> volume group was designed to be a group, because we're looking for good
>> reasons based in performance of future actions, it's not important... or
>> am
>> I mistaken???
>>
>> I don't know if I explained myself very well, so thanks all anyway!
>>
>> Regards,
>> Abraham Pérez
>>
>> 2009/10/30 <malahal@us.ibm.com>
>>
>> > Ray Morris [support@bettercgi.com] wrote:
>> > >     I don't know about a whitepaper, but I can address
>> > > your example.
>> > >
>> > > > he makes one volume group for each logical volume (more or less)
>> > >
>> > >     If each one has one volume, that's not exactly a volume
>> > > GROUP, is it?  If groups and volumes are basically synomous,
>> > > he gives up all the benfits of groups.  In fact, he gives
>> > > up most of the benefits of logical volumes, since each PV
>> > > has to be in one group, and each VG is one LV, you're left
>> > > with one LV per PV - might as well just use partitions
>> > > directly.
>> >
>> > I agree, you lose some flexibility but it has some advantage compared to
>> > plain partitions without LVM. E.g. he can make a file system larger than
>> > any disk with multiple disks in the above LVM (one LV per VG)
>> > configuration.  There are other advantages. I am not sure the reason for
>> > making only one LV per VG though!
>> >
>> > Thanks, Malahal.
>> >
>> > _______________________________________________
>> > 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/
>> >
>>
>>
> ------quoted attachment------
>
>  _______________________________________________
>> 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/
>>
>
>
> _______________________________________________
> 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/
>

[-- Attachment #2: Type: text/html, Size: 9043 bytes --]