SEO

Hiya experts

I have a client that would like to implement AIX LVM based disk
replication for an Oracle based app over 2 sites about 20miles apart.
Much the same concept as GLVM ( www-03.ibm.com/systems/p/os/aix/
whitepapers/aix_glvm.html ) but without the TCP encapsulation
There is already a good (quick/low latency links) SAN fabric that
spans both sites

Can anyone think of reasons why this is not a good idea? eg situations
where a FS buffer may not flush, jfs2 logs may not sync, MWC log issue
(MWC would be switched to passive mode and mirroring would be
asynchronous)
(No ora resilvering as JFS2 is used not raw devices)

ie Whats wrong with poor mans disk replication from 5.3s LVM? (ora
dataguard, SAN based replications, HACMP/XD and GLVM are all out for
various reasons)

I'm not too worried about lost data (as RPO/RTO are in order of days
as it stands) but what I don't EVER want is to have the VG/PV/JFS2/ORA/
etc in some state I cant syncvg -f or fsck or redo log my way out
of!!!

Would JFS2 snapshots and logshipping provide a safer
alternative? ...or rsync

Thanks in advance for any and all thought time and comments!

On Apr 23, 4:21 am, Chris <ruthe...@gmail.com> wrote:
> Hiya experts
>
> I have a client that would like to implement AIX LVM based disk
> replication for an Oracle based app over 2 sites about 20miles apart.
> Much the same concept as GLVM ( www-03.ibm.com/systems/p/os/aix/
> whitepapers/aix_glvm.html ) but without the TCP encapsulation
> There is already a good (quick/low latency links) SAN fabric that
> spans both sites
>
> Can anyone think of reasons why this is not a good idea? eg situations
> where a FS buffer may not flush, jfs2 logs may not sync, MWC log issue
> (MWC would be switched to passive mode and mirroring would be
> asynchronous)
> (No ora resilvering as JFS2 is used not raw devices)
>
> ie Whats wrong with poor mans disk replication from 5.3s LVM? (ora
> dataguard, SAN based replications, HACMP/XD and GLVM are all out for
> various reasons)
>
> I'm not too worried about lost data (as RPO/RTO are in order of days
> as it stands) but what I don't EVER want is to have the VG/PV/JFS2/ORA/
> etc in some state I cant syncvg -f or fsck or redo log my way out
> of!!!
>
> Would JFS2 snapshots and logshipping provide a safer
> alternative? ...or rsync
>
> Thanks in advance for any and all thought time and comments!

I do not know about GLVM but i would take a lool at GPFS from IBM.

GPFS ( latest version 3.2 )
http://www-03.ibm.com/systems/cluste...pfs/index.html

Oracle on GPFS ( Using GPFS v2.3 but fits also to 3.2 )
http://www-03.ibm.com/servers/enable...b726/zb726.pdf

hth
Hajo

On Apr 23, 12:46 pm, Chris <ruthe...@gmail.com> wrote:
> On Apr 23, 6:21 pm, Hajo Ehlers <serv...@metamodul.com> wrote:
>
>
>
> > On Apr 23, 4:21 am, Chris <ruthe...@gmail.com> wrote:
>
> > > Hiya experts
>
> > > I have a client that would like to implement AIX LVM based disk
> > > replication for an Oracle based app over 2 sites about 20miles apart.
> > > Much the same concept as GLVM ( www-03.ibm.com/systems/p/os/aix/
> > > whitepapers/aix_glvm.html ) but without the TCP encapsulation
> > > There is already a good (quick/low latency links) SAN fabric that
> > > spans both sites
>
> > > Can anyone think of reasons why this is not a good idea? eg situations
> > > where a FS buffer may not flush, jfs2 logs may not sync, MWC log issue
> > > (MWC would be switched to passive mode and mirroring would be
> > > asynchronous)
> > > (No ora resilvering as JFS2 is used not raw devices)
>
> > > ie Whats wrong with poor mans disk replication from 5.3s LVM? (ora
> > > dataguard, SAN based replications, HACMP/XD and GLVM are all out for
> > > various reasons)
>
> > > I'm not too worried about lost data (as RPO/RTO are in order of days
> > > as it stands) but what I don't EVER want is to have the VG/PV/JFS2/ORA/
> > > etc in some state I cant syncvg -f or fsck or redo log my way out
> > > of!!!
>
> > > Would JFS2 snapshots and logshipping provide a safer
> > > alternative? ...or rsync
>
> > > Thanks in advance for any and all thought time and comments!
>
> > I do not know about GLVM but i would take a lool at GPFS from IBM.
>
> > GPFS ( latest version 3.2 )http://www-03.ibm.com/systems/cluste...pfs/index.html
>
> > Oracle on GPFS ( Using GPFS v2.3 but fits also to 3.2 )http://www-03.ibm.com/servers/enable...b726/zb726.pdf
>
> > hth
> > Hajo- Hide quoted text -
>
> > - Show quoted text -
>
> thanks for the links Hajo, we used to have GPFS on the ole' SP2s back
> in the day
>
> That would be nice, it does not look like its free and the customer
> wont spend much to get this working,

I do not think that GPFS is that expensive.

But some thoughts:

Do i assume correctly that at both side you will have SAN storage
connected through a SAN fabric . Thus each side has local san storage
but can access the san storage on the other side as well ?

Is synchronous or asynchronous mode required ?

In case for async i would look for LVM mirroring ( Strict allocation
policy ) and JFS2 snapshots ( Using more then one ) because MWC does
not matter at this point because you can use the latest snapshot
before the crash.
Its also easy since at the DR site the VG just has to be imported
( may a scsi reservation break has to be done ) and the latest
snapshot activated.

Also no OS has to be running at the DR site. This means in case you
are booting from the SAN you can boot the initial OS from the
production side so switching the whole configuration from side to the
other.


In case for synchronous mode i would look at least at GLVM since its
major advantage is that it is using synchronous write. Thus the data
at the DR is up to date all the time. The initial drawback for GLVM is
that it does need a high bandwidth tcpcip connection. In case you have
no high bandwidth network between both side you might be able to
configure your san for "tcpip over fc".

If the later is also not possible i would look at GPFS because since
you have already a san fabric the network between both sides must
handle only gpfs communication traffic. ( No NSD server and other
advanced stuff ).

But for sync mode a OS must be running at the DR side.

But like a said - only some thoughts.

regards
Hajo

On Apr 23, 6:21*pm, Hajo Ehlers <serv...@metamodul.com> wrote:
> On Apr 23, 4:21 am, Chris <ruthe...@gmail.com> wrote:
>
>
>
>
>
> > Hiya experts
>
> > I have a client that would like to implement AIX LVM based disk
> > replication for an Oracle based app over 2 sites about 20miles apart.
> > Much the same concept as GLVM ( www-03.ibm.com/systems/p/os/aix/
> > whitepapers/aix_glvm.html ) but without the TCP encapsulation
> > There is already a good (quick/low latency links) SAN fabric that
> > spans both sites
>
> > Can anyone think of reasons why this is not a good idea? eg situations
> > where a FS buffer may not flush, jfs2 logs may not sync, MWC log issue
> > (MWC would be switched to passive mode and mirroring would be
> > asynchronous)
> > (No ora resilvering as JFS2 is used not raw devices)
>
> > ie Whats wrong with poor mans disk replication from 5.3s LVM? (ora
> > dataguard, SAN based replications, HACMP/XD and GLVM are all out for
> > various reasons)
>
> > I'm not too worried about lost data (as RPO/RTO are in order of days
> > as it stands) but what I don't EVER want is to have the VG/PV/JFS2/ORA/
> > etc in some state I cant syncvg -f or fsck or *redo log my way out
> > of!!!
>
> > Would JFS2 snapshots and logshipping provide a safer
> > alternative? ...or rsync
>
> > Thanks in advance for any and all thought time and comments!
>
> I do not know about GLVM but i would take a lool at GPFS from IBM.
>
> GPFS ( latest version 3.2 )http://www-03.ibm.com/systems/cluste...pfs/index.html
>
> Oracle on GPFS ( Using GPFS v2.3 but fits also to 3.2 )http://www-03.ibm.com/servers/enable...b726/zb726.pdf
>
> hth
> Hajo- Hide quoted text -
>
> - Show quoted text -

thanks for the links Hajo, we used to have GPFS on the ole' SP2s back
in the day

That would be nice, it does not look like its free and the customer
wont spend much to get this working, also I have to retro fit the
existing legacy system with minimal disruption/change

On Apr 23, 10:14 pm, Hajo Ehlers <serv...@metamodul.com> wrote:
> On Apr 23, 12:46 pm, Chris <ruthe...@gmail.com> wrote:
>
>
>
> > On Apr 23, 6:21 pm, Hajo Ehlers <serv...@metamodul.com> wrote:
>
> > > On Apr 23, 4:21 am, Chris <ruthe...@gmail.com> wrote:
>
> > > > Hiya experts
>
> > > > I have a client that would like to implement AIX LVM based disk
> > > > replication for an Oracle based app over 2 sites about 20miles apart.
> > > > Much the same concept as GLVM ( www-03.ibm.com/systems/p/os/aix/
> > > > whitepapers/aix_glvm.html ) but without the TCP encapsulation
> > > > There is already a good (quick/low latency links) SAN fabric that
> > > > spans both sites
>
> > > > Can anyone think of reasons why this is not a good idea? eg situations
> > > > where a FS buffer may not flush, jfs2 logs may not sync, MWC log issue
> > > > (MWC would be switched to passive mode and mirroring would be
> > > > asynchronous)
> > > > (No ora resilvering as JFS2 is used not raw devices)
>
> > > > ie Whats wrong with poor mans disk replication from 5.3s LVM? (ora
> > > > dataguard, SAN based replications, HACMP/XD and GLVM are all out for
> > > > various reasons)
>
> > > > I'm not too worried about lost data (as RPO/RTO are in order of days
> > > > as it stands) but what I don't EVER want is to have the VG/PV/JFS2/ORA/
> > > > etc in some state I cant syncvg -f or fsck or redo log my way out
> > > > of!!!
>
> > > > Would JFS2 snapshots and logshipping provide a safer
> > > > alternative? ...or rsync
>
> > > > Thanks in advance for any and all thought time and comments!
>
> > > I do not know about GLVM but i would take a lool at GPFS from IBM.
>
> > > GPFS ( latest version 3.2 )http://www-03.ibm.com/systems/cluste...pfs/index.html
>
> > > Oracle on GPFS ( Using GPFS v2.3 but fits also to 3.2 )http://www-03.ibm.com/servers/enable...b726/zb726.pdf
>
> > > hth
> > > Hajo- Hide quoted text -
>
> > > - Show quoted text -
>
> > thanks for the links Hajo, we used to have GPFS on the ole' SP2s back
> > in the day
>
> > That would be nice, it does not look like its free and the customer
> > wont spend much to get this working,
>
> I do not think that GPFS is that expensive.
>
> But some thoughts:
>
> Do i assume correctly that at both side you will have SAN storage
> connected through a SAN fabric . Thus each side has local san storage
> but can access the san storage on the other side as well ?
>
> Is synchronous or asynchronous mode required ?
>
> In case for async i would look for LVM mirroring ( Strict allocation
> policy ) and JFS2 snapshots ( Using more then one ) because MWC does
> not matter at this point because you can use the latest snapshot
> before the crash.
> Its also easy since at the DR site the VG just has to be imported
> ( may a scsi reservation break has to be done ) and the latest
> snapshot activated.
>
> Also no OS has to be running at the DR site. This means in case you
> are booting from the SAN you can boot the initial OS from the
> production side so switching the whole configuration from side to the
> other.
>
> In case for synchronous mode i would look at least at GLVM since its
> major advantage is that it is using synchronous write. Thus the data
> at the DR is up to date all the time. The initial drawback for GLVM is
> that it does need a high bandwidth tcpcip connection. In case you have
> no high bandwidth network between both side you might be able to
> configure your san for "tcpip over fc".
>
> If the later is also not possible i would look at GPFS because since
> you have already a san fabric the network between both sides must
> handle only gpfs communication traffic. ( No NSD server and other
> advanced stuff ).
>
> But for sync mode a OS must be running at the DR side.
>
> But like a said - only some thoughts.
>
> regards
> Hajo

Thanks Hajo, food for thought synchronous is out b'cos of performance
on the prd end, this is a temp solution(tm) needs to work for a year
or 2 while the new system is developed with minimal cost,risk or
change to the existing system