Prod 18c on ACFS

[Ricard Martinez]

Hi list,

In our 12.1c RAC clusters we define 3 Diskgroups (data/fra/redo) on ASM
that have up to 100 DB on them (noncdb). This been working fine for us, but
with 18c we thought about using ACFS in order to be able to use snapshots,
etc. If we keep the same DG structure we will end with 3 mountpoints for
each DB, meaning we will have over 300 mountpoints on the nodes. Not a good
call on my experience as cluster takes ages to stop/start that many acfs.
We can use 1 for data for each DB and a global one for fra/redo, but not
sure if acfs will be happy with 100 dB writing to only one volume. Of
course there is also the I/O impact on going to ACFS, meaning we will lose
around 3% based in our tests. Is anyway running a similar environment using
ACFS? Even if not, but you have some insight will be appreciated.

Regards

[Seth Miller]

Ricard,

It might be worth getting in touch with the ODA team since the ODA uses
ACFS by default. I'm sure they must have done some testing in the area
already.


Seth

Post by Ricard Martinez
Hi list,
In our 12.1c RAC clusters we define 3 Diskgroups (data/fra/redo) on ASM
that have up to 100 DB on them (noncdb). This been working fine for us, but
with 18c we thought about using ACFS in order to be able to use snapshots,
etc. If we keep the same DG structure we will end with 3 mountpoints for
each DB, meaning we will have over 300 mountpoints on the nodes. Not a good
call on my experience as cluster takes ages to stop/start that many acfs.
We can use 1 for data for each DB and a global one for fra/redo, but not
sure if acfs will be happy with 100 dB writing to only one volume. Of
course there is also the I/O impact on going to ACFS, meaning we will lose
around 3% based in our tests. Is anyway running a similar environment using
ACFS? Even if not, but you have some insight will be appreciated.
Regards

[Ricard Martinez]

Thanks, as far I been reading on the 12.1/12.2 ODA config docs seems they
give you the option to create DB on ASM or ACFS now, instead of ACFS as in
11. That makes me wonder if it is because users demanded it or other
reasons. Anyone with ODA can verify?

On 19 Nov 2018 16:01, "Seth Miller" <***@gmail.com> wrote:

Ricard,

It might be worth getting in touch with the ODA team since the ODA uses
ACFS by default. I'm sure they must have done some testing in the area
already.


Seth

Post by Ricard Martinez
Hi list,
In our 12.1c RAC clusters we define 3 Diskgroups (data/fra/redo) on ASM
that have up to 100 DB on them (noncdb). This been working fine for us, but
with 18c we thought about using ACFS in order to be able to use snapshots,
etc. If we keep the same DG structure we will end with 3 mountpoints for
each DB, meaning we will have over 300 mountpoints on the nodes. Not a good
call on my experience as cluster takes ages to stop/start that many acfs.
We can use 1 for data for each DB and a global one for fra/redo, but not
sure if acfs will be happy with 100 dB writing to only one volume. Of
course there is also the I/O impact on going to ACFS, meaning we will lose
around 3% based in our tests. Is anyway running a similar environment using
ACFS? Even if not, but you have some insight will be appreciated.
Regards

[Seth Miller]

There has always been the option of using ASM disk groups directly, but you
are correct that the option of choosing between ASM and ACFS wasn't part of
the templates until 12.1.0.2.

However, I still think that the ODA group would be a good resource to
provide you with performance and scalability numbers. Tammy Bednar is
probably a good place to start.


Seth

Post by Ricard Martinez
Thanks, as far I been reading on the 12.1/12.2 ODA config docs seems they
give you the option to create DB on ASM or ACFS now, instead of ACFS as in
11. That makes me wonder if it is because users demanded it or other
reasons. Anyone with ODA can verify?
Ricard,
It might be worth getting in touch with the ODA team since the ODA uses
ACFS by default. I'm sure they must have done some testing in the area
already.
Seth

Post by Ricard Martinez
Hi list,
In our 12.1c RAC clusters we define 3 Diskgroups (data/fra/redo) on ASM
that have up to 100 DB on them (noncdb). This been working fine for us, but
with 18c we thought about using ACFS in order to be able to use snapshots,
etc. If we keep the same DG structure we will end with 3 mountpoints for
each DB, meaning we will have over 300 mountpoints on the nodes. Not a good
call on my experience as cluster takes ages to stop/start that many acfs.
We can use 1 for data for each DB and a global one for fra/redo, but not
sure if acfs will be happy with 100 dB writing to only one volume. Of
course there is also the I/O impact on going to ACFS, meaning we will lose
around 3% based in our tests. Is anyway running a similar environment using
ACFS? Even if not, but you have some insight will be appreciated.
Regards

[Pete Sharman]

Tammy has moved away from the ODA team and now reports direct to Penny Avril in Database Cloud Services. She told me a month or so ago that Kevin Deihl is stepping into her old role.



Pete



From: oracle-l-***@freelists.org <oracle-l-***@freelists.org> On Behalf Of Seth Miller
Sent: Tuesday, November 20, 2018 03:51 AM
To: ***@gmail.com
Cc: Oracle-L Freelists <oracle-***@freelists.org>
Subject: Re: Prod 18c on ACFS



There has always been the option of using ASM disk groups directly, but you are correct that the option of choosing between ASM and ACFS wasn't part of the templates until 12.1.0.2.



However, I still think that the ODA group would be a good resource to provide you with performance and scalability numbers. Tammy Bednar is probably a good place to start.





Seth



On Mon, Nov 19, 2018 at 10:36 AM Ricard Martinez <***@gmail.com <mailto:***@gmail.com> > wrote:

Thanks, as far I been reading on the 12.1/12.2 ODA config docs seems they give you the option to create DB on ASM or ACFS now, instead of ACFS as in 11. That makes me wonder if it is because users demanded it or other reasons. Anyone with ODA can verify?



On 19 Nov 2018 16:01, "Seth Miller" <***@gmail.com <mailto:***@gmail.com> > wrote:

Ricard,



It might be worth getting in touch with the ODA team since the ODA uses ACFS by default. I'm sure they must have done some testing in the area already.





Seth



On Mon, Nov 19, 2018 at 9:13 AM Ricard Martinez <***@gmail.com <mailto:***@gmail.com> > wrote:

Hi list,



In our 12.1c RAC clusters we define 3 Diskgroups (data/fra/redo) on ASM that have up to 100 DB on them (noncdb). This been working fine for us, but with 18c we thought about using ACFS in order to be able to use snapshots, etc. If we keep the same DG structure we will end with 3 mountpoints for each DB, meaning we will have over 300 mountpoints on the nodes. Not a good call on my experience as cluster takes ages to stop/start that many acfs. We can use 1 for data for each DB and a global one for fra/redo, but not sure if acfs will be happy with 100 dB writing to only one volume. Of course there is also the I/O impact on going to ACFS, meaning we will lose around 3% based in our tests. Is anyway running a similar environment using ACFS? Even if not, but you have some insight will be appreciated.



Regards

[Mladen Gogala]

Hi Ricard!

I've used both and I haven't seen any significant difference in
performance. ACFS performs well and is much easier to use than ASM. I
have a problem with ASM because it is not possible to overwrite a file.
That means that if I have to restore a file, I first have to delete the
existing version. And since ACFS is running on top of ASM, you get all
the perks of balanced disks. You can also do snapshots. There is a
separate license for ACFS but ACFS does support snapshots. You can also
use classic file system utilities like lsof, ls, tar, cpio, cp and many,
many others. You don't need ASMCMD.

Regards

Post by Ricard Martinez
Thanks, as far I been reading on the 12.1/12.2 ODA config docs seems
they give you the option to create DB on ASM or ACFS now, instead of
ACFS as in 11. That makes me wonder if it is because users demanded it
or other reasons. Anyone with ODA can verify?

[Michael McMullen]

I will verify that 12.2 ODA now creates databases as ASM as the default. We went to 12.2 as a patch bundle. You need to add the -acfs flag to the oakcli create database for acfs. The problem is that the ODA deployment was configured for ACFS so there was actually very little ASM space available for pure ASM databases.
I would have thought Oracle would keep ACFS as the default as they seem to be pushing everything to ACFS.
________________________________
From: oracle-l-***@freelists.org <oracle-l-***@freelists.org> on behalf of Ricard Martinez <***@gmail.com>
Sent: November 19, 2018 11:36 AM
To: Seth Miller
Cc: Oracle-L Freelists
Subject: Re: Prod 18c on ACFS

Thanks, as far I been reading on the 12.1/12.2 ODA config docs seems they give you the option to create DB on ASM or ACFS now, instead of ACFS as in 11. That makes me wonder if it is because users demanded it or other reasons. Anyone with ODA can verify?

On 19 Nov 2018 16:01, "Seth Miller" <***@gmail.com<mailto:***@gmail.com>> wrote:
Ricard,

It might be worth getting in touch with the ODA team since the ODA uses ACFS by default. I'm sure they must have done some testing in the area already.


Seth

On Mon, Nov 19, 2018 at 9:13 AM Ricard Martinez <***@gmail.com<mailto:***@gmail.com>> wrote:
Hi list,

In our 12.1c RAC clusters we define 3 Diskgroups (data/fra/redo) on ASM that have up to 100 DB on them (noncdb). This been working fine for us, but with 18c we thought about using ACFS in order to be able to use snapshots, etc. If we keep the same DG structure we will end with 3 mountpoints for each DB, meaning we will have over 300 mountpoints on the nodes. Not a good call on my experience as cluster takes ages to stop/start that many acfs. We can use 1 for data for each DB and a global one for fra/redo, but not sure if acfs will be happy with 100 dB writing to only one volume. Of course there is also the I/O impact on going to ACFS, meaning we will lose around 3% based in our tests. Is anyway running a similar environment using ACFS? Even if not, but you have some insight will be appreciated.

Regards

[Ricard Martinez]

Yes, Oracle pushing ACFS is one of the main reason we thinking on switching
to it, apart for the snap ability for PDBs, but I'm worried about how will
OS cope with only one filesystem for that many DBs if we use the ODA DG
distribution (1 Data mountpoints for each DB, 1 shared for recording/redo
between all DBs).

Post by Michael McMullen
I will verify that 12.2 ODA now creates databases as ASM as the default.
We went to 12.2 as a patch bundle. You need to add the -acfs flag to the
oakcli create database for acfs. The problem is that the ODA deployment was
configured for ACFS so there was actually very little ASM space available
for pure ASM databases.
I would have thought Oracle would keep ACFS as the default as they seem to
be pushing everything to ACFS.
------------------------------
*Sent:* November 19, 2018 11:36 AM
*To:* Seth Miller
*Cc:* Oracle-L Freelists
*Subject:* Re: Prod 18c on ACFS
Thanks, as far I been reading on the 12.1/12.2 ODA config docs seems they
give you the option to create DB on ASM or ACFS now, instead of ACFS as in
11. That makes me wonder if it is because users demanded it or other
reasons. Anyone with ODA can verify?
Ricard,
It might be worth getting in touch with the ODA team since the ODA uses
ACFS by default. I'm sure they must have done some testing in the area
already.
Seth
Hi list,
In our 12.1c RAC clusters we define 3 Diskgroups (data/fra/redo) on ASM
that have up to 100 DB on them (noncdb). This been working fine for us, but
with 18c we thought about using ACFS in order to be able to use snapshots,
etc. If we keep the same DG structure we will end with 3 mountpoints for
each DB, meaning we will have over 300 mountpoints on the nodes. Not a good
call on my experience as cluster takes ages to stop/start that many acfs.
We can use 1 for data for each DB and a global one for fra/redo, but not
sure if acfs will be happy with 100 dB writing to only one volume. Of
course there is also the I/O impact on going to ACFS, meaning we will lose
around 3% based in our tests. Is anyway running a similar environment using
ACFS? Even if not, but you have some insight will be appreciated.
Regards