RE: AW: Copy-on-write file systems

From: Noveljic Nenad <nenad.noveljic_at_vontobel.ch>
Date: Tue, 5 Sep 2017 18:45:09 +0000
Message-ID: <3708_1504637119_59AEF0BF_3708_8089_1_ECDEF0CC6716EC4596FCBC871F48292AB18F6200_at_ZRH-S231>



Hey Tim,

Thank you for your answers.

Times are changing indeed. In any case, it is wise to evaluate all of the options when facing a major change.

It has been a bit unconventional, though very smart decision to put databases on NAS. You can snapshot/clone databases between servers without moving a single byte across!

What where the arguments against iSCSI for Unix/Linux platforms?

Nenad

Twitter: _at_NenadNoveljic
Home page: http://nenadnoveljic.com/

From: Tim Gorman [mailto:tim.evdbt_at_gmail.com] Sent: Dienstag, 5. September 2017 19:41
To: Noveljic Nenad
Cc: ORACLE-L
Subject: Re: AW: Copy-on-write file systems

Nenad,

Agreed on how solutions evolve over the years. Back in the 1980s I worked on custom-built in-house CRM solutions for medical and travel businesses, but that would be foolish today. In 1994, I helped a large telcomm build a custom in-house general ledger application, which even at the time was raising eyebrows even on the team, asking "why are we doing this?"

Among our best customers are those who have been using ZFS for a long time, know it inside and out, but now prefer to focus on the results of what ZFS does, not the process of getting it to do what they want.

As to your questions...

  1. Are OpenZFS and DxFS fully integrated into the Linux Kernel?
    • DxFS is based on OpenZFS, and Delphix DxOS is based on Illumos, a variant of OpenSolaris
      • DxFS and DxOS are indeed very tightly integrated
  2. Have you seen many Oracle installations running on OpenZFS_at_Linux apart from Delphix?
    • DxOS/DxFS provides block storage for database files from database platforms like Oracle, SQL Server, SAP ASE (Sybase), SAP HANA, IBM DB2, MySQL, PostgreSQL, and full-stack applications like Oracle E-Business Suites
      • Unfortunately, I personally have never worked with OpenZFS on Linux at all (yet)
  3. How do Oracle databases at Delphix access the database files - via NFS or are the file systems locally mounted?
    • Delphix makes database and application files available to database instances and application software via network-attached storage (NAS)
      • For UNIX/Linux platforms, that is indeed NFS; for Windows, that is iSCSI.

Many thanks!

-Tim

On 9/5/17 10:42, Noveljic Nenad wrote:
Tim,

I fully agree that there is more than one way to approach the problem. A self-engineered system is just one option and actually not even the preferred one. However, 10-15 years ago there wasn't any turn-key solution around for the requirements similar to ours. As a consequence, we've been (successfully) engineering our database platforms with the building blocks that were available at the time and developing automation to glue everything together.

I'd be grateful if it'd be possible for you to answer the following questions:

  • Are OpenZFS and DxFS fully integrated into the Linux Kernel?
  • Have you seen many Oracle installations running on OpenZFS_at_Linux apart from Delphix?
  • How do Oracle databases at Delphix access the database files - via NFS or are the file systems locally mounted?

Many thanks,

Nenad

Twitter: _at_NenadNoveljic
Home page: http://nenadnoveljic.com/

P.S. I'm also aware that you have an All Star engineering team at Delphix - Matt Ahrens is not the only ex Solaris engineer there :).

From: Tim Gorman [mailto:tim.evdbt_at_gmail.com] Sent: Dienstag, 5. September 2017 17:39
To: Noveljic Nenad
Cc: ORACLE-L
Subject: Re: AW: Copy-on-write file systems

Nenad,

Thanks for the clarification!

There is more to ZFS than the version supported by Oracle on Solaris, as evidenced HERE<https://en.wikipedia.org/wiki/ZFS#List_of_operating_systems_supporting_ZFS>. There is likely cause for concern due to the layoffs last week, but only for Oracle ZFS products. Please be aware that, alongside Oracle ZFS, there is OpenZFS<https://en.wikipedia.org/wiki/OpenZFS> for Linux, Mac, FreeBSD, and OpenSolaris platforms, and OpenZFS on Linux will survive whatever direction Oracle chooses for its proprietary version.

The other thing to keep in mind is that ZFS is a building block toward solutions, just as Oracle database is a building block toward solutions. For example, nobody builds a custom ERP application in-house anymore, and likewise the same is true with what you're discussing. The use-case you are considering (i.e. dev/test agility) is the primary use-case for which Delphix is designed. Delphix is based upon OpenZFS and one of the original inventors of ZFS (Matt Ahrens) is a principal and founding member of our engineering team. Over the past 7 years, he and this team have significantly enhanced DxFS (based on OpenZFS).

Full disclosure: I work for Delphix and my job is to install/deploy for customers; please consider my responses with that in mind.

Please let me know if you want more information?

Thanks!

-Tim

On 9/3/17 16:15, Noveljic Nenad wrote:
Tim,

I'm apologising for not being precise enough.

The main requirements are related to agility: 1. Fast fallback and rollback in the case of upgrades and tests. 2. Fast provisioning of test and development databases based on a source database.

Both requirements have been achieved by the means of ZFS snapshot, rollback and clone commands.

Furthermore, ZFS has significantly simplified capacity management. What I mean by that is Unix admins provision the space in the pools (zpools), and the space is being allocated to individual ZFSs as the databases are growing. This makes the DBA life easier if there are 20-30 databases in a Solaris container, as it is much cheaper to always keep some spare capacity in the pool to account for an unexpected growth than to micromanage individual databases.

On the other hand, the main disadvantage has been so far that sometimes the quality of new releases left something to be desired for. As a consequence, we've been occasionally spending more time for troubleshooting and performance tuning than we had budgeted for. These are the examples of the problems we've been encountering over the past years:

http://nenadnoveljic.com/blog/arc-resizing-user_reserve_hint_pct/

http://nenadnoveljic.com/blog/solaris-11-3-hang-kernel-object-manager/

I've been using ZFS since it was released in Solaris 10. In my opinion, its benefits far outweigh its disadvantages. The reason I'm thinking about alternatives are the recent layouts of Solaris engineers by the Oracle Corporation.

Finally, let me mention that I'm willing to accept some penalty in performance which is inherent to copy-on-write file systems (see Bart Sjerps' blog post https://bartsjerps.wordpress.com/2013/02/26/zfs-ora-database-fragmentation/ ) in exchange for the features which would help fulfilling the aforementioned agility requirements.

Thanks,

Nenad

Gesendet über BlackBerry Work
(www.blackberry.com<http://www.blackberry.com>) Von: Noveljic Nenad <nenad.noveljic_at_vontobel.ch<mailto:nenad.noveljic_at_vontobel.ch>> Datum Montag, 04. Sep. 2017, 12:14 AM
An: gogala.mladen_at_gmail.com<mailto:gogala.mladen_at_gmail.com> <gogala.mladen_at_gmail.com<mailto:gogala.mladen_at_gmail.com>>, oracle-l_at_freelists.org<mailto:oracle-l_at_freelists.org> <oracle-l_at_freelists.org<mailto:oracle-l_at_freelists.org>> Betreff: AW: Copy-on-write file systems

Hi Mladen,

Thank you for your feedback!

ACFS seems to be a viable alternative to ZFS then.

I'm surprised to hear about good benchmark results of ZFS on Linux. I thought that ZFS has not been integrated into the Linux Kernel, i.e. that the ZFS processes are running in the user space.

As already mentioned, I've been successfully using ZFS from its very beginning, but I've got the impression that its future is uncertain.

Thank you,

Nenad

Gesendet über BlackBerry Work
(www.blackberry.com<http://www.blackberry.com>) Von: Mladen Gogala <gogala.mladen_at_gmail.com<mailto:gogala.mladen_at_gmail.com>> Datum Sonntag, 03. Sep. 2017, 11:54 PM
An: oracle-l_at_freelists.org<mailto:oracle-l_at_freelists.org> <oracle-l_at_freelists.org<mailto:oracle-l_at_freelists.org>> Betreff: Re: Copy-on-write file systems

Not being able to afford a T5 super-cluster, I am playing with ZFS on Linux. It is surprisingly good, no complaints at all. Here are two pages about ZFS beating Ext4 on benchmark:

http://www.phoronix.com/scan.php?page=article&item=ubuntu-xenial-zfs&num=1 https://www.phoronix.com/forums/forum/software/general-linux-open-source/35594-zfs-vs-ext4-zfs-wins

ZFS is a very decent file system.
On 09/03/2017 02:47 PM, Tim Gorman wrote: Nenad,

It is helpful to share those requirements, as it is not useful to make recommendations without them?

What is it that you're trying to accomplish? What did ZFS do right, and at what did it fail?

Thanks!

-Tim

On 9/3/17 04:16, Noveljic Nenad wrote:
Hi,

Is anybody running databases on ACFS or some other copy-on-write file system?

I'd be greatful if you could share your experiences. I'm looking for alternatives to ZFS, which has had fulfilled most of our requirements so far.

Many thanks in advance,

Nenad

Gesendet über BlackBerry Work
(www.blackberry.com<http://www.blackberry.com>)



Please consider the environment before printing this e-mail. Bitte denken Sie an die Umwelt, bevor Sie dieses E-Mail drucken.

Important Notice

This message is intended only for the individual named. It may contain confidential or privileged information. If you are not the named addressee you should in particular not disseminate, distribute, modify or copy this e-mail. Please notify the sender immediately by e-mail, if you have received this message by mistake and delete it from your system.

E-mail transmission may not be secure or error-free as information could be intercepted, corrupted, lost, destroyed, arrive late or incomplete. Also processing of incoming e-mails cannot be guaranteed. All liability of the Vontobel Group and its affiliates for any damages resulting from e-mail use is excluded. You are advised that urgent and time sensitive messages should not be sent by e-mail and if verification is required please request a printed version.

--

Mladen Gogala

Oracle DBA

Tel: (347) 321-1217

Important Notice

This message is intended only for the individual named. It may contain confidential or privileged information. If you are not the named addressee you should in particular not disseminate, distribute, modify or copy this e-mail. Please notify the sender immediately by e-mail, if you have received this message by mistake and delete it from your system.

E-mail transmission may not be secure or error-free as information could be intercepted, corrupted, lost, destroyed, arrive late or incomplete. Also processing of incoming e-mails cannot be guaranteed. All liability of the Vontobel Group and its affiliates for any damages resulting from e-mail use is excluded. You are advised that urgent and time sensitive messages should not be sent by e-mail and if verification is required please request a printed version.

Important Notice

This message is intended only for the individual named. It may contain confidential or privileged information. If you are not the named addressee you should in particular not disseminate, distribute, modify or copy this e-mail. Please notify the sender immediately by e-mail, if you have received this message by mistake and delete it from your system.

E-mail transmission may not be secure or error-free as information could be intercepted, corrupted, lost, destroyed, arrive late or incomplete. Also processing of incoming e-mails cannot be guaranteed. All liability of the Vontobel Group and its affiliates for any damages resulting from e-mail use is excluded. You are advised that urgent and time sensitive messages should not be sent by e-mail and if verification is required please request a printed version.


Important Notice

This message is intended only for the individual named. It may contain confidential or privileged information. If you are not the named addressee you should in particular not disseminate, distribute, modify or copy this e-mail. Please notify the sender immediately by e-mail, if you have received this message by mistake and delete it from your system.

E-mail transmission may not be secure or error-free as information could be intercepted, corrupted, lost, destroyed, arrive late or incomplete. Also processing of incoming e-mails cannot be guaranteed. All liability of the Vontobel Group and its affiliates for any damages resulting from e-mail use is excluded. You are advised that urgent and time sensitive messages should not be sent by e-mail and if verification is required please request a printed version.



<html xmlns="http://www.w3.org/1999/xhtml">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1">
<style type="text/css">p { font-family: Arial;font-size:9pt }</style>
</head>
<body>
<p>
<br>Important Notice</br>
<br>This message is intended only for the individual named. It may contain confidential or privileged information. If you are not the named addressee you should in particular not disseminate, distribute, modify or copy this e-mail. Please notify the sender immediately by e-mail, if you have received this message by mistake and delete it from your system.</br>
<br>E-mail transmission may not be secure or error-free as information could be intercepted, corrupted, lost, destroyed, arrive late or incomplete. Also processing of incoming e-mails cannot be guaranteed. All liability of the Vontobel Group and its affiliates for any damages resulting from e-mail use is excluded. You are advised that urgent and time sensitive messages should not be sent by e-mail and if verification is required please request a printed version.<br/>
</p>
</body>
</html>
-- http://www.freelists.org/webpage/oracle-l
Received on Tue Sep 05 2017 - 20:45:09 CEST

Original text of this message