Re: Diag dest
Date: Thu, 13 Jul 2023 00:07:56 -0400
Message-ID: <5139b183-bc2d-c2bc-d926-ea95758d8c62_at_gmail.com>
On 7/12/23 08:54, Ricard Martinez wrote:
> Also having these files in an ACFS allows you to access all files from
> any single node, instead of having to jump between nodes.
> Isn't that something to consider too?
>
>
Well, that depends on the file system. ACFS is a fuse file system (user
mode) ACFS implementation of the normal file functions like open, close,
copy, remove or rename is not put into the kernel or installed into
kernel using a kernel module like nvidia or KVM modules. That makes the
use of the Linux file cache much harder, which means that normal access
will be much slower. Also, ACFS is built on top of ASM, which is a
volume manager. ACFS supports direct and asynchronous access which,
combined with the fact that Linux file cache is not used, means that
every read from the file system is a direct read and cannot be fulfilled
from the FS cache. So, instead of reading into the FS cache, you will be
reading into the user's address space and will be waiting for the
physical disk I/O for each and every read. That isn't a problem if you
put DB files onto ACFS because, with Oracle, you don't want double
buffering. Oracle has phenomenal caching mechanisms so any OS cache or
even general purpose caching programs like memcached will. generally
speaking, perform worse than Oracle RDBMS. and take the necessary
resources from it. Oracle probably has some caching within ASM, but you
should definitely test ACFS against something like GPFS (also a
commercial cluster file system, available on Linux).
Furthermore, file systems with the snapshot ability, like ACFS, tend to need to store changed blocks to the separate disk area. That means that GPFS maybe stupider than ACFS but might also be faster and larger on the same drive. File systems with cache tend to far outperform the ones without, given the normal use (which Oracle RDBMS redo, data and control files are not). The performance of ACFS based $ORACLE_HOME might just be performing worse than having a local $ORACLE_HOME on something like XFS, F2FS or Ext4. The decision depends on your choice between the ease of administration or performance.
-- Mladen Gogala Database Consultant Tel: (347) 321-1217 https://dbwhisperer.wordpress.com -- http://www.freelists.org/webpage/oracle-lReceived on Thu Jul 13 2023 - 06:07:56 CEST