Oracle 19c OMS and PMEM

[Anton Gavriliuk]

Hi all

I'm going to test Oracle 19c (19.13) using Oracle Memory Speed File (OMS) system for PMEM devices.

Database has size ~5 TB and I have 2 socket box with 6 TB PMEM total.

So how should I organize 6 TB PMEM space ?

2 x 3TB separated XFS 

2 x 3TB concatenated volume and than XFS

2 x 3TB striped volume and than XFS

.........................

Anton

[ranjit...@gmail.com]

Hi Anton,

thank you for testing OMS.

I would suggest using   2 x 3TB separated XFS. Then create uber files on each separated XFS and mount them as different OMS filesystems.  

[Anton Gavriliuk]

Hi

> I would suggest using   2 x 3TB separated XFS. Then create uber files on each separated XFS and mount them as different OMS filesystems.  

Well, I think that it is not so flexible.

What if I will create LVM concatenate setup for those 2 x 3 TB pmem devices ?

pvcreate /dev/pmem0

pvcreate /dev/pmem1

vgcreate vg_pmem /dev/pmem0 /dev/pmem1

lvcreate <all available free space in the vg> -n lv_pmem vg_pmem

fdisk for check 2mb align

mkfs.xfs -d extszinherit=512,su=2m,sw=1 -f /dev/vg_pmem/lv_pmem

mount -o dax,lazytime /dev/vg_pmem/lv_pmem <mount point>

Anton

вт, 9 нояб. 2021 г. в 06:07, ranjit...@gmail.com <ranjit...@gmail.com>:

--
You received this message because you are subscribed to the Google Groups "pmem" group.
To unsubscribe from this group and stop receiving emails from it, send an email to pmem+uns...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/pmem/dcf64ecf-db68-4118-ba87-b25750c5b704n%40googlegroups.com.

[Ranjit Noronha]

A striped volume will work; however:

There are two reasons for this. First, it allows us to place different redo log files from a group on two different devices. So a failure of one persistent memory device allows us to get to the other redo log file in the group.

The other reason is that we can distribute different files according to the numa zone. 

Using a striped volume does not permit this flexibility.

Is having a single mount point with OMS handling multiple persistent memory devices directly something desirable?

--

thanks,

--ranjit

[Anton Gavriliuk]

Hi 

> First, it allows us to place different redo log files from a group on two different devices. So a failure of one persistent memory device allows us to get to the other redo log file in the group.

Correct.  But first of all, we want to measure performance without any redundancy.  

All on PMEM and all on NVMe SSD.  If PMEM setup will provide 2+ times in performance, we will think about redundancy/ODG etc.  Otherwise, if PMEM setup will provide only +10% performance vs SSD setup, I don't want to waste time on redundancy setup.  

> The other reason is that we can distribute different files according to the numa zone

If I have two XFS, then I must start thinking about which files should I place on the 1st or 2nd XFS.  Do you have any ideas/recommendations how to distribute all DB files across two XFS ?

> Is having a single mount point with OMS handling multiple persistent memory devices directly something desirable?

I don't know exactly yet.  I'm not Oracle expert.

Anton

вт, 9 нояб. 2021 г. в 15:44, Ranjit Noronha <ranjit...@gmail.com>: