Re: [pbs-devel] Status of Garbage Collection marks in memory (not atime based)

Re: [pbs-devel] Status of Garbage Collection marks in memory (not atime based)

[Dietmar Maurer]

> We use a CephFS as datastore in our test environment and this is rather slow because of the metadata updates. The in-memory based approach could possibly make a huge difference for CephFS based datastores.

Besides, Proxmox Backup is designed to work with local storage. 
Using any Network based storage create a big amount of additional network 
traffic, and makes no real sense to me. You want(should) backup your Cepfs data 
to Proxmox Backup, not the other way around.

Re: [pbs-devel] Status of Garbage Collection marks in memory (not atime based)

[Marco Gabriel]

>> We use a CephFS as datastore in our test environment and this is rather slow because of the metadata updates. The in-memory based approach could possibly make a huge difference for CephFS based datastores.
> Using any Network based storage create a big amount of additional network
> traffic, and makes no real sense to me. You want(should) backup your Cepfs data
> to Proxmox Backup, not the other way around.
>
While this is totally true in most environments, it prevents a scale-out 
approach where one server is not able to handle everything. That's the 
whole wish behind it.



inett GmbH
>> Ihr IT Systemhaus in  Saarbrücken

Marco Gabriel
Geschäftsführer

t: 0681-410993-0
e: mgabriel@inett.de
www.inett.de

Technische Fragen:
Support E-Mail: support@inett.de
Support Hotline: 0681-410993-42


inett GmbH | Geschäftsführung: Marco Gabriel | Amtsgericht Saarbrücken HRB 16588

Re: [pbs-devel] Status of Garbage Collection marks in memory (not atime based)

[Maximilian Hill]

[-- Attachment #1: Type: text/plain, Size: 835 bytes --]

On Fri, Apr 09, 2021 at 10:13:54AM +0200, Dietmar Maurer wrote:
> > We use a CephFS as datastore in our test environment and this is rather slow because of the metadata updates. The in-memory based approach could possibly make a huge difference for CephFS based datastores.
> 
> Besides, Proxmox Backup is designed to work with local storage. 
> Using any Network based storage create a big amount of additional network 
> traffic, and makes no real sense to me. You want(should) backup your Cepfs data 
> to Proxmox Backup, not the other way around.
> 
Well, if you are already running seperate ceph cluster with erasure coded
pools for other backups, that's not that bad of an idea.
But it shouldn't be the same cluster as the hot production storage,
thats right.

GC jobs cause a lot of writes in the metadata pool.


[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [pbs-devel] Status of Garbage Collection marks in memory (not atime based)

[Dietmar Maurer]

> >> We use a CephFS as datastore in our test environment and this is rather slow because of the metadata updates. The in-memory based approach could possibly make a huge difference for CephFS based datastores.
> > Using any Network based storage create a big amount of additional network
> > traffic, and makes no real sense to me. You want(should) backup your Cepfs data
> > to Proxmox Backup, not the other way around.
> >
> While this is totally true in most environments, it prevents a scale-out 
> approach where one server is not able to handle everything. That's the 
> whole wish behind it.

But the idea behind backup is to have redundant copies of your data at
different physical locations...

Anyways, I got your point. As mentioned in the code, it would be possible to 
optimize that using an in-memory hash.

Re: [pbs-devel] Status of Garbage Collection marks in memory (not atime based)

[Dietmar Maurer]

> on https://git.proxmox.com/?p=proxmox-backup.git;a=blob;f=src/backup.rs;h=cca438818b49724e10594ae2e12302f5fbf021df;hb=HEAD I saw, that besides the currently implemented atime based garbage collection approach, another approach using hashes in memory is under investigation.
> 
> I couldn't find any references besides the lines under that link. Is this investigation still ongoing? Is there any publicly available source?

Nobody is working on that - it is just an idea so far.

[pbs-devel] Status of Garbage Collection marks in memory (not atime based)

[Marco Gabriel - inett GmbH]

[-- Attachment #1: Type: text/plain, Size: 659 bytes --]

Hi,

on https://git.proxmox.com/?p=proxmox-backup.git;a=blob;f=src/backup.rs;h=cca438818b49724e10594ae2e12302f5fbf021df;hb=HEAD I saw, that besides the currently implemented atime based garbage collection approach, another approach using hashes in memory is under investigation.

I couldn't find any references besides the lines under that link. Is this investigation still ongoing? Is there any publicly available source?

We use a CephFS as datastore in our test environment and this is rather slow because of the metadata updates. The in-memory based approach could possibly make a huge difference for CephFS based datastores.

Thanks,
Marco



[-- Attachment #2: Type: text/html, Size: 2454 bytes --]