[pve-devel] [RFC docs 13/13] vzdump: add section about backup fleecing

[Fiona Ebner]

Am 25.01.24 um 19:18 schrieb Dietmar Maurer:
>>>> Stupid question: Wouldn't It be much easier to add a simple IO-buffer
>>>> with limited capacity, implemented inside the RUST backup code?
>>
>> At work, we are running a backup cluster on remote location with hdd , 
>> and a production cluster with super fast nvme,
>> and sometimes I have really big write spikes (in GB/s), so it's
>> impossible for the backup storage or network to handle it without
>> increase latency or saturate link.
>>  
>> So with limited capacity (how much ? in memory ?), I don't think it
>> solve the problem. If the buffer is full, the vm write will hang.
> 
> Ok, I can see the problem...

Yes, exactly. Sure, it could be done with an in-memory buffer. And while
it would help some people, it would help in much fewer scenarios
compared to fleecing, because RAM is almost always more expensive/more
limited than storage space.