[pve-devel] [PATCH 1/4] RFC: Efficient VM backup for qemu

[Dietmar Maurer]

> >>- mirror copies new data
> >>- backup copies old data (before new data is written).
> 
> ok,So we have a backup of the state of disk at the time of the backup start ?

yes

> Do you keep some blocks maps in memory ? (if yes, what is the memory
> usage with big volume?)

Yes, we use a bitmap, one bit for one cluster (cluster size = 65536 bytes)

1 TB ==> 16MB bitmap

2^40/2^16 = 2^(40-16) = 2^24
 
> >>This can slow down writes. But I guess this is not that bad (currently, we
> run
> >>out of snapshot space if storage is slow)
> >>
> >>> (with snapshot, we can backup slowly, new writes are only going to vm
> storage).
> >>
> >>if you backup slowly, you need much space for snapshots.
> 
> Yes, this is space vs speed.
> I have in mind some high database load, like 5000iops or more, so it need a
> fast backup storage in this case.

I guess that totally depends on the access pattern.

> I the vm does random io, does the backup storage write blocks sequentially ?

yes, sequential writes only :-)

>  (if yes, maybe it can be fast with a slow random io backup storage but with
> high sequential throughput)

Anyways, you can solve that problem easily. All you need is some fast
temporary storage, and you buffer writes by writing to that storage first.
Should be really easy to implement.