[pbs-devel] applied: [PATCH proxmox-backup] backup/verify: improve speed by sorting chunks by inode

[Thomas Lamprecht]

On 13.04.21 16:35, Dominik Csapak wrote:
> before reading the chunks from disk in the order of the index file,
> stat them first and sort them by inode number.
> 
> this can have a very positive impact on read speed on spinning disks,
> even with the additional stat'ing of the chunks.
> 
> memory footprint should be tolerable, for 1_000_000 chunks
> we need about ~16MiB of memory (Vec of 64bit position + 64bit inode)
> (assuming 4MiB Chunks, such an index would reference 4TiB of data)
> 
> two small benchmarks (single spinner, ext4) here showed an improvement from
> ~430 seconds to ~330 seconds for a 32GiB fixed index
> and from
> ~160 seconds to ~120 seconds for a 10GiB dynamic index
> 
> Signed-off-by: Dominik Csapak <d.csapak at proxmox.com>
> ---
> it would be great if other people could also benchmark this patch on
> different setups a little (in addition to me), to verify or disprove my results
> 
>  src/backup/datastore.rs |  5 +++++
>  src/backup/verify.rs    | 32 +++++++++++++++++++++++++++++---
>  2 files changed, 34 insertions(+), 3 deletions(-)
> 
>

applied this for now, did already so before Fabians feedback.

Actually I had a a slight regression here too, but not as bad as Fabian reported
and also on a plain SSD-backed ext4, where I expected that the overhead of getting
the inodes out weights the advantages for storage which is already good at random IO.

I booted an older test-server and with lots of data and a more complex spinner
setup, lets see what that one reports.

Any how, we could, and probably should, make this a switch very easily, either as a
datastore option, or by checking the underlying storage - the latter is easy for single
disk storage (just check the rotational flag in /sys/block/...) but gets quickly ugly
with zfs/btrfs/... and the special devices they support.

If we further add such optimizations for sync (to remote and tape) then those would also
fall under that option-switch. Admins like to tune and this would give them a knob to
check what's best for a setup of theirs.