duplicity-talk
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Duplicity-talk] Performance issues w/ S3


From: tink
Subject: Re: [Duplicity-talk] Performance issues w/ S3
Date: Wed, 24 Jul 2024 10:46:43 +1200



On Tue, 23 Jul 2024 at 20:53, Thomas Laubrock via Duplicity-talk <duplicity-talk@nongnu.org> wrote:

Hi Tink,

Am 23.07.24 um 04:13 schrieb tink via Duplicity-talk:
On Mon, 22 Jul 2024 at 21:57, Thomas Laubrock via Duplicity-talk <duplicity-talk@nongnu.org> wrote:
Am 21.07.24 um 23:21 schrieb tink via Duplicity-talk:
And without compression and encryption the local version takes 55 seconds.

The compression and encryption is done by feeding the data into GnuPG binary installed on your system.

I suggest following tests:

- use `gpg` and encrypt a file manually and check the timing.

Encrypting a 10GB file (using compression) with GPG took  6m34s (again reading from rust, writing to SSD). 
gpg with no compression takes 37s

OK, there is not much duplicity can do here, as it rely on gpg. Maybe you have any strange gpg setting in your system. Working with gpg-key in opposite with symmetric passphrase may also have an advantage.

IMHO gpg itself has no option to run on multiple cores.
Because of the architecture of duplicity volumes must be created in sequence.

Long story short, if you are able to speed up gpg, you duplicity will get the same advantage.

Upload to AWS seems not to be your bottleneck, but if try `--concurrency=X`. In my env volume creation worked fast enough to feed 2-3 uploads. Keep in mind that parallel upload will pretty fast saturate other limits like network bandwidth or server limits.


Hi Thomas,

I can't quite do `--concurrency` on the server proper yet, haven't managed to get 3.0.0 to behave.
Will keep trying to experiment.

As an aside: using 3.0.0 on a debian 12 VM and testing a 10G transfer I get twice the speed to AWS S3 that I get to Wasabi, which is somewhat worrying.


reply via email to

[Prev in Thread] Current Thread [Next in Thread]