Re: slow USB vtapes

Am 01.06.22 um 12:04 schrieb Jose M Calhariz:

> How many DLEs and how many run in parallel (inparallel parameter)?

inparallel 4

basically 3 linux hosts involved, 20 DLEs, some of them via amsamba

I even have to encrypt things ...

Maybe I should decrease inparallel, yes

And only flush after the DLEs are on disk.

> How fast is your holdingdisk?

rather slow. 4x 5400 rpm drives in a HW RAID1 -> /dev/sda

> What "iostat -k -x 2" command gives when amanda is running?

*thanks* for that pointer

It tells that /dev/sda is 100% utilized.

So that seems to be the bottleneck.

Especially as there is a rsnapshot-cronjob also on that box. This is 
quite heavy for that CPU and the storage.

Right now as I disabled rsnapshot, I was able to flush a 200GB DLE 
successfully .. and in acceptable time:

43 minutes

82753.8 KB/s

Better that multiple hours each night. Much better.

-

changes today:

* changed blocksize from 4m to 1m .. maybe I can go back here

* downgraded to 3.5.1 (just to minimize the moving parts)

* made sure the holding disk is mounted with "noatime"

changes planned:

* bigger and faster disk(s) for the holding disk

* split off rsnapshot-ting: either move that to another server, or at 
least make sure it never runs in parallel to amdumps etc

I test an rsnapshot run now and also watch iostat ... same bottleneck: 
/dev/sda