Re: [Scst-devel] Strange performance behavior on top of encrypted raid

[Göran B. <goe...@gm...>]

Hello Bart,

Am 10.06.2019 um 17:00 schrieb Bart Van Assche:
> Are you aware that using file I/O implies using the read-ahead and
> page cache mechanisms and that using these mechanisms can cause the
> I/O depth to exceed 1 ?
No I am not aware of this. Read-ahead should work with block-io also but
page cache not as far as I know.
>
> Have you considered to adjust the file I/O readahead and/or page cache
> writeback throttling parameters?
>
Read-ahead of raid device had been tweaked before for best general
performance.
The raid is accessed via iscsi/samba/nfs on 1Gbps network. For the usual
work loads (sequential write and sequential read) it got close to
maximum bandwith ... until I set up dm-crypt.
I did not change any writeback throttling parameters so far. I will try
to dig up some information on this.
But it seems wrong to use file io when windows uses the drive as block
device with its own file system caches.
To be honest I am disappointed as I don't see the cause for the
performance issue and begin to think something might be wrong dm-crypt
implementation.
As last resort I will switch back to unencrypted raid and maybe encrypt
individual logical volumes.

Thanks again for your time

Göran