Re: Ext4: Buffered random writes performance regression with dioread_nolock enabled

On 22/09/19 04:18PM, hazem ahmed mohamed wrote:

Hey Team,

I am sending this e-mail to report a performance regression that’s caused by commit 244adf6426(ext4: make dioread_nolock the default) , I am listing the performance regression symptoms below & our analysis for the reported regression.

Although you did mention you already tested on the latest kernel too and there too you saw a similar performance regression. But no harm in also double checking if you have this patch [1]. This does fixes a similar problem AFAIU.

[1]: https://lore.kernel.org/linux-ext4/20200520133119.1383-1-jack@suse.cz/

-ritesh