If the performance difference between the file system and the raw device is not large in this scenario, then my question is meaningless. But this is not what I can conclude by simply testing two devices. I think your experience will be more reliable. From experience, is there a big performance gap between the two?
According to Jiangpeng's opinion, provide raw devices for each image, such as LVM, on the premise that the performance is greatly improved.
-----Original Message-----
From: Ilya Dryomov <idryomov@xxxxxxxxx>
Sent: Friday, March 4, 2022 6:05 PM
To: Yin, Congmin <congmin.yin@xxxxxxxxx>
Cc: dev <dev@xxxxxxx>; Ma, Jianpeng <jianpeng.ma@xxxxxxxxx>
Subject: Re: consult the performance of creating bdev with file
On Fri, Mar 4, 2022 at 10:04 AM Yin, Congmin <congmin.yin@xxxxxxxxx> wrote:
>
> Hi Ilya,
>
>
>
> Consult a performance question. It is the way of creating bdev that SSD mode of pwl cache is using.
>
> Is it appropriate to create bdev devices by truncate files on the file system? Is there a performance gap compared with directly creating bdev device by using /dev /nvme0n1?
I think the file system is used because each RBD image gets should get its own cache. What would handle partitioning the block device between RBD images otherwise?
Thanks,
Ilya
_______________________________________________
Dev mailing list -- dev@xxxxxxx
To unsubscribe send an email to dev-leave@xxxxxxx
