Re: NetBSD block vs raw disk I/O

[Sad Clouds <cryintothebluesky%gmail.com@localhost>]

On Fri, 10 Aug 2018 12:29:04 -0000 (UTC)
mlelstv%serpens.de@localhost (Michael van Elst) wrote:

> cryintothebluesky%gmail.com@localhost (Sad Clouds) writes:
> 
> >Reading from the virtual RAID drive gives throughput of 892MB/sec for
> >raw device, but only 53MB/sec for block device. This is NetBSD-8
> 
> >Any ideas why block device I/O is so abysmal? Is this something
> >specific to NetBSD?
> 
> When using the block device, everything is funneled through the
> classic buffer cache. The buffer size is retrieved from the disklabel
> and defaults to 2kbyte only. Finally, reads are done synchronously
> without significant read-ahead.
> 
> This is specific to NetBSD, but it is hardly used because file I/O
> is done by the virtual memory system, bypassing the classic buffer
> cache. So there is no incentive to improve the situation as it also
> requires a redesign of the buffer cache and some people would rather
> get rid of the block device interface.
> 
> So, just don't use it. In particular not for benchmarks :)

OK, thanks for the info, I was trying to see what NetBSD I/O would be
like on this device if I were to use it for FFS file system, but looks
like reading from block device is not the right way to benchmark it. I
didn't want to format it with FFS since it has Linux installed, which
I use on daily basis.