Skip site navigation (1)Skip section navigation (2)
Date:      Thu, 07 Feb 2008 10:19:38 -0600
From:      Eric Anderson <anderson@freebsd.org>
To:        "fluffles.net" <bsd@fluffles.net>
Cc:        freebsd-fs@freebsd.org
Subject:   Re: delayed write buffer
Message-ID:  <47AB2F9A.8070909@freebsd.org>
In-Reply-To: <47A95BC0.1060307@fluffles.net>
References:  <47A95BC0.1060307@fluffles.net>

next in thread | previous in thread | raw e-mail | index | archive | help
fluffles.net wrote:
> Hello kind list,
> 
> I was wondering how to tune FreeBSD's VFS write buffer. I would like a
> large amount of RAM (say 500MB out of 1GB) to be reserved or allocated
> to be used as write buffer for my backup NAS system.
> 
> If i understand the mechanism correctly, a "dd if=/dev/zero of=/mnt/fs
> bs=1m count=500" would act as if i used a malloc-backed md device. This
> leads to very nice performance gains, especially in my case because i'm
> using encryption causing throughput to be limited to 22MB/s. But if the
> first 500MB is free, i can mask this limitation and experience a fast
> drive, when writing that is.
> 
> Anyone can point me to the right directions? I tried playing with some
> sysctl vfs (notably the vfs.maxmallocbufspace tunable) but did not
> achieve the desired effect. And yes, i do know a lot of dirty buffers is
> dangerous but my storage setup is redundant enough. Besides i'm just
> curious in this topic. :) Could a regular BIO-FLUSH caused by UFS
> metadata sync be the curlpit?
> 
> Thanks for any assistance!
> Kind greetings,


GEOM cache maybe?  I can't recall what the status is on it, or what it's 
real benefits were.  I recall it being used for some RAID5 stuff somehow 
though.

Or, how about making a gjournal, with the journal being a MD-backed 'disk'?


Eric




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?47AB2F9A.8070909>