Date: Wed, 25 Jul 2018 23:23:21 -0600 From: Warner Losh <imp@bsdimp.com> To: Kirk McKusick <mckusick@mckusick.com> Cc: "Mikhail T." <mi+thun@aldan.algebra.com>, fs@freebsd.org Subject: Re: core dumps onto ZFS Message-ID: <CANCZdfpfjAtoyWddZTj-t79L2CHEraX7XLvt5qXzEqG5b86Sgw@mail.gmail.com> In-Reply-To: <201807252259.w6PMxhDt055842@chez.mckusick.com> References: <a3574554-731b-6aeb-b7e1-ec29e79d8a00@aldan.algebra.com> <201807252259.w6PMxhDt055842@chez.mckusick.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Jul 25, 2018 4:56 PM, "Kirk McKusick" <mckusick@mckusick.com> wrote:
Is there any chance that your filesystem was more than 90% full?
Once ZFS gets past 90% full its performance drops precipitously.
As it approaches completely full it locks up for minutes per write
doing huge numbers of writes to its pool in an attempt to consolodate
what little free space it has left into something big enough for
it to use (which it sounds like is happening to you).
Write amplification (the amount of extra writing described here) approaches
1 / free-space as free-space goes to zero... so 1% empty is 10x slower than
10% empty, but that is only 2x slower than 80% full. I say tends because
the more free space you have, the greater the odds you have ready space to
write without GC.
Warner
Kirk McKusick
_______________________________________________
freebsd-fs@freebsd.org mailing list
https://lists.freebsd.org/mailman/listinfo/freebsd-fs
To unsubscribe, send any mail to "freebsd-fs-unsubscribe@freebsd.org"
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CANCZdfpfjAtoyWddZTj-t79L2CHEraX7XLvt5qXzEqG5b86Sgw>