Date: Wed, 9 Sep 2015 11:45:27 -0400 From: Paul Kraus <paul@kraus-haus.org> To: FreeBSD Questions <freebsd-questions@freebsd.org> Subject: Re: Storage question Message-ID: <C48993B7-EF03-4EBE-B949-601509711EF3@kraus-haus.org> In-Reply-To: <55F04D78.8070508@hiwaay.net> References: <55EF3D23.5060009@hiwaay.net> <20150908220639.20412cbd@gumby.homeunix.com> <55EF5409.8020007@yahoo.com> <55EFC2DA.3020101@hiwaay.net> <08B351DD-AA48-4F30-B0D6-C500D0877FB3@lafn.org> <55F02DC8.7000706@hiwaay.net> <20150909150626.5c3b99e5.freebsd@edvax.de> <55F031A0.40500@hiwaay.net> <20150909145820.c3b48aafad4f70553c1c1fd8@sohara.org> <55F0451A.5080709@hiwaay.net> <20150909160005.d3b84775c3d0748014a871e5@sohara.org> <55F04D78.8070508@hiwaay.net>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sep 9, 2015, at 11:15, William A. Mahaffey III <wam@hiwaay.net> = wrote: > I have heard that filling your zpool is a *BAD* thing, Filling a zpool is _fatal_. With Copy on Write and no free space you = cannot even remove a file. > but it can be for any FS, just maybe a bit worse for ZFS. I create a dummy dataset (filesystem) with a reservation of 1 GB and a = quota of 1 GB and no mount point and call it =93do-not-remove=94 or = =93dnr=94 for short :-) If the zpool is small I make the quota and = reservation 512 MB. If I accidentally fill the zpool, I can then _stop_ = whatever is filling it, remove and reservation from =93dnr=94 and = proceed to remove files. The other thing to remember is that ZFS (without compression) is _less_ = space efficient. The checksums and metadata take up more room than for = UFS. ZFS also has a steep performance drop when you hit a certain % in = use. And that is NOT a fixed number but varies with workload. For my VM = hosts I need to keep the zpool less than 85-90% full or the performance = drops so far that the VMs=92 I/O start timing out. > I am going to study that option a bit more. The online docs all seem = to show swap within the zpool as well, does that work OK, performance = wise ? It would simplify installation, however I am planning to script = that, so a bit of 'extra' effort for separate swap partitions is not an = issue. I have always thought that separate swap partitions directly = kernel managed were the best for swap performance if/when it gets down = to that, no ? Swap on zvol is a bad idea (as you may have already found). The issue, = as best I can tell, is that since ZFS grabs all the RAM it can, and SWAP = is used in low RAM situations, there are times were SWAP is trying to = write to the zvol and ZFS is trying to decrease it=92s RAM usage = (because of memory pressure) so it=92s performance drops. Functionally = it becomes a race state that ends badly. -- Paul Kraus paul@kraus-haus.org
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?C48993B7-EF03-4EBE-B949-601509711EF3>