Date: Fri, 11 Jul 2014 12:18:44 -0700 From: javocado <javocado@gmail.com> To: Daniel Kalchev <daniel@digsys.bg> Cc: FreeBSD Filesystems <freebsd-fs@freebsd.org> Subject: Re: Using 2 SSD's to create a SLOG Message-ID: <CAP1HOmQuy7kZGg1kKW2=yct_3K50mnws7vQ369jNN3FpMczMvw@mail.gmail.com> In-Reply-To: <53BD0AFB.3000909@digsys.bg> References: <CAP1HOmRe5hEts6qA=0ZiNbMt%2B%2BDoyJO1f0fs=Wnx4ceC%2BygWeg@mail.gmail.com> <53BD0AFB.3000909@digsys.bg>
next in thread | previous in thread | raw e-mail | index | archive | help
Thank you for your feedback. We understand that the devices are mirrored and tolerant. To further my understanding, confirm that I should be able to yank one of the SLOG SSD drives and zfs won't care, and I can subsequently replace it (issuing what command?) and zfs will then re-mirror it? And if both drives are removed then only the sync data would be lost? But if the drives are restored, or in the case of a power loss when the server comes back up, the sync data on the SSD's will be processed once zfs starts? On Wed, Jul 9, 2014 at 2:27 AM, Daniel Kalchev <daniel@digsys.bg> wrote: > > On 08.07.14 04:06, javocado wrote: > >> I am hoping to dumbly plug two SSDs onto motherboard SATA ports and just >> hand them over, raw, to ZFS. >> >> > Others already commented how you should setup mirrored SLOG. In addition > to that, because of the nature of SSDs and SLOG, I would recommend the > following: > > The SLOG size does not need to be large, it should only cover several > seconds of your synchronous write throughput -- usually few GB are plenty. > Today's SSDs are much large than needed for SLOG. But, today's SSDs also > suffer severe performance degradation, especially for writing when you fill > them up with data and they need to do garbage collection. Also, most SSDs > have "good performance" only when using an 8GB span, not the whole drive. > All of this only makes sense if the drive has TRIM. FreeBSD already > supports TRIM for ZFS SLOG. Therefore ensure you do TRIM of the entire > drive, then partition it with GPT to only use (say) 8GB for the SLOG. Leave > the rest unallocated -- you will never write there but the drive's > controller will use those blocks as spares for TRIM and this will both > improve performance and make the drive last much longer. Then add both > slices as a mirrored log device to your ZFS pool. > > Daniel > > _______________________________________________ > freebsd-fs@freebsd.org mailing list > http://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?CAP1HOmQuy7kZGg1kKW2=yct_3K50mnws7vQ369jNN3FpMczMvw>