[zfs-discuss] txg_sync is blocked when copying data to an iSCSI disk

Gregor Kopka (@zfs-discuss) zfs-discuss at kopka.net
Thu Aug 4 16:12:48 EDT 2016



Am 03.08.2016 um 10:57 schrieb Sebastian Bayer:
>
>     # Create pool
>     zpool create -o ashift=12 tank
>     /dev/disk/by-id/scsi-3600140519581e55ec004cbb80c32784d
>     # Some tuning
>     zfs set sync=disabled tank
>     zfs set recordsize=1M tank
>     zfs set atime=off tank
>     zfs set compression=lz4 tank
>     zfs set xattr=sa tank
>     zfs set logbias=throughput
>
>
> With those settings I achieve about 35 - 40MB/s. Good enough for my
> purposes and very close to the ext4 speeds. Do you have any comments
> on those settings? Is anything considered unsafe, given that I work on
> a block device exported from some hardware raid?
sync=disabled can lead to data loss of txg_timeout seconds in case the
system crashes.
recordsize=1M is good for having big files which are written once in one
go, never change and are read sequentially.
logbias=throughput dosn't matter since sync=disabled turns every write
into an async one, effectively disabling the ZIL.

Should the backend be classic RAID and you care about the data you
should think about either giving the pool redundancy or setting copies>1
to give ZFS a chance to repair dud data should the iSCSI device decide
to deliver garbage.

Gregor

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.zfsonlinux.org/pipermail/zfs-discuss/attachments/20160804/a66ffc1d/attachment.html>


More information about the zfs-discuss mailing list