> From: zfs-discuss-boun...@opensolaris.org [mailto:zfs-discuss-
> boun...@opensolaris.org] On Behalf Of Jim Klimov
> Thanks... but doesn't your description imply that the sync writes
> would always be written twice? 

That is correct, regardless of whether you have slog or not.  In the case of 
slog, it gets written first to a dedicated device, and then to the pool.  In 
the case of no-slog, it gets written to dedicated ZIL recyclable blocks in the 
main pool, and then it gets written to the main pool again as part of the next 

All of this is true, except when the sync write is sufficiently large.  When 
it's larger than a configurable threshold (I forget which parameter, but I 
could look it up) then it goes directly to the main pool and skips the ZIL 
completely.  I don't know how exactly that is implemented - Maybe it goes into 
the next TXG and simply forces the next TXG to flush immediately.
zfs-discuss mailing list

Reply via email to