Re: [PATCH v2] xfs: enable WQ_MEM_RECLAIM on m_sync_workqueue

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Sat, Jul 06, 2024 at 10:32:52AM +1000, NeilBrown wrote:
> Would it be reasonable for the partial pages to be written over RPC and
> for only full pages to be sent directly to the server-side file using
> O_DIRECT writes?  Presumably the benefits of localio are most pronounced
> with large writes which will mostly be full-page, or even full-folio.

Yes.  Note that we already have infrastructure to always read the
entire page for pnfs block (PNFS_READ_WHOLE_PAGE /
pnfs_ld_read_whole_page) to make this a lot more common at least for
reads.

> O_DIRECT writes on the NFS side would be more awkward.  open(2)
> documents that NFS places no alignment restrictions on O_DIRECT I/O.  If
> applications depend on that then some copying will have to be done
> before the data is written to a block filesystem - possibly into the
> page cache, possibly into some other buffer.  This wouldn't be more
> copying that we already do.

Yes.  There is precedence for that in pNFS as well:
fs/nfs/blocklayout/blocklayout.c:is_aligned_req().





[Index of Archives]     [XFS Filesystem Development (older mail)]     [Linux Filesystem Development]     [Linux Audio Users]     [Yosemite Trails]     [Linux Kernel]     [Linux RAID]     [Linux SCSI]


  Powered by Linux