On Fri, 25 Mar 2011 14:20:04 -0700, Venkateswararao Jujjuri <jvrao@xxxxxxxxxxxxxxxxxx> wrote: > On 03/25/2011 04:30 AM, Aneesh Kumar K.V wrote: > > Now that we use write_inode to flush server > > cache related to fid, we don't need tsyncfs. > > This help us to do a more efficient server flush > > for dotu protocol > Why are you singling out dotu only? won't it be applicable to dotl too? > With dotl we can have new operations and so we added tsyncfs. The primary goal is to add an operation that can flush server cache. We hooked that to sync(2) on the client. With dotu we cannot add new operations so we always forced the write on the server in case of dotu to O_SYNC. That is much slower than doing an fsync on write_inode. But whether doing an fsync on write inode is better than doing tsyncfs on sync(2) on client is something i haven't yet measured. Stefan Hajnoczi wants to see some numbers before we push tsyncfs in the server(qemu). We also don't want a kernel release with 9p operation which we may remove later. So the plan now is to get write_inode changes upstream in this merge window and later get numbers against tsyncfs/write_inode -> fsync and add tsyncfs only if we see a benefit. BTW NFS use the write_inode approach. -aneesh -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
