On Mon, Mar 25, 2019 at 8:02 PM Christoph Hellwig <hch@xxxxxx> wrote: > > On Mon, Mar 25, 2019 at 07:56:33PM +0200, Amir Goldstein wrote: > > Sure, let's give that a shot. But allow me to stay skeptical, because > > I don't think there is a one-size-fits-all solution. > > If application doesn't need >4K atomicity and xfs imposes file-wide > > read locks, there is bound to exist a workload where ext4 can guaranty > > lower latencies than xfs. > > > > Then again, if we fix rw_semaphore to do a good enough job for my > > workload, I may not care about those worst case workloads... > > Downgrading these long standing guarantees is simply not an option. Right. Not without user opt-in. > > Not quite sure what the I/O pattern of your workload is, but if it > is reads from other regions than you write to you should look into > implementing range locks. Workload is a VM image file (VDI) accessed with random IO pattern over SMB. I am still studying the accurate rd/wr mix patterns that really matter to customers. Thanks, Amir.
