On Wed, May 09, 2012 at 07:12:43AM -0400, Christoph Hellwig wrote: > On Wed, May 09, 2012 at 05:30:16PM +0800, Zheng Liu wrote: > > IMHO, FUA provides a solution that we can get better performance in write > > cache mode when we do many flush operations. If I set the disk to write > > through mode, I won't get this benefit. Am I missing something? > > If you set the disk to write through mode you never have to flush the > cache. So as soonas your number of flushes gets close to the number of > writes it tends to be a clear win - for ATA the tradeoff is even more in > favour of write through because the flush command can't be queued yetin > commonly available standards versions. So if you have a workload that > basically needs to flush out every write you win - if you have workloads > where you have a lot more writes than cache flushes write back mode > wins. Thanks for your explanation. It seems that there still has a problem. If I set the disk to write through mode, I need to modify my application to remove all of flush/sync operations. It is unacceptable for us. > > Currently, the key issue is that we disable FUA detection for SATA disk. > > We almost have no chance to change it because it is too complicated to > > set libata_fua variable when this module is loaded. So why not give > > SATA disk an opportunity to enable this feature? After all, there is a > > lot of SATA disks that support this feature. > > I'm all in favour of your patch, I just wanted to point out that the > argument in the description wasn't quite correct. Thank you. I will fix it. :-) Regards, Zheng -- To unsubscribe from this list: send the line "unsubscribe linux-ide" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html