On 07/31/2017 09:29 PM, Jianjian Huo wrote: > On Sat, Jul 29, 2017 at 8:34 PM, Mark Nelson <mark.a.nelson@xxxxxxxxx> wrote: >> >> https://drive.google.com/uc?export=download&id=0B2gTBZrkrnpZbE50QUdtZlBxdFU > Thanks for sharing this data, Mark. > From your data of last March, for RBD EC overwrite on NVMe, EC > sequential writes are faster than 3X for all IO sizes including small > 4K/16KB. Is this right? but I am not seeing this on my setup(all nvme > drives, 12 of them per node), in my case EC sequential writes are 2~3 > times slower than 3X. Maybe I have too many drives per node? > FWIW, we've seen EC random writes being 3x to 4x slower than replication in terms of IOPS for a block size of 4kb. Similar setup: 10 NVMe disks per node. Mohamad -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
