On 2017/5/25 上午9:12, tang.junhui@xxxxxxxxxx wrote: > > Hello Coly Eric, > > > I find the perfomace is same after testing with `fio` by using 512GB data, > > so 'dd' may be inaccurate, but accutually, I also test it with Ceph > applicatitangon, > > the performance is also promoted after the modification. So whatever, I > think > > the bcache devices would better inherit underlying queue limits to show > the real > > limits to upper layer. You can remove the description of the performace > promotion > > in the patch since it maybe depend on test scenarios and applications. > > Junhui, bcache_device_init() already does something overlapped with blk_queue_stack_limits(), which means your patch introduces redundant code execution. Also blk_queue_stack_limits() may fail, so if you do want to add this function call, you also need to check the return value and handle the failure condition. Currently I don't see benefit from this patch. It is probably I miss something from your patch, please correct me if you think so. Thanks. Coly > Regards > > Tang Junhui > > > ＞＞＞ On 2017/5/23 下午4:14, tang.junhui@xxxxxxxxxx wrote: > > ＞＞＞ ＞ From: "tang.junhui" ＜tang.junhui@xxxxxxxxxx＞ > > ＞＞＞ ＞ > > ＞＞＞ ＞ We'd better let the stacking driver inherit underlying queue > limits, > > ＞＞＞ ＞ so, we can avoid the unnessary mergering in the top layer, and > It leads > > ＞＞＞ ＞ a 10% performance promotion in large sequential IO reading > testing as > > ＞＞＞ ＞ follows: > > ＞＞＞ ＞ > > ＞＞＞ > > ＞＞＞ Hi Junhui, > > ＞＞＞ > > ＞＞＞ I don't get the point why there is unnecessary merging in the upper > > ＞＞＞ layer, and how the unnecessary merging can be avoided by set a > queue limit. > > ＞＞＞ > > ＞＞＞ And from the testing, I see total data size is around 21GB which > is not > > ＞＞＞ convinced enough. Is it possible to use a large enough data size, > e.g. > > ＞＞＞ 512GB or 1TB ? > > ＞＞ > > ＞＞ > > ＞＞Please test with `fio` with directio instead of `dd`. I've seen too > many > > ＞＞times where DD is inaccurate for unknown reasons. > > ＞＞ > > ＞＞Also, this should be tested with your cache detached from bcache. > If it > > ＞＞really does improve performance that would be great. > > ＞＞ > > ＞＞-- > > ＞＞Eric Wheeler > > ＞ > > -- Coly Li -- To unsubscribe from this list: send the line "unsubscribe linux-bcache" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
