juicefs performs much more worse than cephfs in scenario of buffer+end_fsync writing mode

[liujiangang01]

Test Consulsion:
using same backend storage pool（ceph pool）, when execute buffer + end_fsync write fio test, results show that performance of cephfs is more better than juicefs.
Test Environment：
juicefs：juicefs + tikv + cephrados （tikv use sata ssd as data disk）
cephfs：cephfs + cephrados（cephfs metadata pool use sata ssd）
anotation：juicefs and cephfs use the same ceph pool
Test Result：


Thinking：
In the scenario of large file rand writing，we always use mode of buffer and end_fsync to write，but the result show that juicefs performs much more worse than cephfs ，"juicefs client made some trade-offs sacrificing the performance in this scenario？