View Issue Details
|ID||Project||Category||View Status||Date Submitted||Last Update|
|0016996||CentOS-7||kernel||public||2020-02-01 03:16||2020-02-01 03:16|
|Target Version||Fixed in Version|
|Summary||0016996: SSD performance drop if FIO io queue depth bigger than SSD recommanded arbitration burst.|
|Description||for SSD sequential read, base on SSD round robin arbitration, if host send io command size bigger than recommanded arbitration burst in each io queue, then will cause device handled LBA sequence is not sequential, thus, device sequential read performance will drop.|
in Centos7, nvme driver can fetch recommanded arbitration burst value with SSD identify device command, but this value is not referenced in kernel, which may be caused by blk-mq feature porting.
so when we test SSD performance with FIO, if io queue depth is bigger than device recommanded arbitration burst, performan is not stable.
|Steps To Reproduce||install nvme-cli and FIO|
fetch SSD recommanded arbitration burst value with nvme-cli
test SSD sequential read performance with io queue depth bigger than recommanded arbitration burst(libaio engine)
|Tags||No tags attached.|