On 2022/10/13 22:32, Chaitanya Kulkarni wrote:
On 10/13/22 02:44, Chao Leng wrote:
This set improves the quiesce time when using a large set of
namespaces, which also improves I/O failover time in a multipath environment.
by how much and what are the problems exits with current timing needs to
be documented, i.e. add quantitative data if you are posting patches for
time/performance improvement.
Theoretically, every synchronize_rcu/synchronize_srcu need to wait more than 10ms.
Now nvme quiesce all queues of namespace one by one, The more namespace,
the longer the waiting time.
The test result:
Test Scenario: nvme over roce with multipathing, 256 namespaces
When send I/Os to all namespaces, simulate a path fault, the fail over waiting time
is more than 20 seconds.
After analysis, it was found that the total of quiesce waiting time for 256 namespace is
more than 20 seconds.
After optimization, same scenario the fail over waiting time is less than 1 second.
-ck
