File tree Expand file tree Collapse file tree 2 files changed +25
-0
lines changed
Expand file tree Collapse file tree 2 files changed +25
-0
lines changed Original file line number Diff line number Diff line change @@ -2890,6 +2890,19 @@ I/O depth
28902890 if none of I/O has been completed yet, we will NOT wait and immediately exit
28912891 the system call. In this example we simply do polling.
28922892
2893+ .. option :: iodepth_batch_complete_omit=int
2894+
2895+ The number of in-flight IOs that need not be retrieved on quiescing. It
2896+ defaults to 0 which means all in-flight IOs will be retrieved until completion
2897+ as not to skew the latency. After that, if in the rate-limiting context, the
2898+ next IO will be set after a fixed delay. But in some cases, the in-flight IOs
2899+ may need be hanging for a while to merge more subsequent IOs in order to get
2900+ a better bandwidth or reach the EC stripe length, and so on. And the hanging
2901+ time may delay the next IO for long if all in-flight IOs must be retrieved.
2902+ Therefore, this option will be used to skip quiescing partially(IOs in merge
2903+ waiting) for a smooth data flow, and as such, it is not suitable for the
2904+ latency-sensitive scenarios.
2905+
28932906.. option :: iodepth_low=int
28942907
28952908 The low water mark indicating when to start filling the queue
Original file line number Diff line number Diff line change @@ -2650,6 +2650,18 @@ if none of I/O has been completed yet, we will NOT wait and immediately exit
26502650the system call. In this example we simply do polling.
26512651.RE
26522652.TP
2653+ .BI iodepth_batch_complete_omit \fR =\fP int
2654+ The number of in-flight IOs that need not be retrieved on quiescing. It
2655+ defaults to 0 which means all in-flight IOs will be retrieved until completion
2656+ as not to skew the latency. After that, if in the rate-limiting context, the
2657+ next IO will be set after a fixed delay. But in some cases, the in-flight IOs
2658+ may need be hanging for a while to merge more subsequent IOs in order to get
2659+ a better bandwidth or reach the EC stripe length, and so on. And the hanging
2660+ time may delay the next IO for long if all in-flight IOs must be retrieved.
2661+ Therefore, this option will be used to skip quiescing partially(IOs in merge
2662+ waiting) for a smooth data flow, and as such, it is not suitable for the
2663+ latency-sensitive scenarios.
2664+ .TP
26532665.BI iodepth_low \fR =\fP int
26542666The low water mark indicating when to start filling the queue
26552667again. Defaults to the same as \fB iodepth \fR , meaning that fio will
You can’t perform that action at this time.
0 commit comments