| Message ID | 20210805145407.15234-1-tim.gardner@canonical.com |
|---|---|
| Headers | show |
| Series | [Impish:azure] UBUNTU: [Config] azure: CONFIG_RT_GROUP_SCHED=y | expand |
On 05/08/2021 15:54, Tim Gardner wrote: > BugLink: https://bugs.launchpad.net/bugs/1939024 > > [Impact] > > Microsoft has requested that CONFIG_RT_GROUP_SCHED be enabled. This option has > no discernible runtime impact in the default case. As quoted from research > performed by Colin King, "I ran some synthetic stress-ng throughput and > latency tests with and without CONFIG_RT_GROUP_SCHED on CONFIG_RT_GROUP_SCHED. > I don't see enough variation in the throughput or latency to warrant any concern > about enabling this option. We have quite a bit of small jitter on the results, > but overall the kernel results are similar enough to make me feel OK to enable > this option for Focal." > > [Fix] > > CONFIG_RT_GROUP_SCHED=y > > [Test Case] > > Microsoft has tested this option using kernels at > https://kernel.ubuntu.com/~rtg/azure-rt-group-sched-sf00314473/. The customer > requesting this config option verified that it satisfied their requirements. > Microsft also determined that there were no performance regressions for the default > case, i.e., when no RT priority threads were created. > > [Where problems could occur] > > Adding options to the scheduler could impact work loads that we have yet to observe. > > [Other Info] > > SF: 00314473 > > I'm happy with this change. I don't believe it will negatively impact performance. Acked-by: Colin Ian King <colin.king@canonical.com>
On 05.08.21 16:54, Tim Gardner wrote: > BugLink: https://bugs.launchpad.net/bugs/1939024 > > [Impact] > > Microsoft has requested that CONFIG_RT_GROUP_SCHED be enabled. This option has > no discernible runtime impact in the default case. As quoted from research > performed by Colin King, "I ran some synthetic stress-ng throughput and > latency tests with and without CONFIG_RT_GROUP_SCHED on CONFIG_RT_GROUP_SCHED. > I don't see enough variation in the throughput or latency to warrant any concern > about enabling this option. We have quite a bit of small jitter on the results, > but overall the kernel results are similar enough to make me feel OK to enable > this option for Focal." > > [Fix] > > CONFIG_RT_GROUP_SCHED=y > > [Test Case] > > Microsoft has tested this option using kernels at > https://kernel.ubuntu.com/~rtg/azure-rt-group-sched-sf00314473/. The customer > requesting this config option verified that it satisfied their requirements. > Microsft also determined that there were no performance regressions for the default > case, i.e., when no RT priority threads were created. > > [Where problems could occur] > > Adding options to the scheduler could impact work loads that we have yet to observe. > > [Other Info] > > SF: 00314473 > > Just as a comment for Impish. This could be getting complicated as right now it probably is still the forward-port which will change when hirsute changes. But (not knowing when) at some point it is becoming a 5.13 based kernel and then it might need to be remembered individually to apply it. Acked-by: Stefan Bader <stefan.bader@canonical.com>
Applied to Focal azure/master-next, Hirsute azure/master-next. Thanks. Waiting for Impish to transition to 5.13 before setting this config option. -rtg On 8/5/21 8:54 AM, Tim Gardner wrote: > BugLink: https://bugs.launchpad.net/bugs/1939024 > > [Impact] > > Microsoft has requested that CONFIG_RT_GROUP_SCHED be enabled. This option has > no discernible runtime impact in the default case. As quoted from research > performed by Colin King, "I ran some synthetic stress-ng throughput and > latency tests with and without CONFIG_RT_GROUP_SCHED on CONFIG_RT_GROUP_SCHED. > I don't see enough variation in the throughput or latency to warrant any concern > about enabling this option. We have quite a bit of small jitter on the results, > but overall the kernel results are similar enough to make me feel OK to enable > this option for Focal." > > [Fix] > > CONFIG_RT_GROUP_SCHED=y > > [Test Case] > > Microsoft has tested this option using kernels at > https://kernel.ubuntu.com/~rtg/azure-rt-group-sched-sf00314473/. The customer > requesting this config option verified that it satisfied their requirements. > Microsft also determined that there were no performance regressions for the default > case, i.e., when no RT priority threads were created. > > [Where problems could occur] > > Adding options to the scheduler could impact work loads that we have yet to observe. > > [Other Info] > > SF: 00314473 >