Message ID | 20180703151503.2549-14-josef@toxicpanda.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
On 07/03/18 08:15, Josef Bacik wrote: > From: Josef Bacik <jbacik@fb.com> > > A basic documentation to describe the interface, statistics, and > behavior of io.latency. > > Signed-off-by: Josef Bacik <jbacik@fb.com> > --- > Documentation/admin-guide/cgroup-v2.rst | 79 +++++++++++++++++++++++++++++++++ > 1 file changed, 79 insertions(+) > > diff --git a/Documentation/admin-guide/cgroup-v2.rst b/Documentation/admin-guide/cgroup-v2.rst > index 8a2c52d5c53b..569ce27b85e5 100644 > --- a/Documentation/admin-guide/cgroup-v2.rst > +++ b/Documentation/admin-guide/cgroup-v2.rst > @@ -51,6 +51,9 @@ v1 is available under Documentation/cgroup-v1/. > 5-3. IO > 5-3-1. IO Interface Files > 5-3-2. Writeback > + 5-3-3. IO Latency > + 5-3-3-1. How IO Latency Throttling Works > + 5-3-3-2. IO Latency Interface Files > 5-4. PID > 5-4-1. PID Interface Files > 5-5. Device > @@ -1446,6 +1449,82 @@ writeback as follows. > vm.dirty[_background]_ratio. > > > +IO Latency > +~~~~~~~~~~ > + > +This is a cgroup v2 controller for IO workload protection. You provide a group > +with a latency target, and if the average latency exceeds that target the > +controller will throttle any peers that have a lower latency target than the > +protected workload. > + > +The limits are only applied at the peer level in the hierarchy. This means that > +in the diagram below, only groups A, B, and C will influence each other, and > +groups D and F will influence each other. Group G will influence nobody. > + > + [root] > + / | \ > + A B C > + / \ | > + D F G > + > + > +So the ideal way to configure this is to set io.latency in groups A, B, and C. > +Generally you do not want to set a value lower than the latency your device > +supports. Experiment to find the value that works best for your workload. > +Start at higher than the expected latency for your device and watch the > +total_lat_avg value in io.stat for your workload group to get an idea of the > +latency you see during normal operation. Use this value as a basis for your > +real setting, setting at 10-15% higher than the value in io.stat. > +Experimentation is key here because total_lat_avg is a running total, so is the > +"statistics" portion of "lies, damned lies, and statistics." > + > +How IO Latency Throttling Works > +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ > + > +io.latency is work conserving; so as long as everybody is meeting their latency > +target the controller doesn't do anything. Once a group starts missing its > +target it begins throttling any peer group that has a higher target than itself. > +This throttling takes 2 forms: > + > +- Queue depth throttling. This is the number of outstanding IO's a group is > + allowed to have. We will clamp down relatively quickly, starting at no limit > + and going all the way down to 1 IO at a time. > + > +- Artificial delay induction. There are certain types of IO that cannot be > + throttled without possibly adversely affecting higher priority groups. This > + includes swapping and metadata IO. These types of IO are allowed to occur > + normally, however they are "charged" to the originating group. If the > + originating group is being throttled you will see the use_delay and delay > + fields in io.stat increase. The delay value is how many microseconds that are > + being added to any process that runs in this group. Because this number can > + grow quite large if there is a lot of swapping or metadata IO occurring we > + limit the individual delay events to 1 second at a time. > + > +Once the victimized group starts meeting its latency target again it will start > +unthrottling any peer groups that were throttled previously. If the victimized > +group simply stops doing IO the global counter will unthrottle appropriately. > + > +IO Latency Interface Files > +~~~~~~~~~~~~~~~~~~~~~~~~~~ > + > + io.latency > + This takes a similar format as the other controllers. > + > + "MAJOR:MINOR target=<target time in microseconds" (repeat comment:) in microseconds>" > + > + io.stat > + If the controller is enabled you will see extra stats in io.stat in > + addition to the normal ones. > + > + depth > + This is the current queue depth for the group. > + > + avg_lat > + The running average IO latency for this group in microseconds. > + Running average is generally flawed, but will give an > + administrator a general idea of the overall latency they can > + expect for their workload on the given disk. > + > PID > --- > >
On 03.07.2018 18:15, Josef Bacik wrote: > From: Josef Bacik <jbacik@fb.com> > > A basic documentation to describe the interface, statistics, and > behavior of io.latency. > Request size also has significant effect on latency of following requests. It's worth to notice that smaller max_sectors_kb gives more control over latency. > Signed-off-by: Josef Bacik <jbacik@fb.com> > --- > Documentation/admin-guide/cgroup-v2.rst | 79 +++++++++++++++++++++++++++++++++ > 1 file changed, 79 insertions(+) > > diff --git a/Documentation/admin-guide/cgroup-v2.rst b/Documentation/admin-guide/cgroup-v2.rst > index 8a2c52d5c53b..569ce27b85e5 100644 > --- a/Documentation/admin-guide/cgroup-v2.rst > +++ b/Documentation/admin-guide/cgroup-v2.rst > @@ -51,6 +51,9 @@ v1 is available under Documentation/cgroup-v1/. > 5-3. IO > 5-3-1. IO Interface Files > 5-3-2. Writeback > + 5-3-3. IO Latency > + 5-3-3-1. How IO Latency Throttling Works > + 5-3-3-2. IO Latency Interface Files > 5-4. PID > 5-4-1. PID Interface Files > 5-5. Device > @@ -1446,6 +1449,82 @@ writeback as follows. > vm.dirty[_background]_ratio. > > > +IO Latency > +~~~~~~~~~~ > + > +This is a cgroup v2 controller for IO workload protection. You provide a group > +with a latency target, and if the average latency exceeds that target the > +controller will throttle any peers that have a lower latency target than the > +protected workload. > + > +The limits are only applied at the peer level in the hierarchy. This means that > +in the diagram below, only groups A, B, and C will influence each other, and > +groups D and F will influence each other. Group G will influence nobody. > + > + [root] > + / | \ > + A B C > + / \ | > + D F G > + > + > +So the ideal way to configure this is to set io.latency in groups A, B, and C. > +Generally you do not want to set a value lower than the latency your device > +supports. Experiment to find the value that works best for your workload. > +Start at higher than the expected latency for your device and watch the > +total_lat_avg value in io.stat for your workload group to get an idea of the > +latency you see during normal operation. Use this value as a basis for your > +real setting, setting at 10-15% higher than the value in io.stat. > +Experimentation is key here because total_lat_avg is a running total, so is the > +"statistics" portion of "lies, damned lies, and statistics." > + > +How IO Latency Throttling Works > +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ > + > +io.latency is work conserving; so as long as everybody is meeting their latency > +target the controller doesn't do anything. Once a group starts missing its > +target it begins throttling any peer group that has a higher target than itself. > +This throttling takes 2 forms: > + > +- Queue depth throttling. This is the number of outstanding IO's a group is > + allowed to have. We will clamp down relatively quickly, starting at no limit > + and going all the way down to 1 IO at a time. > + > +- Artificial delay induction. There are certain types of IO that cannot be > + throttled without possibly adversely affecting higher priority groups. This > + includes swapping and metadata IO. These types of IO are allowed to occur > + normally, however they are "charged" to the originating group. If the > + originating group is being throttled you will see the use_delay and delay > + fields in io.stat increase. The delay value is how many microseconds that are > + being added to any process that runs in this group. Because this number can > + grow quite large if there is a lot of swapping or metadata IO occurring we > + limit the individual delay events to 1 second at a time. > + > +Once the victimized group starts meeting its latency target again it will start > +unthrottling any peer groups that were throttled previously. If the victimized > +group simply stops doing IO the global counter will unthrottle appropriately. > + > +IO Latency Interface Files > +~~~~~~~~~~~~~~~~~~~~~~~~~~ > + > + io.latency > + This takes a similar format as the other controllers. > + > + "MAJOR:MINOR target=<target time in microseconds" > + > + io.stat > + If the controller is enabled you will see extra stats in io.stat in > + addition to the normal ones. > + > + depth > + This is the current queue depth for the group. > + > + avg_lat > + The running average IO latency for this group in microseconds. > + Running average is generally flawed, but will give an > + administrator a general idea of the overall latency they can > + expect for their workload on the given disk. > + > PID > --- > >
diff --git a/Documentation/admin-guide/cgroup-v2.rst b/Documentation/admin-guide/cgroup-v2.rst index 8a2c52d5c53b..569ce27b85e5 100644 --- a/Documentation/admin-guide/cgroup-v2.rst +++ b/Documentation/admin-guide/cgroup-v2.rst @@ -51,6 +51,9 @@ v1 is available under Documentation/cgroup-v1/. 5-3. IO 5-3-1. IO Interface Files 5-3-2. Writeback + 5-3-3. IO Latency + 5-3-3-1. How IO Latency Throttling Works + 5-3-3-2. IO Latency Interface Files 5-4. PID 5-4-1. PID Interface Files 5-5. Device @@ -1446,6 +1449,82 @@ writeback as follows. vm.dirty[_background]_ratio. +IO Latency +~~~~~~~~~~ + +This is a cgroup v2 controller for IO workload protection. You provide a group +with a latency target, and if the average latency exceeds that target the +controller will throttle any peers that have a lower latency target than the +protected workload. + +The limits are only applied at the peer level in the hierarchy. This means that +in the diagram below, only groups A, B, and C will influence each other, and +groups D and F will influence each other. Group G will influence nobody. + + [root] + / | \ + A B C + / \ | + D F G + + +So the ideal way to configure this is to set io.latency in groups A, B, and C. +Generally you do not want to set a value lower than the latency your device +supports. Experiment to find the value that works best for your workload. +Start at higher than the expected latency for your device and watch the +total_lat_avg value in io.stat for your workload group to get an idea of the +latency you see during normal operation. Use this value as a basis for your +real setting, setting at 10-15% higher than the value in io.stat. +Experimentation is key here because total_lat_avg is a running total, so is the +"statistics" portion of "lies, damned lies, and statistics." + +How IO Latency Throttling Works +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + +io.latency is work conserving; so as long as everybody is meeting their latency +target the controller doesn't do anything. Once a group starts missing its +target it begins throttling any peer group that has a higher target than itself. +This throttling takes 2 forms: + +- Queue depth throttling. This is the number of outstanding IO's a group is + allowed to have. We will clamp down relatively quickly, starting at no limit + and going all the way down to 1 IO at a time. + +- Artificial delay induction. There are certain types of IO that cannot be + throttled without possibly adversely affecting higher priority groups. This + includes swapping and metadata IO. These types of IO are allowed to occur + normally, however they are "charged" to the originating group. If the + originating group is being throttled you will see the use_delay and delay + fields in io.stat increase. The delay value is how many microseconds that are + being added to any process that runs in this group. Because this number can + grow quite large if there is a lot of swapping or metadata IO occurring we + limit the individual delay events to 1 second at a time. + +Once the victimized group starts meeting its latency target again it will start +unthrottling any peer groups that were throttled previously. If the victimized +group simply stops doing IO the global counter will unthrottle appropriately. + +IO Latency Interface Files +~~~~~~~~~~~~~~~~~~~~~~~~~~ + + io.latency + This takes a similar format as the other controllers. + + "MAJOR:MINOR target=<target time in microseconds" + + io.stat + If the controller is enabled you will see extra stats in io.stat in + addition to the normal ones. + + depth + This is the current queue depth for the group. + + avg_lat + The running average IO latency for this group in microseconds. + Running average is generally flawed, but will give an + administrator a general idea of the overall latency they can + expect for their workload on the given disk. + PID ---