| Message ID | 20201104120554.944255-1-andrea.righi@canonical.com |
|---|---|
| Headers | show |
| Series | aws: update patch to batch hibernate and resume IO requests | expand |
On 04.11.20 13:05, Andrea Righi wrote: > BugLink: https://bugs.launchpad.net/bugs/1902864 > > [Impact] > > During hibernation and resume the kernel is submitting an individual IO > request for each page of data. In the aws kernel we are using a custom > SAUCE patch to batch IO requests together in order to achieve better > performance. > > Recently a patch designed to achieve the same goal has been applied > upstream. This patch has been acknowledged by Amazon and it has shown a > performance improvement. > > Moreover, this patch looks much cleaner compared to the custom patch > that we are using and it's upstream, so it makes sense to drop the > previous patch and apply this new one. > > [Test case] > > Hibernate + resume and measure the time required to perform these > operations. > > Performance result reported by Amazon: > > One hibernate and resume cycle for 16GB RAM out of 32GB in use takes > around 21 minutes before the change, and 1 minutes after the change on > a system with limited storage IOPS. > > [Fix] > > Apply the following upstream commit: > > 55c4478a8f0ecedc0c1a0c9379380249985c372a ("PM: hibernate: Batch hibernate and resume IO requests") > > Drop the custom aws SAUCE patch: > > 11c3fa3b29722124f5c9122671983614383686db ("UBUNTU: SAUCE: [aws] PM / hibernate: Speed up hibernation by batching requests") > > [Regression potential] > > Upstream patch that allows to drop a custom patch that is doing the same > thing. The only potential regression would be a performance drop, but > according to Amazon's tests and our tests, we didn't notice any > performance regression. Any other kind of regression would be considered > as upstream regressions. > > ---------------------------------------------------------------- > Andrea Righi (1): > Revert "UBUNTU: SAUCE: [aws] PM / hibernate: Speed up hibernation by batching requests" > > Xiaoyi Chen (1): > PM: hibernate: Batch hibernate and resume IO requests > > kernel/power/swap.c | 250 ++++++++++------------------------------------------ > 1 file changed, 49 insertions(+), 201 deletions(-) > > Acked-by: Stefan Bader <stefan.bader@canonical.com>
On Wed, Nov 04, 2020 at 01:05:52PM +0100, Andrea Righi wrote: > BugLink: https://bugs.launchpad.net/bugs/1902864 > > [Impact] > > During hibernation and resume the kernel is submitting an individual IO > request for each page of data. In the aws kernel we are using a custom > SAUCE patch to batch IO requests together in order to achieve better > performance. > > Recently a patch designed to achieve the same goal has been applied > upstream. This patch has been acknowledged by Amazon and it has shown a > performance improvement. > > Moreover, this patch looks much cleaner compared to the custom patch > that we are using and it's upstream, so it makes sense to drop the > previous patch and apply this new one. > > [Test case] > > Hibernate + resume and measure the time required to perform these > operations. > > Performance result reported by Amazon: > > One hibernate and resume cycle for 16GB RAM out of 32GB in use takes > around 21 minutes before the change, and 1 minutes after the change on > a system with limited storage IOPS. > > [Fix] > > Apply the following upstream commit: > > 55c4478a8f0ecedc0c1a0c9379380249985c372a ("PM: hibernate: Batch hibernate and resume IO requests") > > Drop the custom aws SAUCE patch: > > 11c3fa3b29722124f5c9122671983614383686db ("UBUNTU: SAUCE: [aws] PM / hibernate: Speed up hibernation by batching requests") > > [Regression potential] > > Upstream patch that allows to drop a custom patch that is doing the same > thing. The only potential regression would be a performance drop, but > according to Amazon's tests and our tests, we didn't notice any > performance regression. Any other kind of regression would be considered > as upstream regressions. > > ---------------------------------------------------------------- > Andrea Righi (1): > Revert "UBUNTU: SAUCE: [aws] PM / hibernate: Speed up hibernation by batching requests" > > Xiaoyi Chen (1): > PM: hibernate: Batch hibernate and resume IO requests > > kernel/power/swap.c | 250 ++++++++++------------------------------------------ > 1 file changed, 49 insertions(+), 201 deletions(-) Always nice to replace a SAUCE with an upstream cherry-pick. :-) Thanks! Acked-by: William Breathitt Gray <william.gray@canonical.com>
Applied to G/aws/master Thanks, Ian On 2020-11-04 13:05:52 , Andrea Righi wrote: > BugLink: https://bugs.launchpad.net/bugs/1902864 > > [Impact] > > During hibernation and resume the kernel is submitting an individual IO > request for each page of data. In the aws kernel we are using a custom > SAUCE patch to batch IO requests together in order to achieve better > performance. > > Recently a patch designed to achieve the same goal has been applied > upstream. This patch has been acknowledged by Amazon and it has shown a > performance improvement. > > Moreover, this patch looks much cleaner compared to the custom patch > that we are using and it's upstream, so it makes sense to drop the > previous patch and apply this new one. > > [Test case] > > Hibernate + resume and measure the time required to perform these > operations. > > Performance result reported by Amazon: > > One hibernate and resume cycle for 16GB RAM out of 32GB in use takes > around 21 minutes before the change, and 1 minutes after the change on > a system with limited storage IOPS. > > [Fix] > > Apply the following upstream commit: > > 55c4478a8f0ecedc0c1a0c9379380249985c372a ("PM: hibernate: Batch hibernate and resume IO requests") > > Drop the custom aws SAUCE patch: > > 11c3fa3b29722124f5c9122671983614383686db ("UBUNTU: SAUCE: [aws] PM / hibernate: Speed up hibernation by batching requests") > > [Regression potential] > > Upstream patch that allows to drop a custom patch that is doing the same > thing. The only potential regression would be a performance drop, but > according to Amazon's tests and our tests, we didn't notice any > performance regression. Any other kind of regression would be considered > as upstream regressions. > > ---------------------------------------------------------------- > Andrea Righi (1): > Revert "UBUNTU: SAUCE: [aws] PM / hibernate: Speed up hibernation by batching requests" > > Xiaoyi Chen (1): > PM: hibernate: Batch hibernate and resume IO requests > > kernel/power/swap.c | 250 ++++++++++------------------------------------------ > 1 file changed, 49 insertions(+), 201 deletions(-) > > > -- > kernel-team mailing list > kernel-team@lists.ubuntu.com > https://lists.ubuntu.com/mailman/listinfo/kernel-team
BugLink: https://bugs.launchpad.net/bugs/1902864 [Impact] During hibernation and resume the kernel is submitting an individual IO request for each page of data. In the aws kernel we are using a custom SAUCE patch to batch IO requests together in order to achieve better performance. Recently a patch designed to achieve the same goal has been applied upstream. This patch has been acknowledged by Amazon and it has shown a performance improvement. Moreover, this patch looks much cleaner compared to the custom patch that we are using and it's upstream, so it makes sense to drop the previous patch and apply this new one. [Test case] Hibernate + resume and measure the time required to perform these operations. Performance result reported by Amazon: One hibernate and resume cycle for 16GB RAM out of 32GB in use takes around 21 minutes before the change, and 1 minutes after the change on a system with limited storage IOPS. [Fix] Apply the following upstream commit: 55c4478a8f0ecedc0c1a0c9379380249985c372a ("PM: hibernate: Batch hibernate and resume IO requests") Drop the custom aws SAUCE patch: 11c3fa3b29722124f5c9122671983614383686db ("UBUNTU: SAUCE: [aws] PM / hibernate: Speed up hibernation by batching requests") [Regression potential] Upstream patch that allows to drop a custom patch that is doing the same thing. The only potential regression would be a performance drop, but according to Amazon's tests and our tests, we didn't notice any performance regression. Any other kind of regression would be considered as upstream regressions. ---------------------------------------------------------------- Andrea Righi (1): Revert "UBUNTU: SAUCE: [aws] PM / hibernate: Speed up hibernation by batching requests" Xiaoyi Chen (1): PM: hibernate: Batch hibernate and resume IO requests kernel/power/swap.c | 250 ++++++++++------------------------------------------ 1 file changed, 49 insertions(+), 201 deletions(-)