mbox series

[SRU,Bionic,0/1] s390/dasd: reduce the default queue depth and nr of hardware queues (LP: 1852257)

Message ID 1573731982-7370-1-git-send-email-frank.heimes@canonical.com
Headers show
Series s390/dasd: reduce the default queue depth and nr of hardware queues (LP: 1852257) | expand

Message

Frank Heimes Nov. 14, 2019, 11:46 a.m. UTC
Buglink: https://bugs.launchpad.net/bugs/1852257

SRU Justification:

[Impact]

* On s390x systems with a small memory footprint, but large amounts of DASD disks,

* the memory can get depleted (even during installation) and can eventually lead to a situation where the OOM kicks in (followed by even more problems).

* Starting with kernel 4.18 the patch below leads to 90% memory consumption savings per active DASD device.

* The below backport is needed to fix this and get the improvement into bionic's kernel 4.15.

[Fix]

* 3284da34a87ab7a527a593f89bbdaf6debe9e713 3284da3 "s390/dasd: reduce the default queue depth and nr of hardware queues"

* Backport: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1852257/+attachment/5304830/+files/0001-s390-dasd-reduce-the-default-queue-depth-and-nr-of-h.patch

[Test Case]

* Configure a s390x system (z/VM guest or LPAR) with only a bit RAM, but lot's of DASDs devices.

* Now gradually enable more and more DASDs and monitor the memory usage (be sure to exclude shared memory and cache).

* One can notice a difference in mem usage of about 10:1 per activated DASD comparing the current stock 4.15 kernel with a patches kernel 4.15.

* With about less than 1GB memory and 40+ DASD devices one may start to run into an OOM situation w(o the patch.

[Regression Potential] 

* The regression potential can be considered as moderate, since:

* this is purely s390x specific

* it again only affects DASD disk storage (no zFCP/SCSI disk storage)

* and it's again only limited to smaller systems (so more z/VM guests rather than LPARs).

[Other Info]

* A cherry-pick to 4.15 wasn't clean (problem in one line), hence the backport, which applied, compiled and worked fine.

Stefan Haberland (1):
  s390/dasd: reduce the default queue depth and nr of hardware queues

 drivers/s390/block/dasd.c     | 13 +++++++++++--
 drivers/s390/block/dasd_int.h |  8 --------
 2 files changed, 11 insertions(+), 10 deletions(-)

Comments

Khalid Elmously Dec. 2, 2019, 8:16 a.m. UTC | #1
On 2019-11-14 12:46:21 , frank.heimes@canonical.com wrote:
> Buglink: https://bugs.launchpad.net/bugs/1852257
> 
> SRU Justification:
> 
> [Impact]
> 
> * On s390x systems with a small memory footprint, but large amounts of DASD disks,
> 
> * the memory can get depleted (even during installation) and can eventually lead to a situation where the OOM kicks in (followed by even more problems).
> 
> * Starting with kernel 4.18 the patch below leads to 90% memory consumption savings per active DASD device.
> 
> * The below backport is needed to fix this and get the improvement into bionic's kernel 4.15.
> 
> [Fix]
> 
> * 3284da34a87ab7a527a593f89bbdaf6debe9e713 3284da3 "s390/dasd: reduce the default queue depth and nr of hardware queues"
> 
> * Backport: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1852257/+attachment/5304830/+files/0001-s390-dasd-reduce-the-default-queue-depth-and-nr-of-h.patch
> 
> [Test Case]
> 
> * Configure a s390x system (z/VM guest or LPAR) with only a bit RAM, but lot's of DASDs devices.
> 
> * Now gradually enable more and more DASDs and monitor the memory usage (be sure to exclude shared memory and cache).
> 
> * One can notice a difference in mem usage of about 10:1 per activated DASD comparing the current stock 4.15 kernel with a patches kernel 4.15.
> 
> * With about less than 1GB memory and 40+ DASD devices one may start to run into an OOM situation w(o the patch.
> 
> [Regression Potential] 
> 
> * The regression potential can be considered as moderate, since:
> 
> * this is purely s390x specific
> 
> * it again only affects DASD disk storage (no zFCP/SCSI disk storage)
> 
> * and it's again only limited to smaller systems (so more z/VM guests rather than LPARs).
> 
> [Other Info]
> 
> * A cherry-pick to 4.15 wasn't clean (problem in one line), hence the backport, which applied, compiled and worked fine.
> 
> Stefan Haberland (1):
>   s390/dasd: reduce the default queue depth and nr of hardware queues
> 
>  drivers/s390/block/dasd.c     | 13 +++++++++++--
>  drivers/s390/block/dasd_int.h |  8 --------
>  2 files changed, 11 insertions(+), 10 deletions(-)
> 
> -- 
> 2.7.4
> 
> 
> -- 
> kernel-team mailing list
> kernel-team@lists.ubuntu.com
> https://lists.ubuntu.com/mailman/listinfo/kernel-team