Hi,
Here is my current wip on the throttle group support.
For the user interface I implemented Stefanha's idea proposed in Stuttgart.
For the throttling algorithm I use a cooperative round robin scheduler.
Classical round robin works with a fixed HZ ticks and it's totaly incompatible
with the throttling algorithm.
So the cooperative round robin scheduler is a way for each block device to
decide
if a pause must be done and a timer be armed and most important of all which
other block device of the group must resume the work once the timer is fired.
The advantages of this algorigthm are:
-only one timer active at a given time (no more cpu usage than regular
throttling)
-no central place didacting the sheduling policy like a didactureship:
we love collaboration isn't it ?:)
-No need to deal with incoming queues to collect requests before scheduling
then with and dispatchs queues
-Compatible with the throttling code with almost no changes
-As you go scheduling
Best regards
Benoît
Benoît Canet (5):
throttle: Extract timers from ThrottleState into a separate
ThrottleTimers structure
throttle: Add throttle group infrastructure.
throttle: Add throttle group infrastructure tests
throttle: Prepare to have multiple timers for one ThrottleState
throttle: Add throttle group support.
block.c | 173 +++++++++++++++++++++++++++++++-----
block/qapi.c | 2 +-
blockdev.c | 16 +++-
hmp.c | 4 +-
include/block/block.h | 2 +-
include/block/block_int.h | 9 +-
include/qemu/throttle-groups.h | 43 +++++++++
include/qemu/throttle.h | 42 ++++++---
qapi/block-core.json | 5 +-
qemu-options.hx | 1 +
qmp-commands.hx | 3 +-
tests/test-throttle.c | 133 ++++++++++++++++++++--------
util/Makefile.objs | 1 +
util/throttle-groups.c | 195 +++++++++++++++++++++++++++++++++++++++++
util/throttle.c | 103 +++++++++++++++-------
15 files changed, 616 insertions(+), 116 deletions(-)
create mode 100644 include/qemu/throttle-groups.h
create mode 100644 util/throttle-groups.c
--
2.1.0.rc1
