diff options
author | Matias Elo <matias.elo@nokia.com> | 2018-12-03 14:01:25 +0200 |
---|---|---|
committer | Maxim Uvarov <maxim.uvarov@linaro.org> | 2018-12-13 12:30:18 +0300 |
commit | 20c1f922a47c91d2bff0e8822a18c1ad0f4b3b43 (patch) | |
tree | a6ea4bb3b9b8ca47350e1dc4eef3d030f91233de /config | |
parent | f59cd5efc9fd3b64e9ca16a5850f6d406c31e454 (diff) |
Port 52d54fae5 "linux-gen: sched: add spread weight config file option"
Signed-off-by: Matias Elo <matias.elo@nokia.com>
Reviewed-by: Bill Fischofer <bill.fischofer@linaro.org>
Signed-off-by: Maxim Uvarov <maxim.uvarov@linaro.org>
Diffstat (limited to 'config')
-rw-r--r-- | config/odp-linux-dpdk.conf | 26 |
1 files changed, 20 insertions, 6 deletions
diff --git a/config/odp-linux-dpdk.conf b/config/odp-linux-dpdk.conf index 6566b4767..01523d423 100644 --- a/config/odp-linux-dpdk.conf +++ b/config/odp-linux-dpdk.conf @@ -16,7 +16,7 @@ # Mandatory fields odp_implementation = "linux-dpdk" -config_file_version = "0.1.1" +config_file_version = "0.1.2" # Pool options pool: { @@ -50,13 +50,27 @@ queue_basic: { } sched_basic: { - # Priority level spread. Each priority level is spread into multiple - # scheduler internal queues. A higher spread value typically improves - # parallelism and thus is better for high thread counts, but causes - # uneven service level for low thread counts. Typically, optimal - # value is the number of threads using the scheduler. + # Priority level spread + # + # Each priority level is spread into multiple scheduler internal queues. + # This value defines the number of those queues. Minimum value is 1. + # Each thread prefers one of the queues over other queues. A higher + # spread value typically improves parallelism and thus is better for + # high thread counts, but causes uneven service level for low thread + # counts. Typically, optimal value is the number of threads using + # the scheduler. prio_spread = 4 + # Weight of the preferred scheduler internal queue + # + # Each thread prefers one of the internal queues over other queues. + # This value controls how many times the preferred queue is polled + # between a poll to another internal queue. Minimum value is 1. A higher + # value typically improves parallelism as threads work mostly on their + # preferred queues, but causes uneven service level for low thread + # counts as non-preferred queues are served less often + prio_spread_weight = 63 + # Burst size configuration per priority. The first array element # represents the highest queue priority. The scheduler tries to get # burst_size_default[prio] events from a queue and stashes those that |