[ovs-dev] [PATCH v6 5/6] dpif-netdev: Change pmd selection order.

[Kevin Traynor]

Up to his point rxqs are sorted by processing cycles they
consumed and assigned to pmds in a round robin manner.

Ian pointed out that on wrap around the most loaded pmd will be
the next one to be assigned an additional rxq and that it would be
better to reverse the pmd order when wraparound occurs.

In other words, change from assigning by rr to assigning in a forward
and reverse cycle through pmds.

Also, now that the algorithm has finalized, document an example.

Suggested-by: Ian Stokes <ian.stokes at intel.com>
Signed-off-by: Kevin Traynor <ktraynor at redhat.com>
---
 Documentation/howto/dpdk.rst | 16 ++++++++++++++++
 lib/dpif-netdev.c            | 27 ++++++++++++++++++++++++++-
 tests/pmd.at                 |  2 +-
 3 files changed, 43 insertions(+), 2 deletions(-)

diff --git a/Documentation/howto/dpdk.rst b/Documentation/howto/dpdk.rst
index a67f3a1..bac51de 100644
--- a/Documentation/howto/dpdk.rst
+++ b/Documentation/howto/dpdk.rst
@@ -124,4 +124,20 @@ will be used where known to assign rxqs to pmd based on a round robin of the
 sorted rxqs.
 
+For example, in the case where here there are 5 rxqs and 3 cores (e.g. 3,7,8)
+available, and the measured usage of core cycles per rxq over the last
+interval is seen to be:
+
+- Queue #0: 30%
+- Queue #1: 80%
+- Queue #3: 60%
+- Queue #4: 70%
+- Queue #5: 10%
+
+The rxqs will be assigned to cores 3,7,8 in the following order:
+
+Core 3: Q1 (80%) |
+Core 7: Q4 (70%) | Q5 (10%)
+core 8: Q3 (60%) | Q0 (30%)
+
 Rxq to pmds assignment takes place whenever there are configuration changes.
 
diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c
index 4b1cb85..0023f19 100644
--- a/lib/dpif-netdev.c
+++ b/lib/dpif-netdev.c
@@ -3288,4 +3288,5 @@ struct rr_numa {
 
     int cur_index;
+    bool idx_inc;
 };
 
@@ -3344,11 +3345,35 @@ rr_numa_list_populate(struct dp_netdev *dp, struct rr_numa_list *rr)
         numa->pmds = xrealloc(numa->pmds, numa->n_pmds * sizeof *numa->pmds);
         numa->pmds[numa->n_pmds - 1] = pmd;
+        /* At least one pmd so initialise curr_idx and idx_inc. */
+        numa->cur_index = 0;
+        numa->idx_inc = true;
     }
 }
 
+/* Returns the next pmd from the numa node in
+ * incrementing or decrementing order. */
 static struct dp_netdev_pmd_thread *
 rr_numa_get_pmd(struct rr_numa *numa)
 {
-    return numa->pmds[numa->cur_index++ % numa->n_pmds];
+    int numa_idx = numa->cur_index;
+
+    if (numa->idx_inc == true) {
+        /* Incrementing through list of pmds. */
+        if (numa->cur_index == numa->n_pmds-1) {
+            /* Reached the last pmd. */
+            numa->idx_inc = false;
+        } else {
+            numa->cur_index++;
+        }
+    } else {
+        /* Decrementing through list of pmds. */
+        if (numa->cur_index == 0) {
+            /* Reached the first pmd. */
+            numa->idx_inc = true;
+        } else {
+            numa->cur_index--;
+        }
+    }
+    return numa->pmds[numa_idx];
 }
 
diff --git a/tests/pmd.at b/tests/pmd.at
index b6732ea..e39a23a 100644
--- a/tests/pmd.at
+++ b/tests/pmd.at
@@ -54,5 +54,5 @@ m4_define([CHECK_PMD_THREADS_CREATED], [
 
 m4_define([SED_NUMA_CORE_PATTERN], ["s/\(numa_id \)[[0-9]]*\( core_id \)[[0-9]]*:/\1<cleared>\2<cleared>:/"])
-m4_define([SED_NUMA_CORE_QUEUE_PATTERN], ["s/\(numa_id \)[[0-9]]*\( core_id \)[[0-9]]*:/\1<cleared>\2<cleared>:/;s/\(queue-id: \)0 2 4 6/\1<cleared>/;s/\(queue-id: \)1 3 5 7/\1<cleared>/"])
+m4_define([SED_NUMA_CORE_QUEUE_PATTERN], ["s/\(numa_id \)[[0-9]]*\( core_id \)[[0-9]]*:/\1<cleared>\2<cleared>:/;s/\(queue-id: \)1 2 5 6/\1<cleared>/;s/\(queue-id: \)0 3 4 7/\1<cleared>/"])
 m4_define([DUMMY_NUMA], [--dummy-numa="0,0,0,0"])
 
-- 
1.8.3.1