diff options
| author | Ethan Jackson <ethan@nicira.com> | 2013-03-21 19:40:49 -0700 |
|---|---|---|
| committer | Ethan Jackson <ethan@nicira.com> | 2013-04-01 12:22:58 -0700 |
| commit | 37419873b42963ff9da99b7f47cbde39f20f6184 (patch) | |
| tree | aaf29e438cf0f6568dc587737d2e001103b511c2 /ofproto | |
| parent | dffa93198738a0cae64cd2343d53445e3161d92d (diff) | |
ofproto-dpif: Rate limit calls to facet_learn().
In the TCP_CRR benchmark, ovs-vswitchd spends so much time in
update_stats() that it has a significant impact on flow setup
performance. Further work is needed in this area, but for now,
simply rate limiting facet_learn() has a roughly 10% improvement
with complex flow tables.
Signed-off-by: Ethan Jackson <ethan@nicira.com>
Diffstat (limited to 'ofproto')
| -rw-r--r-- | ofproto/ofproto-dpif.c | 10 |
1 files changed, 10 insertions, 0 deletions
diff --git a/ofproto/ofproto-dpif.c b/ofproto/ofproto-dpif.c index 6fe55481..62c7f316 100644 --- a/ofproto/ofproto-dpif.c +++ b/ofproto/ofproto-dpif.c @@ -486,6 +486,8 @@ struct facet { * overhead. (A facet always has at least one subfacet and in the common * case has exactly one subfacet.) */ struct subfacet one_subfacet; + + long long int learn_rl; /* Rate limiter for facet_learn(). */ }; static struct facet *facet_create(struct rule_dpif *, @@ -4363,6 +4365,8 @@ facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash) netflow_flow_init(&facet->nf_flow); netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); + facet->learn_rl = time_msec() + 500; + return facet; } @@ -4439,6 +4443,12 @@ facet_learn(struct facet *facet) struct subfacet, list_node); struct action_xlate_ctx ctx; + if (time_msec() < facet->learn_rl) { + return; + } + + facet->learn_rl = time_msec() + 500; + if (!facet->has_learn && !facet->has_normal && (!facet->has_fin_timeout |