On 19/09/2020 14:07, Flavio Leitner wrote: > The EMC is not large enough for current production cases > and they are scaling up, so this change switches over from > EMC to SMC by default, which provides better results. > > The EMC is still available and could be used when only a > few number of flows is used. >
Andrew did perf testing of this a couple of years ago. It could be added as a reference. http://www.openvswitch.org/support/ovscon2018/5/1330-theurer.pdf A lot of benchmarks run with a small number of flows that will drop a lot by default with this. So people may get a surprise when they test the first OVS version that includes it. Checked the logs and we don't currently print out the default EMC/SMC settings, only when they are changed. Perhaps it would be good to add a log for the default state of EMC/SMC on startup. At least it may be a clue for when someone hits this. > Signed-off-by: Flavio Leitner <f...@sysclose.org> > --- > Documentation/topics/dpdk/bridge.rst | 10 +++++----- > NEWS | 3 +++ > lib/dpif-netdev.c | 6 +++--- > tests/pmd.at | 4 ++-- > vswitchd/vswitch.xml | 4 ++-- > 5 files changed, 15 insertions(+), 12 deletions(-) > > diff --git a/Documentation/topics/dpdk/bridge.rst > b/Documentation/topics/dpdk/bridge.rst > index 526d5c959..0d9aed4eb 100644 > --- a/Documentation/topics/dpdk/bridge.rst > +++ b/Documentation/topics/dpdk/bridge.rst > @@ -130,13 +130,13 @@ SMC cache > SMC cache or signature match cache is a new cache level after EMC cache. > The difference between SMC and EMC is SMC only stores a signature of a flow > thus it is much more memory efficient. With same memory space, EMC can store > 8k > -flows while SMC can store 1M flows. When traffic flow count is much larger > than > -EMC size, it is generally beneficial to turn off EMC and turn on SMC. It is > -currently turned off by default. > +flows while SMC can store 1M flows. When traffic flow count is small than Maybe Andrew's results can be used to put a number on 'small', even with YMMV/caveats etc. > +EMC size, it is generally beneficial to turn off SMC and turn on EMC. It is > +currently turned on by default. Not sure the advice should be to turn off the SMC here. I guess it's almost zero overhead to keep the SMC enabled if the flows are cached in the EMC, and if they spill out of the EMC a bit due to collisions or increased flows it may then be beneficial to have the SMC enabled. It might also be helpful initially if the SMC is still enabled at the point where the EMC is first enabled to avoid going to the dpcls, though I didn't check if it would populate the EMC from a hit in the SMC. > > -To turn on SMC:: > +To turn off SMC:: > > - $ ovs-vsctl --no-wait set Open_vSwitch . other_config:smc-enable=true > + $ ovs-vsctl --no-wait set Open_vSwitch . other_config:smc-enable=false > > Datapath Classifier Performance > ------------------------------- > diff --git a/NEWS b/NEWS > index 2f67d5047..5b88937fb 100644 > --- a/NEWS > +++ b/NEWS > @@ -1,6 +1,9 @@ > Post-v2.14.0 > --------------------- > > + - DPDK: > + * The SMC cache is enabled by default. > + * The EMC cache is disabled by default. > > v2.14.0 - 17 Aug 2020 > --------------------- > diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c > index 02df8f11e..0f2bb10fd 100644 > --- a/lib/dpif-netdev.c > +++ b/lib/dpif-netdev.c > @@ -2079,7 +2079,7 @@ port_create(const char *devname, const char *type, > port->netdev = netdev; > port->type = xstrdup(type); > port->sf = NULL; > - port->emc_enabled = true; > + port->emc_enabled = false; > port->need_reconfigure = true; > ovs_mutex_init(&port->txq_used_mutex); > > @@ -4305,7 +4305,7 @@ dpif_netdev_set_config(struct dpif *dpif, const struct > smap *other_config) > } > } > > - bool smc_enable = smap_get_bool(other_config, "smc-enable", false); > + bool smc_enable = smap_get_bool(other_config, "smc-enable", true); > bool cur_smc; > atomic_read_relaxed(&dp->smc_enable_db, &cur_smc); > if (smc_enable != cur_smc) { > @@ -4440,7 +4440,7 @@ dpif_netdev_port_set_config(struct dpif *dpif, > odp_port_t port_no, > struct dp_netdev_port *port; > int error = 0; > const char *affinity_list = smap_get(cfg, "pmd-rxq-affinity"); > - bool emc_enabled = smap_get_bool(cfg, "emc-enable", true); > + bool emc_enabled = smap_get_bool(cfg, "emc-enable", false); > > ovs_mutex_lock(&dp->port_mutex); > error = get_port_by_number(dp, port_no, &port); > diff --git a/tests/pmd.at b/tests/pmd.at > index 5b612f88f..20c179b39 100644 > --- a/tests/pmd.at > +++ b/tests/pmd.at > @@ -238,8 +238,8 @@ pmd thread numa_id <cleared> core_id <cleared>: > packets received: 20 > packet recirculations: 0 > avg. datapath passes per packet: 1.00 > - emc hits: 19 > - smc hits: 0 > + emc hits: 0 > + smc hits: 19 > megaflow hits: 0 > avg. subtable lookups per megaflow hit: 0.00 > miss with success upcall: 1 > diff --git a/vswitchd/vswitch.xml b/vswitchd/vswitch.xml > index 81c84927f..009f0e550 100644 > --- a/vswitchd/vswitch.xml > +++ b/vswitchd/vswitch.xml > @@ -513,7 +513,7 @@ > when flow count is larger than EMC capacity. > </p> > <p> > - Defaults to false but can be changed at any time. > + Defaults to true but can be changed at any time. > </p> > </column> > > @@ -3296,7 +3296,7 @@ ovs-vsctl add-port br0 p0 -- set Interface p0 > type=patch options:peer=p1 \ > key="emc-insert-inv-prob"/> will have effect on this interface. > </p> > <p> > - Defaults to true. > + Defaults to false. > </p> > </column> > </group> > _______________________________________________ dev mailing list d...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-dev
