Re: [ovs-discuss] What is a l2 pkt microflow
Thank you Ben Is this L2 flow is an entry in one of the kernel megaflow tables ? otherwise if it an entry in the kernel microflow table - then there should be multiple microflow tables, because L2 flow keys consist only of L2 header, while for other flow the keys are l2+l3 headers. i thought that there is only a single microflow table so classification takes O(1). --Sara On Mon, Nov 20, 2017 at 7:38 PM, Ben Pfaff wrote: > On Thu, Nov 16, 2017 at 07:46:05PM +0200, Sara Gittlin wrote: >> I understand that microflow kernel cache entries consist of full set of >> header keys. >> What about a l2 pkt e.g cfm pkt. What are the hash keys for cfm pkt in the >> single microflow table? > > Just the Ethernet header. ___ discuss mailing list disc...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-discuss
Re: [ovs-discuss] OpenFlow “select” group
Hi Ben, Thanks for replying me back! Right now, I don't care that much with performance. I just want to figure out a way to send each packet to a different interface within a group. Your idea to edit the hash function might work, I'll try this out. On 20 November 2017 at 12:37, Ben Pfaff wrote: > On Thu, Nov 16, 2017 at 04:41:07PM -0500, Ronaldo Resende wrote: > > I know that Open vSwitch 2.4 and later + OpenFlow 1.4, by default, hashes > > some headers fields to choose a bucket in a select group. > > > > What if I want to change the way this is being done? What would be the > > right approach to accomplish this? (i.e distribute packets among buckets > > using some king of counter, to create an evenly load balance). > > > > Which codes should I considering poking around? > > You would need to start by figuring out how to implement this at the > datapath level, that is, in the Linux kernel module. The datapath > doesn't currently have any concept that can be used to serialize > packets. (Serializing packets will slow things down because it > inherently prevents parallelism, but maybe you don't care about > performance?) Maybe you could add some kind of datapath action to > increment a counter and store it in dp_hash and then recirculate, > analogous to the way that there's an action to hash fields and > recirculate. > ___ discuss mailing list disc...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-discuss
Re: [ovs-discuss] OpenFlow “select” group
No. On Mon, Nov 20, 2017 at 11:41:26AM -0600, Prabhu wrote: > Just curios to know, does Group selection method support weighted round > robin selection method ? > > > On 11/20/2017 11:37 AM, Ben Pfaff wrote: > >On Thu, Nov 16, 2017 at 04:41:07PM -0500, Ronaldo Resende wrote: > >>I know that Open vSwitch 2.4 and later + OpenFlow 1.4, by default, hashes > >>some headers fields to choose a bucket in a select group. > >> > >>What if I want to change the way this is being done? What would be the > >>right approach to accomplish this? (i.e distribute packets among buckets > >>using some king of counter, to create an evenly load balance). > >> > >>Which codes should I considering poking around? > >You would need to start by figuring out how to implement this at the > >datapath level, that is, in the Linux kernel module. The datapath > >doesn't currently have any concept that can be used to serialize > >packets. (Serializing packets will slow things down because it > >inherently prevents parallelism, but maybe you don't care about > >performance?) Maybe you could add some kind of datapath action to > >increment a counter and store it in dp_hash and then recirculate, > >analogous to the way that there's an action to hash fields and > >recirculate. > >___ > >discuss mailing list > >disc...@openvswitch.org > >https://mail.openvswitch.org/mailman/listinfo/ovs-discuss > > ___ > discuss mailing list > disc...@openvswitch.org > https://mail.openvswitch.org/mailman/listinfo/ovs-discuss ___ discuss mailing list disc...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-discuss
Re: [ovs-discuss] OpenFlow “select” group
Just curios to know, does Group selection method support weighted round robin selection method ? On 11/20/2017 11:37 AM, Ben Pfaff wrote: On Thu, Nov 16, 2017 at 04:41:07PM -0500, Ronaldo Resende wrote: I know that Open vSwitch 2.4 and later + OpenFlow 1.4, by default, hashes some headers fields to choose a bucket in a select group. What if I want to change the way this is being done? What would be the right approach to accomplish this? (i.e distribute packets among buckets using some king of counter, to create an evenly load balance). Which codes should I considering poking around? You would need to start by figuring out how to implement this at the datapath level, that is, in the Linux kernel module. The datapath doesn't currently have any concept that can be used to serialize packets. (Serializing packets will slow things down because it inherently prevents parallelism, but maybe you don't care about performance?) Maybe you could add some kind of datapath action to increment a counter and store it in dp_hash and then recirculate, analogous to the way that there's an action to hash fields and recirculate. ___ discuss mailing list disc...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-discuss ___ discuss mailing list disc...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-discuss
Re: [ovs-discuss] What is a l2 pkt microflow
On Thu, Nov 16, 2017 at 07:46:05PM +0200, Sara Gittlin wrote: > I understand that microflow kernel cache entries consist of full set of > header keys. > What about a l2 pkt e.g cfm pkt. What are the hash keys for cfm pkt in the > single microflow table? Just the Ethernet header. ___ discuss mailing list disc...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-discuss
Re: [ovs-discuss] OpenFlow “select” group
On Thu, Nov 16, 2017 at 04:41:07PM -0500, Ronaldo Resende wrote: > I know that Open vSwitch 2.4 and later + OpenFlow 1.4, by default, hashes > some headers fields to choose a bucket in a select group. > > What if I want to change the way this is being done? What would be the > right approach to accomplish this? (i.e distribute packets among buckets > using some king of counter, to create an evenly load balance). > > Which codes should I considering poking around? You would need to start by figuring out how to implement this at the datapath level, that is, in the Linux kernel module. The datapath doesn't currently have any concept that can be used to serialize packets. (Serializing packets will slow things down because it inherently prevents parallelism, but maybe you don't care about performance?) Maybe you could add some kind of datapath action to increment a counter and store it in dp_hash and then recirculate, analogous to the way that there's an action to hash fields and recirculate. ___ discuss mailing list disc...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-discuss
Re: [ovs-discuss] in_port=local never matches for flows
On Fri, Nov 17, 2017 at 02:16:55PM +, Jan De Landtsheer wrote: > Hello again, > > I'm trying to define a conntrack flow that allows connections from linux > namespace1 to a test namespace, and block everything from the test > namespace to the host > > for that I create a bridge, add a port, send the port into the namespace, > give it an IP. on the host I add an IP ont the local interface of the > bridge: > > ``` > ovs-vsctl add-br test > ovs-vsctl add-port test tst -- set Interface test type=internal That's a curious set of commands. Is 'tst' in the second line a typo? Is "test" in the "set Interface" command a typo? And then, later on, when you match on "local", do you expect that to match on your "tst" interface or on the built-in "test" interface? It is the latter that it will match. ___ discuss mailing list disc...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-discuss