Hello Vinayak > Generally, a service should not send packets to other service > directly. > > They are always multiplexed via dispatchers to avoid too maintain > pipeline simplicity.
Agreed. I guess we did consider at the time to go through dispatcher but found it either not possible or not very reasonable. But yes, let's discuss on wenesday how can it could be improved. BR Jaime. -----Original Message----- From: Vinayak Joshi <[email protected]> To: [email protected] <[email protected]>, jc [email protected] Subject: RE: [sfc-dev] Question on OF pipeline Date: Mon, 7 May 2018 06:34:49 +0000 Hi Brady & Jaimie, Can we discuss this is Wed call? Generally, a service should not send packets to other service directly. They are always multiplexed via dispatchers to avoid too maintain pipeline simplicity. Additionally, I want to understand how the egress classifier can be used for the non-NSH SFC proposed in (https://git.opendaylight.org/gerr it/#/c/66552/). Can we discuss this too. Regards, Vinayak -----Original Message----- From: Jaime Caamaño Ruiz [mailto:[email protected]] Sent: Friday, May 04, 2018 2:11 PM To: Vinayak Joshi <[email protected]>; [email protected] ght.org Subject: Re: [sfc-dev] Question on OF pipeline No, it does send it to the ingress sfc service directly in that case. -----Original Message----- From: Vinayak Joshi <[email protected]> To: [email protected] <[email protected]>, jc [email protected] Subject: RE: [sfc-dev] Question on OF pipeline Date: Fri, 4 May 2018 04:29:59 +0000 Hi Jaime, What Egress SFC classifier does if the pkt has NSH and if SFF is local? (The slide says it gives is to Ingress SFC Service, it should be “gives it to Ingress Dispatcher” is that correct?) Regards, Vinayak -----Original Message----- From: Jaime Caamaño Ruiz [mailto:[email protected]] Sent: Thursday, May 03, 2018 8:07 PM To: Vinayak Joshi <[email protected]>; [email protected] ght.org Subject: Re: [sfc-dev] Question on OF pipeline Hello Vinayak The slide is probably a bit updated. > Doesn’t this cause a packet loop? Classifier service index is higher than SFC service index, so this can't happen. We also set a magic value in NSH C1 on ingress classifier, and egress classifier only acts when this magic value is present. > Isn’t this incorrect, packet with NSH should always be forwarded by > SFC services? The packet is intercepted and forwarded by the egress SFC classifier service, to the first SFF. Nevertheless, the egress SFC classifier needs to know which is the final destination (lport) of the packet, and thats determined by the remaining ingress services so thats the reason why the packet is resubmitted to the dispatcher by the ingress SFC classifier service. BR Jaime. -----Original Message----- From: Vinayak Joshi <[email protected]> To: [email protected] <[email protected]>, jc [email protected] Subject: [sfc-dev] Question on OF pipeline Date: Thu, 3 May 2018 10:28:00 +0000 Hi, I have a few questions on the OF pipeline described in Slide 1 at https ://docs.google.com/presentation/d/1gN8GnpVGwku4mp1on7EBZiE41RI7lZ- FFmFS2QlUTKk/edit#slide=id.g210ac90bb8_0_78. Sorry for the long mail. Please point out if the scenarios/my packet flows are incorrect. Say a regular packet arrives at br-int1 on port P1 from VM1 (i.e. VM1 is not an SF). Say the SFC Classifier (ingress) matches ACL ant it stamps NSH into the packet and resubmits to the ingress-dispatcher. Now there are two possible scenarios: Port P1 also has SFC service enabled on it. i. Packet gets into Ingress SFC service which sends the packet to egress dispatcher table. ii. Egress SFC Classifier acts on it. Since the packet has NSH and if the fist SFF is on the same br-int the packet is given packet to Ingress SFC. This is as per- iii. Question: Doesn’t this cause a packet loop? Port P1 does not have SFC service enabled on it. i. In this case a packet with NSH would be given to some other service (say L3) which for forward packet to final destination with NSH. ii. Question: Isn’t this incorrect, packet with NSH should always be forwarded by SFC services? Thanks in advance, Vinayak _______________________________________________ sfc-dev mailing list [email protected] https://lists.opendaylight.org/mailman/listinfo/sfc-dev _______________________________________________ sfc-dev mailing list [email protected] https://lists.opendaylight.org/mailman/listinfo/sfc-dev
