RARE user and assistance email list

[Frédéric LOUI <>]

Hi,

>  i'm planning to add nat as an option soo

Nice. This could be great if this « just » require a additional P4 control.
TBH NAT at 100GE is somewhat irrelevant but for 1GE,10GE or in DC space it 
might be (?)

>  opens up the possibility for me to have more images for a test set and the 
> tester
> be able to sselect one that believed to be albe to execute it successfully

This is great.

>  the first one that depends on the recirculation,

IIRC, in the VM I gave you, you’ll see a « resubmit » example from 
INTEL/BAREFOOT.
As per BA-102 training, it is strictly the same in terms of code 
implementation. You can get a good grasp of it.
If needed, we can ask some help in FASTER portal. I had the confirmation that 
KIFU is INTEL/BAREFOOT partner.
Is Janos gave you access to BAREFOOT/FATER portal ? If not can he gives you 
access to it ? The process is straightforward.

> second part are simply a kind of nexthop … seems easy-peasy

This is re-assuring. It could be great though to have this mainstream in a PE 
SDM, or if the it fits into the NPU to have this in the default MPLS SDM.
Or maybe this would make more sense to create a BRAS SDM.

> the third category is the mix of the two above.

So when the 2 above would occurs then :)

Keep up the good work !

À bientôt,
--  Frederic




> Le 29 août 2020 à 06:25, mc36 <> a écrit :
> 
> hi,
> 
> 
> On 8/28/20 6:48 PM, Frédéric LOUI wrote:
>> Hi
>> Great !
>> My KVM_ONIE VM was not useless after all :)
> absolutely!!!
> 
>> So this is nice, because that we can have a dynamic feature set in a list 
>> for that I can integrate in the RARE WIKI.
> the newer ones are 'nicer', you'll see... :)
> 
>> By dynamic, I mean that in a CI way:
>> * each commits
>> * adding new features
>> * will trigger test against TNA (bf_switchd) platform
>> * these test results reflect the features list and their status. (RFS or 
>> not )
> external ci tools would go crazy with this: it takes 2 hours to run on the 
> current
> p4 feature list, requires a lot of cpu, ram and diskio because of the qemu 
> instances
> continously spawned and stopped, and the most critical, requires the 
> image(s) to be
> present in the ci environment, and surely the execution of my tester, since 
> i've
> not aware of any other automatic testing solution that does anything 
> similar this... :)
> 
>> Note that due to the use case nature, there will be some RARE/freeRouter 
>> profile (SDM)
>> For example, we have SDM=MPLS, SDM=SRv6
>> In some cases having GRE or NAT at 100GE makes no senses, some we will 
>> have more SDM.
>> i.e All the features that boast BMv2 cannot enter into TOFINO. BMv2 has 
>> virtually no memory/resources constraints which is not the case for 
>> TOFINO. So we will end up in various SDM profiles for TNA architecture 
>> having a subset of features of the whole feature set that BMv2/P4DPDK  has.
> to be honest, nat is straightforward since it's just another control 
> executed at the right place,
> and all the infra needed for it is the acls, and that's already there so i 
> modularized bridging
> in the bfswd code like we modularized mpls and srv6 to have some stages so 
> yess, i'm planning to
> add nat as an option soon...
> also introduced features in my auto-tester by the way we discussed 
> yesterday that is, images now
> describe which features they provide and test cases describe which features 
> they require and a
> test is onyl started if requirements meets provided features... it's a huge 
> speedup for my runs
> on bfswd butt opens up the possibility for me to have more images for a 
> test set and the tester
> be able to sselect one that believed to be albe to execute it 
> successfully... as soon as i'll
> get to the nat on bfswd part, it'll happen...
> 
> regarding the rest of these features introduced recently, they fall on 
> three categories: the
> first one that depends on the recirculation, which started with hairpin and 
> believed that we have
> in tofino so since i have now a convenient development environment for 
> bfswd, will give it a try...
> the second part are simply a kind of nexthop, for example gre, pppoe and 
> l2tp (and vxlan will be
> the same!) so surely seems easy-peasy... the third category is the mix of 
> the two above.... :)
> 
> regards,
> cs
> 
> 
>> Thanks for this excellent job Csaba !
>> À bientôt,
>> --  Frederic
>>> Le 28 août 2020 à 18:34, mc36 <> a écrit :
>>> 
>>> hi,
>>> finally got a working tester vm with barefoot tofino emulator installed 
>>> within...
>>> it runs exactly the same binary generated from the same p4s as our wedge 
>>> switches...
>>> please find attached the test runs against it... don't be scared of the 
>>> lot of
>>> failing ones, those features are simply not available in our current 
>>> tofino p4
>>> code tree  (or not enabled in the tested mpls sdm profile, like srv6), 
>>> but all
>>> exists for bmv2 so each could be implemented, and some are already 
>>> selected to
>>> arrive sooner or later...
>>> regards,
>>> cs
>>> <bfswd.html>