From: Nitin Singh <[email protected]<mailto:[email protected]>>
Date: Sunday, March 15, 2015 at 9:43 AM
To: Cisco Employee <[email protected]<mailto:[email protected]>>, 
"[email protected]<mailto:[email protected]>"
 
<[email protected]<mailto:[email protected]>>
Cc: Hassan Hosseini <[email protected]<mailto:[email protected]>>, Ryan 
Bickhart <[email protected]<mailto:[email protected]>>, Manoj Sharma 
<[email protected]<mailto:[email protected]>>, John E Drake 
<[email protected]<mailto:[email protected]>>
Subject: RE: draft-sajassi-bess-evpn-etree

Hi Ali ,

How would the Etree service would work for Single Active Ethernet Segments .  
Please note that AD route per ESI for single Active segments have ESI label 
encoded as zero as defined in RFC 7432 .

IMO  the ESI label is useful for loop prevention even in case of S/A MH 
segments for ELAN service too, not sure why it was decided to encode this as 
zero in RFC 7432 .

Thanks
Nitin



From: Ali Sajassi (sajassi) [mailto:[email protected]]
Sent: Thursday, March 12, 2015 11:31 AM
To: Nitin Singh; 
[email protected]<mailto:[email protected]>
Cc: Hassan Hosseini; Ryan Bickhart; Manoj Sharma; John E Drake
Subject: Re: draft-sajassi-bess-evpn-etree


Exactly!

Cheers,
Ali

From: Nitin Singh <[email protected]<mailto:[email protected]>>
Date: Thursday, March 12, 2015 at 11:02 AM
To: Cisco Employee <[email protected]<mailto:[email protected]>>, 
"[email protected]<mailto:[email protected]>"
 
<[email protected]<mailto:[email protected]>>
Cc: Hassan Hosseini <[email protected]<mailto:[email protected]>>, Ryan 
Bickhart <[email protected]<mailto:[email protected]>>, Manoj Sharma 
<[email protected]<mailto:[email protected]>>, John E Drake 
<[email protected]<mailto:[email protected]>>
Subject: RE: draft-sajassi-bess-evpn-etree

Hi Ali ,

Thanks for your response,  it seems like that the proposal for SH segments  is 
to advertise the 2  AD routes per ESI with the ESI value encoded as 0 , one for 
SH segments in  Leaf role and other one the SH segments in Root mode .

Regards
Nitin

From: Ali Sajassi (sajassi) [mailto:[email protected]]
Sent: Thursday, March 12, 2015 10:39 AM
To: Nitin Singh; 
[email protected]<mailto:[email protected]>
Cc: Hassan Hosseini; Ryan Bickhart; Manoj Sharma; John E Drake
Subject: Re: draft-sajassi-bess-evpn-etree


Hi Nitin,


From: Nitin Singh <[email protected]<mailto:[email protected]>>
Date: Thursday, March 12, 2015 at 10:25 AM
To: Cisco Employee <[email protected]<mailto:[email protected]>>, 
"[email protected]<mailto:[email protected]>"
 
<[email protected]<mailto:[email protected]>>
Cc: Hassan Hosseini <[email protected]<mailto:[email protected]>>, Ryan 
Bickhart <[email protected]<mailto:[email protected]>>, Manoj Sharma 
<[email protected]<mailto:[email protected]>>, John E Drake 
<[email protected]<mailto:[email protected]>>
Subject: RE: draft-sajassi-bess-evpn-etree

Hi Ali ,

Please note that IM routes are advertised at granularity of Ethernet Tag [ IOW 
, there is a common IM route for ALL Ethernet Segments ].

Hence I agree that there is doubling of IM routes per Ethernet Tag per EVI  but 
this does not have the multiplication factor of number of ESIs. Essentially we 
will have two multicast groups one for the leafs other one for the roots .

I said the number of IM routes are doubled (e.g., 100% overhead) which is the 
case.

The issue that I see is that the use ESI label in Etree service is slightly 
different than the one described for ELAN service in RFC 7432 .

In ELAN service , ESI label is only used between the set of MH PEs connected to 
the ESI , but with the approach described in ETree even the PEs not connected 
to the ESI have to create forwarding state/SH procedures using the ESI label .

That’s not correct – i.e., ESI label is NOT just used among MH PEs.  Ether–AD 
per ES route along with ESI label is advertised to all PEs participating in the 
VPNs. There is a forwarding state that is created for ESI label for P2MP LSP 
anyway (but not for ingress replication).

Also there can be single homed AC in root/leaf roles , there will be no AD 
route per ESI routes for these single homed AC hence we cannot use the SH 
mechanisms  based on ESI label .

That’s why we only need to advertise two such labels per PE (one for root and 
another for leaf)!

CHeers,
Ali

Thanks
Nitin





From: Ali Sajassi (sajassi) [mailto:[email protected]]
Sent: Thursday, March 12, 2015 10:01 AM
To: Nitin Singh; 
[email protected]<mailto:[email protected]>
Cc: Hassan Hosseini; Ryan Bickhart; Manoj Sharma; John E Drake; Ali Sajassi 
(sajassi)
Subject: Re: draft-sajassi-bess-evpn-etree


There are bunch of disadvantages for doing it this way:

1.      That’s a lot of advertisements if we do it with IM route - i.e., the 
BGP messages for IM gets increased by 100% !! A given ES can support hundreds 
or thousands of IM routes !!
2.      In data-plane we end up increasing the size of LFIB
3.      There are now to different approaches for single-home versus 
multi-homed sites

Cheers,
Ali

From: Nitin Singh <[email protected]<mailto:[email protected]>>
Date: Wednesday, March 11, 2015 at 5:14 PM
To: 
"[email protected]<mailto:[email protected]>"
 
<[email protected]<mailto:[email protected]>>
Cc: Hassan Hosseini <[email protected]<mailto:[email protected]>>, Ryan 
Bickhart <[email protected]<mailto:[email protected]>>, Manoj Sharma 
<[email protected]<mailto:[email protected]>>, John E Drake 
<[email protected]<mailto:[email protected]>>
Subject: draft-sajassi-bess-evpn-etree
Resent-From: <[email protected]<mailto:[email protected]>>
Resent-To: 
<[email protected]<mailto:[email protected]>>
Resent-Date: Thursday, March 12, 2015 at 4:50 AM

Hi ,

This is regarding the section 3.2 BUM Traffic .   From my reading of the draft 
it seems that the assumption here is that BUM traffic will  be send with the 
ESI label  even to the PEs which are not connected to the ESI.

An alternative solution could be to advertise two  IM routes per Ethernet Tag

1) One for  the AC in leaf mode
2) Second for the AC in Root mode

The advantage of the above solution is that we don’t need the ESI label in 
label stack for the packet send towards the set of PEs not connected to the ESI 
.

The use of ESI label could be restricted  for forwarding the BUM traffic 
between MH PEs connected to ESI operating in all-active mode,  as per the SH 
procedures  described in the RFC 7432 .

Thanks
Nitin




_______________________________________________
BESS mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/bess

Reply via email to