multi-point proposal
Hello NSI team, There was a discussion at Monday’s meeting about revisiting the idea of an NSI multi-point service. I have attached the presentation by Miyamoto where he presented a proposal for such a service in our Chicago meeting. Regards, Guy Guy Roberts PhD Senior Network Architect Tel: +44 (0)1223 371316 Mob: +44 (0)7881 336417 Skype: guy1965 Networks • Services • People Learn more at www.geant.org<http://www.geant.org/> GÉANT is the collective trading name of the GÉANT Association in Amsterdam, NL, and of GEANT Limited in Cambridge, UK GÉANT Vereniging (Association) is registered in the Netherlands with the Chamber of Commerce in Amsterdam<http://www.kvk.nl/english/traderegister/default.asp>. Registration number: 40535155. Registered office: Singel 468 D, Amsterdam 1017 AW, The Netherlands GEANT Limited is registered in England & Wales. Registration number: 2806796. Registered office: City House, 126-130 Hills Road, Cambridge CB2 1PQ, UK.
Hi On Fri, 2 Jun 2017, Guy Roberts wrote:
Hello NSI team,
There was a discussion at Monday’s meeting about revisiting the idea of an NSI multi-point service.
I have attached the presentation by Miyamoto where he presented a proposal for such a service in our Chicago meeting.
While the base idea is fairly simple (which is good) and mixes well into NSI, there is a pressing real-world issue: - What happens when someone create multiple switching points and hooks them together? - In particular, when/if a loop is created? (and this is typically needed for redundancy in the real world - especially since current NSI is based off static demarcation points which make link failover rather tricky to engineer). OpenFlow is not really enough here. AFAIK the only cross-platform protocol that solve this is spanning tree, but it is rather limited with how it works across VLANs which is suggested - typically an STP process is run for every VLAN. You can do VLAN rewrite, but I think that would cause more issues than it solves here. Spanning tree typically becomes a pain in large environments, and mixing implementations across multiple vendors over high-latency links is very likely to cause a lot of issues (this is typically avoided). VPLS is a lot better (as it runs over MPLS/IP, but therefore the equipment is a lot more expensive). Then there is stuff like TRILL, which might or might not be better, but it is not that widespread yet AFAIK. However none of these suggestions mix that well with OpenFlow, which is the suggested switching technology in the presentation. But not having any established protocol for handling loops sounds a bit dangerous. It is infinitely easier cause accidents with multi-point than point-point. Of course there is nothing that prevents someone from doing it - and something very similar has already been done already in GTS. But it also a more isolated infrastructure, and intended for experiments (from the presentation it is not quite clear what the scope of the suggestion is). Best regards, Henrik Henrik Thostrup Jensen <htj at nordu.net> Software Developer, NORDUnet
participants (2)
-
Guy Roberts
-
Henrik Thostrup Jensen