100 likes | 197 Vues
Discusses the challenges and solutions related to implementing PWE3 model, including advantages and disadvantages of UDP PWs. Highlights a discussion on UDP PWs, potential problems, and proposed solutions for enterprise and service provider models.
E N D
UDP Issues PWE3 – 61th IETF 11 - 11 - 2004 Yaakov (J) Stein
Service Provider Model in the standard PWE3 model • emulation is PEto PE • IWF located at PE • AC is native service PW native service native service CE CE IWF PE PE IWF attachment circuit attachment circuit PSN
Enterprise Model there is an alternative model (CE2E) • emulation is CEto CE (see draft-stein-pwce2e-00) • IWF located at CE • what runs over the AC ? PW CE IWF IWF CE PE PE attachment circuit attachment circuit PSN
AC possibilities MPLS AC • extend MPLS towards the customer • set up PWs from CEs to PEs • splice (stitch/switch) the access PWs and core PW UDP/IP AC • leave MPLS in the core network • use UDP/IP from CEs to Pes • terminate UDP/IP at the PE and send over MPLS PW other AC possibilities • L2TP • MPLS over IP • native service over IP using GRE (when defined) • MPLS over IP using GRE
List discussion • there was a lively discussion of this issue on the list • over 50 emails from 16 participants • the following 3 slides summarize what was said
UDP PW advantages • UDP/IP is familiar to enterprise customer base (Stewart) • PW label as UDP Port number reduces overhead (Yaakov) • already extensively deployed for TDM PWs (Yaakov,Stewart) • reuse of AVT protocols (Sasha, Ron, Amnon, Andy) • simplify NAT traversal (Yaakov, Mark)
UDP PW disadvantages • hard to provide QoS assurances w/o co p2p trail (Neil) • there should be no layer networks above UDP • no operator has spoken out • large number of UDP ports - doesn’t scale (Mark) • less than 64K port numbers altogether • increases state maintained in NAT/Firewall • need protocol for UDP port signaling (Yaakov) • UDP checksum introduces processing overhead (Mark) • why introduce new PW type at such a late stage when we already have MPLS and L2TP? (Eric, Richard) • potential security problems (Stewart) • potential congestion control problems (Stewart)
Misc comments • need to reply to ITU liaison (Stewart) • PWE charter aimed at operators/SPs not customers (Ben, Mark) • wrong, but hard to stop customers from using it (Neil) • no consensus here (Eric) • discussion should be diverted to AVT (Ron, Andy) • but CE-CE PWs not in AVT charter (Sasha) • UDP OK for VoIP since adapts an application but for adapting a layer network (Ben) • some comments seem to rule out MPLS PWs too (Yaakov)
Disadvantage rebuttal • hard to provide QoS assurances w/o co p2p trail • QoS similar to LDP based MPLS or L2TP • large number of UDP ports - doesn’t scale • enterprises do not need many PW labels • scales better than VoIP presently being deployed • need protocol for UDP port signaling • can limit to manual provisioning • several simple alternatives (draft-stein-pwe3-udp-00.txt) • UDP checksum introduces processing overhead (Mark) • checksum also useful / may be set to zero • why introduce a new PW type at such a late stage • has been in charter from the beginning • potential security problems • LDP and L2TP protocols are similarly unsafe • potential congestion control problems • similar to L2TP
Proposal • explicitly limit UDP/IP to enterprise (CE-CE) PWs • if present charter is only for SPs then need to update • only allow manual provisioning • enterprise responsible for • security (firewall) • congestion avoidance (admission control) • if the enterprise requires a large number of PWs then MPLS access PWs should be used