1. 1 IETF-56 PWE3 Workgroup 19-Mar-03, San Francisco Edge-to-edge Emulation of Structured TDM Services over PSN: Open Issues Sasha Vainshtein

2. 2 03/19/03 IETF-56 San Francisco What This Is About Consensus on emulation of unstructured TDM (E1, T1, E3, T3) looks very close : Raw TDM payload encapsulated "as is" in packets of fixed size No alignment presumed or preserved The packet payload size is a matter of convenience Different sizes should be supported in order to allow trade-off between BW efficiency and packetization latency Mandatory defaults can be agreed upon Why is the situation with structured TDM so different? Different people mean different things when they talk about "structured TDM emulation" A clear WG position on scope, requirements and architecture seems to be lacking The WG input is solicited on a number of issues

3. 3 03/19/03 IETF-56 San Francisco What is Structured TDM A Frame Alignment Signal structures the trunk bit-stream at two levels: TDM frames: Always present Always octet-aligned Number of octets per frame is trunk-specific DS0 channels use a single octet per frame NxDS0 services use a subset of N octets per frame Some applications depend on NxDS0 frame integrity Always 8000 frames per second An NxDS0 service carries Nx64 kbit/sec, 1 <= N <=31 Multiframes/superframes May be absent on some trunks Not synchronized between different trunks The number of frames per multiframe is trunk-dependent Bit-error detection and OAM channels DS0 channel-associated signaling (CAS)

4. 4 03/19/03 IETF-56 San Francisco Why Structured TDM Emulation? Suggested rationales vary between: Save the PSN BW by skipping unused channels This reasoning has been used for AAL5 SDU in ATM Fidelity of BW-saving emulation may be limited Split a single trunk attachment circuit between several applications Different groups of timeslots are used by different applications and are carried by different PWs Assumes a cross-connect NSP in the PE Improve resilience of the attachment circuit to packet loss The "interface down" condition in the CE never occurs Can be achieved in unstructured emulation under reasonable assumptions Improve resilience of specific applications to packet loss Interpolate between lost packets carrying Voice samples Application-specific BW management E.g., graceful congestion handling

5. 5 03/19/03 IETF-56 San Francisco Splitting the Access Circuit Trunk A DS0 Grooming Application

6. 6 03/19/03 IETF-56 San Francisco Which Structures to Preserve? Answers vary between: NxDS0 frames Supports all application requirements for data integrity Supports interworking of services originating in different trunks Allows extension of N beyond a single trunk Treats CAS as encoding of CE application signaling Signaling phase relative to data is not perfect "Form" of signals can be preserved Signaling may be carried in separate packets or travel appended to the NxDS0 payload NxDS0 frames without CAS or trunk multiframes with CAS: CAS remains interleaved with the TDM payload Additional timeslot required for E1 trunks Problematic with existing NSP blocks Problematic handling of packet loss "Replacement logic" for data and signaling is different

7. 7 03/19/03 IETF-56 San Francisco Which Structures to Preserve (2)? ATM-CES-like: NxDS0 frames for NxDS0 without CAS "Superframe structures with signaling" for NxDS0 with CAS Requires extraction/insertion of CAS as a set of bit-streams Makes services with CAS trunk-specific Multiframe alignment may be preserved at the expense of additional delay in the NSP Requires double capacity of the IWF/NSP interface Handling of lost packets is problematic Always NxDS0 multiframes/superframes Problematic if multiframe structures do not exist in the trunk End services originating in different trunks will not interwork Loop Emulation-like: Frame integrity is not preserved, only DS0 integrity Suits only telephony applications Allows additional BW saving

8. 8 03/19/03 IETF-56 San Francisco Granularity of Packet Latency? Important for low-rate services: Trade-off between packet latency and BW efficiency becomes essential The rate varies from 64 kbit/s to 1920 kbit/s Latency of NxDS0 PWs between the given pair of PEs should not depend on N The answers vary between: 125 microsecond granularity is mandatory 125 microsecond granularity is recommended The granularity is not important at all

9. 9 03/19/03 IETF-56 San Francisco Piece of Wire or Distributed DXC? Trade-off between the IWF and NSP Complexity Answers vary between: A "piece of wire" approach (which wire?) Absolute timeslot positions in the trunk must be preserved CAS alignment with data is fully preserved (NSP may impair it later) Handling of lost packets may be non-trivial A DXC approach Absolute timeslot positions are irrelevant TDM data and signaling may follow different paths Synchronization between data and signals may be impaired Reliable signaling delivery may be an issue A Mix of both Absolute timeslot positions are irrelevant CAS alignment with data is preserved

10. 10 03/19/03 IETF-56 San Francisco Questions?