EiC IEEE April 6-9, 1992 Sun H0tel Phoenix Miyazaki, Japan Q6‘ 4 <‘?/8 THE ZND BROADBAND ISDN G1 in 0/- 5~‘\ 9 , 99$ EE, SuppOY\9¢\O\§ '0 ‘\ TECHNICAL WORKSHOP Conference Record $P0nS0r6d by IEICE, Technical Groups on Switching Systems Engineering, Infonnation Networks, and Communication Systems IEEE Communications Society $l1pP0rt¢d by Miyazaki Prefectural Govemment The Telecommunication Tecimology Committee Intemational Communication Foundation Asian and Pacific Areas Assembly for Communications Systems Miyazaki Convention Bureau <3‘ K

» .~, -.-.»ug~_.m..--¢_..;.».-u-n~»~nn-.>.nv~4r->.-as ..... _ ATM LANS and B-ISDN Tom Lyon Distinguished Engineer Sun Microsystems, Inc. [email protected] AIM LAN! Ind B-ISDN 2nd B-ISDN Wmkslxlp 0 LAN Trends into physical star star - Bandwidth is shared, but wires are not! instead of media access protocols! - Ethernet 10Base-T concentrators make logical bus ‘ - FDDI concentrators make logical ring into physical - If star topologies are prevalent, use switching nodes KIM LAN: Ind B-ISDN anB-ISDN Wdhhqr

0 LAN Problems - Bandwidth crisis! Aggregate b/w > 100 Mbps needed now, individual b/w > 100 Mbps needed soon! - No real-time guarantees in current LANs; required by M.M. applications, e.g., video teleconference - Bus-based routers & bridges of adequate b/w are too hard to build. Use of space switching easier. - Lack of scaling in network b/w causes system imbalance as MIPS grows KIM LANS and B-ISDN 2nd B-ISDN Wwkshop 10 Year System Scaling - CPU: 1982 @ 1 MIP, 1992 @ 30 MIPS - Disk: 1982 @ 5 Mbps, 1992 @ 40 Mbps - Net: 1982 @ 10 Mbps (<1 usable), but 1992 is still @ 10 Mbps (50 usable!) - Growth is exponential and continues rapidly! ATM LAN: md B-ISDN gm] 3.151)}; wa-“hop

~. ,. . i.@-...~.t~,~»»»wlI1€--»»'_: - » Application Trends - Multimedia applications will generate huge amounts of data - B/\N use will grow to capacity if cheap enough, just like CPU and disk - Apply any resource to aid user or programmer productivity - Globalization of business pushing need for better WANs A1'M LANS md B-ISDN 2nd B-ISDN Wakshq) Why Local ATM? - Need for high aggregate bandwidth - Need for multi-sen/ice/multimedia capabilities ~ Need for scalable solution - faster speeds or wider geographies with same architecture - Same needs for LAN as for B-ISDN, only sooner - Leverage R&D investment of B-ISDN community ATM IAN: Ind B-BDN ' 1|“3.131)}: wqgmq,

_ ,, _, an a&m%m »>wm , . The Vision - 155 Mbps to each desktop, n*155 Mbps to sewers, point-to-point wiring only - CPU motherboard connection @ < $100'parts cost, wiring closet switch/concentrator @ $1000/port - Short run from desktop to wiring closet - use unshielded twisted pair wiring for low cost - ATM LAN deployment creates demand for B-ISDN ATM LANs and B-ISDN 2nd B-ISDN Wmkshcp Usage Assumptions - . Traffic will be predominantly existing LAN protocols, e.g., TCP/IP - ATM is just one of many data link level networks, so both end systems and routers must work well - Apps and protocols will evolve to request QOS - pure datagram service no longer sufficient - All true for B-ISDN as well as ATM LANs ATM LAN: lidB-BDN nd B-ISDN Wtrkxqa

0 Challenges - Requirements change from CO switch to wiring closet concentrator: cost, size, physical layer, policing, etc. A - Simplicity and scalability of end-system interfaces: adaptation layer - Congestion: no MAC layer means simultaneous transmitters can overrun receiver " - Ease of use, setup, and ownership A'I'M LANs and B-ISDN 2nd B-EDN Workshop Physical Layer - Single mode fibre, lasers, & connectors too expensive for local area use; ATM/SONET chips not yet ready - For now, use multimode fibre with 8B/10B framing (follows ANSI FibreChannel) ‘ - Develop transceivers for SONET/STM over data- grade unshielded twisted pair (follow ANSI FDDl work) - Multimode fibre feasible even with STS-12c/STM-4 l l A'l'M LAN8 lidB-ISDN x! B-ISDN Wqkshqy

, _..,.- ..¢....“......;... '_..__.,.. 7*""*' *"*7&i—€ Adaptation Layer - Concern about AAL3/4's CRC-10 for different media - Much complexity added by fields such as MID which duplicate functionality elsewhere - Inefficient to transfer data in 44 byte units, especially with 64 bit buses - AAL5 proposal: 48 bytes per cell, CRC-32 per packet, ‘minimalist’ philosophy KIM LAN: md B-ISDN 2nd B-BDN “'4;-kghq; Congestion - The Problem 155 I--,= 155 15$LIpa Cong¢lonhLAN - Speed of host interface equals speed of switch port - Congestion is the norm, unless avoided ATM LAN: ma B-ISDN - 2m B-ISDN wu-mmp

0 Congestion Issues - Statistical multiplexing fails with high peak rates - Increasing peak rates and burst sizes - Congestion management features add cost to switch - Dropping cells/packets wastes too much b/w waiting for retransmission - Unpredictable application/user requirements A'I'M LANs and B-ISDN 2nd B-ISDN Wwkshop Adaptive and Cooperative Bandwidth Management - Peak rate allocation with dynamic adaptation - Allocation for VC adjusts to changing needs - Host shapes traffic to the bandwidth allocation - Host cooperates with network's policies to achieve u fair allocation ; ut~s-¢e¢2~<\-l H/\ 1>.wa-Jj a.(g<>.§l¢\.A ' M\‘°=> 3‘""'*°‘<< V ATM LAN: md B-ISDN 2nd B-ISDN Wllkallp

Q Signalling, Addressing, Etc. - Ease of use critical for LANs: need auto- configuration, self-identification, etc. - Assignment of E.164 addresses in LAN redundant with higher level IP, OSI, Appletalk, etc. addresses. Use multi-address structure as in multi-protocol routers - Need to ensure smooth evolution of signalling rather than trying to get it perfect the first time. Allow an extensible framework for features specific to higher level protocols families. A'I'.M LAN: lid B-ISDN 2nd B-ISDN Won-lcabqa ATM LANs vs. B-ISDN - ATM LANs and B-ISDN need each other for success - But LAN and WAN requirements are different: don't assume the B-ISDN solution for the LAN unless the problem is the same - The WAN should evolve towards the LAN as the LAN evolves towards the WAN KIM LAN: md B-ISDN an13.151»: wqhhq,