Eo S Yaakov J Stein Chief Scientist RAD

Eo. S Yaakov (J) Stein Chief Scientist RAD Data Communications

Course Outline 1) Introduction 2) Background - Ethernet 3) Background – HDLC 4) Background - PPP 5) Background - SONET/SDH 6) VCAT 7) LCAS 8) POS (PPP over SONET/SDH – RFC 1619/2615) 9) LAPS 10) GFP 11) Alternatives Y(J)S Eo. S Slide 2

Introduction Y(J)S Eo. S Slide 3

Motivation Assume that you are a traditional operator n n n n You have an extensive SONET/SDH network This network has cost you Millions-Billions to build This network is highly reliable Your staff is well trained to maintain it You may have not yet reached Return On Investment It supports the service that brings the most revenue – voice It supports the service with the highest margin – leased lines But suddenly customers are asking for something new n “Ethernet handoff” And new competitors are willing to supply it! Y(J)S Eo. S Slide 4

Option 1: install new infrastructure You may choose to build a new IP/MPLS based network (BT 21 CN approach) Yes – this means significant investment, but this is definitely the future! But SONET/SDH has comparative advantages: n Reliable optical transport n Well known technology and protocols n Ubiquitous with present operators n Many supported data rates (from 1 Mbps to many Gbps) n Low overhead n Strong OAM (MPLS isn’t there yet …) So if you replace the existing network n How will you handle the service that brings your main income – voice ? n You may lose your existing leased line customers n You will need to solve the timing distribution problem And if you keep your existing network n You need to maintain two completely different networks ! This sounds problematic ! Y(J)S Eo. S Slide 5

Option 2: leased lines Ethernet Switch I W F A D M SONET RING A D M I W F Ethernet Switch You can try to convince these customers to use leased lines The customer converts traffic into T 1/E 1 (e. g. by using frame relay) n n n You can supply this service now The major expense is for the customer (who needs FRAD, CSU/DSU, etc. ) Leased lines are profitable But this only worked before the new competitors appeared You will probably lose these customers ! Y(J)S Eo. S Slide 6

Option 3: ATM Ethernet Switch A T M A D M SONET RING A D M A T M Ethernet Switch You can offer ATM service The customer converts traffic into ATM (AAL 5) n You can supply this service now n ATM is a well-known technology n ATM is a reliable and high-quality service n ATM maps efficiently onto SONET/SDH n You may even be able to perform the conversion at your POP (but Ethernet is notoriously hard to transport over distances) But ATM has its disadvantages n ATM has high overhead – but you can only charge for user BW n ATM is an additional network n – you will have to train and pay new staff – maintain another operations center ATM usually carries IP, not native Ethernet traffic Y(J)S Eo. S Slide 7

Option 4: Eo. S Ethernet Switch I W F SONET RING I W F Ethernet Switch A new choice is Ethernet over SONET/SDH (Eo. S) The customer’s Ethernet traffic is transported directly by SONET/SDH n You build on your existing network n You transport native Ethernet – needn’t route at network edges – maintain all Ethernet features n New SONET/SDH features make Eo. S highly efficient But Eo. S and related protocols are new technologies n You may need to upgrade existing equipment n Market hasn’t yet stabilized on one technology So you will probably need to take this course ! Y(J)S Eo. S Slide 8

World’s Apart SONET/SDH is presently the most prevalent transport infrastructure Ethernet is by far the most popular user data interface So we need efficient methods for carrying Ethernet over SONET But Ethernet n n comes in bursty “frames” (packets) uses basic rates of 10, 1000 Mbps While SONET/SDH n n is constant bit rate is designed for various rates such as 1. 6, 2. 176, 6. 784 Mbps So the job isn’t easy ! Y(J)S Eo. S Slide 9

Standards we will encounter IEEE 802. 3 Ethernet ISO 3309 HDLC RFC 1661 PPP (ex 1548) RFC 1662 PPP in HDLC framing (ex 1549) RFC 2615 Po. S (ex 1619) G. 707 SDH (especially the new section 11 – VCAT) G. 709 OTN G. 7041 GFP G. 7042 LCAS for SDH G. 7043 VCAT for PDH X. 85 IP over SDH using LAPS X. 86 Ethernet over SDH using LAPS Y(J)S Eo. S Slide 10

Background Ethernet Y(J)S Eo. S Slide 11

Ethernet frame For our purposes, “Ethernet” is any layer 2 protocol using 1 of the following frame formats : 64 – 1518 B DA (6 B) SA (6 B) T/L (2 B) data (0 -1500 B) pad (0 -46) FCS (4 B) 68 – 1522 B DA(6 B) SA(6 B) VT(2 B) VLAN(2 B) T/L(2 B) data (0 -1500 B) pad(0 -46) FCS(4 B) Y(J)S Eo. S Slide 12

Ethernet frame size n Minimum frame is 64 bytes n Maximum payload was 1500 bytes – and maximum frame was 1522 bytes n 802. 3 as lengthened maximum frame to 2000 bytes n Various physical layer modulations and framing n Rates : 10 Mbps, 100 Mbps, 1 Gbps, 10 Gbps, … Y(J)S Eo. S Slide 13

Background HDLC Y(J)S Eo. S Slide 14

Packet to bit stream The first problem in converting Ethernet to TDM: n Ethernet consists of frames carrying packets n TDM is a continuous bit stream We can convert a sequence of packets into a bit stream by using an “idle code” packet 1 packet 2 packet 3 packet 4 For example, we can use a sequence of 1 s as idle indication 1111111111110 packet 1 011111111110 packet 2 0111111111110 01111110 packet 3 011111111 The appearance of a 0 bit indicates that data follows Y(J)S Eo. S Slide 15

Packet to bit stream (cont. ) How does the receiver know when to return to idle? We use a specific “flag” (HDLC uses hex 7 E = 01111110) We can use the flag as the idle code as well 01111110 packet 1 01111110 packet 2 01111110 packet 3 01111110 Some implementations allow “zero sharing” 011111101111110 packet 1 01111110 packet 2 01111110 packet 3 01111110 But the flag must not appear in valid data! If we have access to the physical layer we can mark there (“violations”) Otherwise (we only access bits) we must disallow the idle code by replacing it with something else Y(J)S Eo. S Slide 16

HDLC flags ISO developed High level Data Link C based on IBM’s SDLC HDLC inputs packets of bytes HDLC uses hex 7 E as its idle code (“flag”) 01111110 So an idle HDLC stream repeats 7 E 01111110 packet 1 01111110 packet 2 01111110 packet 3 01111110 Alternatively, 1 s can be sent as idle, flags as delineators 111111111 01111110 packet 1 011111101111110 packet 2 01111110 11111111101111110 packet 3 01111110 There are two methods of disallowing flags n bit stuffing (zero insertion) n byte (octet) stuffing Y(J)S Eo. S Slide 17

Bit stuffing / zero insertion ECMA-40 Whenever the encoder sees 5 successive 1 s it appends a 0 thus there are never 6 successive 1 s in the data When the decoder sees 5 successive 1 s : n If the next bit is a 0 it is deleted n If the next bit is a 1 then this is the closing flag Notes: n bit stream length is no longer necessarily divisible by 8 n bit stream length is not a priori predictable n worst case expansion is 20% n encoding/decoding is easy in HW, hard in SW Y(J)S Eo. S Slide 18

RFC 1549 Byte (octet) stuffing Whenever the encoder sees hex 7 E It replaces it with 7 D 5 E Whenever the encoder sees hex 7 D It replaces it with 7 D 5 D Optionally other codes (e. g. some under hex 20) can be “escaped” Second byte is original with 6 th bit complemented (xor with hex 20) e. g. ^Q = hex 11→ 7 D 31 ^S = hex 13 → 7 D 33 When the receiver sees 7 D xx It replaces it with the original byte (complementing 6 th bit) Notes: n bit stream remains byte oriented n length expansion is typically about 1%, but can range from 0 to 100% ! (there is also a consistent overhead algorithm – but not in use) n encoding/decoding is easy in SW Y(J)S Eo. S Slide 19

HDLC framing HDLC frame is bounded by flags, and has a particular structure flag (8) address (0/8/16) ctrl (8/16) data FCS (16/32) flag (8) Many variants (SDLC, ISO, LAPB, LAPD, LAPF, LAPS, SS 7, PPP-HDLC, Cisco-HDLC, etc) Address: n There may be no address (e. g. SS 7 HDLC) n SDLC always had 8 bit addresses n ISO 3309 HDLC has structured multibyte address SAPI n C/R EA EA – Service Access Point Identifier (MSB of SAPI =1 may indicate broadcast/multicast) – EA=1 means 8 bit, EA=0 means extended address – C/R=1 for commands, C/R=0 for responses The single byte hex FF is recognized as the broadcast address Y(J)S Eo. S Slide 20

HDLC control HDLC networks can be configured: n Balanced – all stations have equal responsibility n Unbalanced – primary and one or more secondary stations and HDLC can operate : n Best effort (datagram) – uses Un-numbered (U) frames n Reliable (Asynchronous Balanced Mode) – uses frames with sequence numbers in control field l Information (I) frames (data + acknowledgement) l Supervisory (S) frames (only acknowledgement) The various frame types are indicated by the control field which varies widely between different protocols Y(J)S Eo. S Slide 21

HDLC FCS HDLC uses a Frame Check Sequence to detect errors The FCS is implemented as a shift-register n n CRC-16 CRC-32 X 16 + X 12 + X 5 + 1 X 32 + X 26 + X 23 + X 22 + X 16 + X 12 + X 11 + X 10 + X 8 + X 7 + X 5 + X 4 + X 2 + X + 1 Some HDLC-based protocols require 32 bit FCS others allow 16 bit but recommend 32 bit FCS Y(J)S Eo. S Slide 22

Background PPP Y(J)S Eo. S Slide 23

Point to Point Protocol (RFC 1661) PPP is a method for transporting datagrams between 2 peers over full-duplex, point-to-point data links – for example: short lines, leased lines, dial-up modems PPP may be used to connect hosts to routers, and routers to routers PPP is made up of 3 components: n encapsulation method for (multiprotocol) datagrams Link Control Protocol for establishing, configuring, and testing data-link connections n Network Control Protocols for establishing and configuring different network-layer protocols n PPP is a suite containing many protocols ML-PPP, PPPo. E, BAP, BCP, IPCP, … Y(J)S Eo. S Slide 24

Basic PPP encapsulation (RFC 1661) protocol (8/16) information padding Encapsulation enables demuxing of different network-layer protocols Only 1 field needs to be examined for protocol determination Protocol field obeys ISO 3309 rules: – protocol value must be odd (for EA=1) – if 16 -bit, then the LSB of first byte must be zero (for EA=0) PPP protocol values managed by IANA (http: //www. iana. org/assignments/ppp-numbers) Padding may be used (e. g. to cause header to fall on 32 -bit boundary) Y(J)S Eo. S Slide 25

PPP using HDLC framing (RFC 1662) flag address ctrl protocol 7 E FF 03 (8/16 b) information padding FCS flag (optional) (16/32 b) 7 E When using PPP over synchronous links we use HDLC-like framing 1 byte Broadcast address is used by default (users may define alternative address) Synchronous Link may be bit-oriented or byte-oriented Basic PPP encapsulation is extended by 8 bytes Bit stuffing or byte stuffing allowed Escape mechanism allows transparent transfer of control data (e. g. ^S/^Q) enables removal of spurious control data (inserted by intermediate boxes) Y(J)S Eo. S Slide 26

RFC 1662 vs. X. 85 ITU-T X. 85 defines IP over SDH using LAPS (will study later) Its encapsulation is similar to RFC 1662 (but can’t co-exist with it) Instead of the protocol ID it has a SAPI = 21 for IPv 4 =57 for IPv 6 The FCS MUST be 32 bits and no padding is used No special escaping is defined PPP frame 1662 X. 85 flag address ctrl protocol 7 E FF 03 (8/16 b) flag address ctrl SAPI 7 E 04 03 (16 b) information padding FCS flag (optional) (16/32 b) 7 E FCS flag (32 b) 7 E IP Packet Y(J)S Eo. S Slide 27

Background SONET/SDH Note: For more information – see SONET/SDH course. Y(J)S Eo. S Slide 28

SONET architecture ADM regenerator ADM Path Line Section Line Path Termination Termination path line section SONET (SDH) has at 3 layers: n path – end-to-end data connection, muxes tributary signals path section – there are STS paths + Virtual Tributary (VT) paths n line – protected multiplexed SONET payload multiplex section n section – physical link between adjacent elements regenerator section Each layer has its own overhead to support needed functionality SDH terminology Y(J)S Eo. S Slide 29

SONET STS-1 frame 9 rows 90 columns Synchronous Transfer Signals are bit-signals (OC are optical) Each STS-1 frame is 90 columns * 9 rows = 810 bytes There are 8000 STS-1 frames per second so each byte represents 64 kbps (each column is 576 kbps) Thus the basic STS-1 rate is 51. 840 Mbps Y(J)S Eo. S Slide 30

SDH STM-1 frame 270 columns 9 rows … Synchronous Transport Modules are the bit-signals for SDH Each STM-1 frame is 270 columns * 9 rows = 2430 bytes There are 8000 STM-1 frames per second Thus the basic STM-1 rate is 155. 520 Mbps 3 times the STS-1 rate! Y(J)S Eo. S Slide 31

SONET/SDH rates SONET SDH STS-1 columns rate 90 51. 84 M STS-3 STM-1 270 155. 52 M STS-12 STM-4 1080 622. 080 M STS-48 STM-16 4320 2488. 32 M STS-192 STM-64 17280 9953. 28 M STS-N has 90 N columns STM-M corresponds to STS-N with N = 3 M SDH rates increase by factors of 4 each time STS/STM signals can carry PDH tributaries, for example: n STS-1 can carry 1 T 3 or 28 T 1 s or 1 E 3 or 21 E 1 s n STM-1 can carry 3 E 3 s or 63 E 1 s or 3 T 3 s or 84 T 1 s Y(J)S Eo. S Slide 32

SONET/SDH tributaries SONET SDH STS-1 T 1 T 3 E 1 E 3 28 1 21 1 E 4 STS-3 STM-1 84 3 63 3 1 STS-12 STM-4 336 12 252 12 4 STS-48 STM-16 1344 48 1008 48 16 STS-192 STM-64 5376 192 4032 192 64 E 3 and T 3 are carried as Higher Order Paths (HOPs) E 1 and T 1 are carried as Lower Order Paths (LOPs) Y(J)S Eo. S Slide 33

STS-1 frame structure 9 rows 6 rows 3 rows 90 columns Synchronous Payload Envelope section + line overhead Section overhead is 3 rows * 3 columns = 9 bytes = 576 kbps framing, performance monitoring, management Line overhead is 6 rows * 3 columns = 18 bytes = 1152 kbps protection switching, line maintenance, mux/concat, SPE pointer SPE is 9 rows * 87 columns = 783 bytes = 50. 112 Mbps Similarly, STM-1 has 9 (different) columns of section+line overhead ! Y(J)S Eo. S Slide 34

STM-1 frame structure 270 columns … Transport Overhead TOH Similarly, STM-1 has 9 (different) columns of transport overhead ! RS overhead is 3 rows * 9 columns Pointer overhead is 1 row * 9 columns MS overhead is 5 rows * 9 columns SPE is 9 rows * 87 columns Y(J)S Eo. S Slide 35

Scrambling SONET/SDH receivers recover clock based on incoming signal Insufficient number of 0 -1 transitions causes degradation of clock performance In order to guarantee sufficient transitions, SONET/SDH employ a scrambler n n n All data except first row of section overhead is scrambled Scrambler is 7 bit self-synchronizing X 7 + X 6 + 1 Scrambler is initialized with ones A short scrambler is sufficient for voice data but NOT for data which may contain long stretches of zeros When sending data an additional payload scrambler is used n n n modern standards use 43 bit X 43 + 1 run continuously on ATM payload bytes (suspended for 5 bytes of cell tax) run continuously on HDLC payloads Xn Yn = Xn + Yn-43 Z-43 Y(J)S Eo. S Slide 36

HOP SPE structure 2 bytes in the line overhead point to the STS path overhead POH pointer (floating) allows frequency/phase compensation (after re-arranging) POH is one column of 9 rows (9 bytes = 576 kbps) Y(J)S Eo. S Slide 37

Path overhead J 1 B 3 C 2 POH is responsible for – path performance monitoring – status (including of mapped payloads) – trace G 1 C 2 (hex) Payload type 00 unequipped 01 nonspecific 02 LOP (TUG) 04 E 3/T 3 12 E 4 F 2 2 bytes are of particular interest to us: H 4 C 2 is the “signal label” indicates path payload type 13 ATM 16 Po. S – RFC 1662 H 4 is the “multiframe indication” used by VCAT/LCAS (discussed later) 18 LAPS X. 85 1 A 10 G Ethernet 1 B GFP CF Po. S - RFC 1619 F 3 K 3 N 1 POH Y(J)S Eo. S Slide 38

STS-1 HOP 1 30 59 87 1 column of SPE is POH 2 more (“fixed stuffing”) columns are reserved We are left with 84 columns = 756 bytes = 48. 384 Mbps for payload This is enough for a E 3 (34. 368 M) or a T 3 (44. 736 M) Y(J)S Eo. S Slide 39

LOP 1 30 59 VTG 87 1 2 3 4 5 6 7 To carry lower rate payloads, divide 84 available columns into 7 * 12 interleaved columns, i. e. 7 Virtual Tributary (VT) groups VT group is 12 columns of 9 rows, i. e. 108 bytes or 6. 912 Mbps VT group is composed of VT(s) n There are different types of VT in order to carry different types of payload n all VTs in VT group must be of the same type n but different VT groups in same SPE can have different VT types A VT can have 3, 4, 6 or 12 columns Y(J)S Eo. S Slide 40

SONET/SDH : VT/VC types VT/STS LOP HOP VC column rate payload VT 1. 5 VC-11 3 1. 728 DS 1 (1. 544) 4 per group VT 2 VC-12 4 2. 304 E 1 (2. 048) 3 per group 6 3. 456 DS 1 C (3. 152) 2 per group 1 per group VT 3 VT 6 VC-2 12 6. 912 DS 2 (6. 312) STS-1 VC-3 48. 384 E 3 (34. 368) STS-1 VC-3 48. 384 DS 3 (44. 736) STS-3 c VC-4 149. 760 E 4 (139. 264) standard PDH rates map efficiently into SONET/SDH ! Y(J)S Eo. S Slide 41

Payload capacity VT 1. 5/VC-11 has 3 columns = 27 bytes = 1. 728 Mbps but 2 bytes are used for overhead so actually only 25 bytes = 1. 6 Mbps are available Similarly VT 2/VC-12 has 4 columns = 36 bytes = 2. 304 Mbps but 2 bytes are used for overhead So actually only 34 bytes = 2. 176 Mbps are available Y(J)S Eo. S Slide 42

VCAT Virtual Concatenation Y(J)S Eo. S Slide 43

Concatenation Payloads that don’t fit into standard VT/VC sizes can be accommodated by concatenating of several VTs / VCs For example, 10 Mbps doesn’t fit into any VT or VC so w/o concatenation we need to put it into an STS-1 (48. 384 Mbps) the remaining 38. 384 Mbps can not be used We would like to be able to divide the 10 Mbps among 7 VT 1. 5/VC-11 s = 7 * 1. 600 = 11. 20 Mbps or 5 VT 2/VC-12 s = 5 * 2. 176 = 10. 88 Mbps Y(J)S Eo. S Slide 44

Concatenation There are 2 ways to concatenate X VTs or VCs: n Contiguous Concatenation (G. 707 11. 1) – HOP – STS-Nc (SONET) or VC-4 -Nc (SDH) or LOP – 1 -7 VC-2 -Nc into a VC-3 – since has to fit into SONET/SDH payload n n l only STS-Nc : N=3 * 4 or VC-4 -Nc : N=4 – components transported together and in-phase – requires support at intermediate network elements n Virtual Concatenation (VCAT G. 707 11. 2) – HOP – STS-1 -Xv or STS-Nc-Xv (SONET) or VC-3/4 -Xv (SDH) or LOP – VT-1. 5/2/3/6 -Xv (SONET) or VC-11/12/2 -Xv (SDH) – HOP: X ≤ 256 LOP: X ≤ 64 (limitation due to bits in header) – payload split over multiple STSs / STMs – fragments may follow different routes – requires support only at path terminations – requires buffering and differential delay alignment Y(J)S Eo. S Slide 45

Contiguous Concatenation: STS-3 c 270 columns 9 rows … 258 columns of SPE 9 columns of section and line overhead 3 columns of path overhead 258 columns * 0. 576 = 148. 608 Mbps STS-3 270 columns 9 rows … 260 columns of SPE 9 columns of section and line overhead 1 column of path overhead 260 columns * 0. 576 = 149. 760 Mbps STS-3 c Y(J)S Eo. S Slide 46

STS-N vs. STS-Nc Although both have raw rates of 155. 520 Mbps STS-3 c has 2 more columns (1. 152 Mbps) available More generally, For STS-Nc gains (N-1) columns e. g. STS-12 c gains 11 columns = 6. 336 Mbps vis a vis STS-12 STS-48 c gains 47 columns = 27. 072 Mbps STS-192 c gains 191 columns = 110. 016 Mbps ! However, an STS-Nc signal is not as easily separable when we want to add/drop component signals Y(J)S Eo. S Slide 47

Virtual Concatenation … H 4 VCAT is an inverse multiplexing mechanism (round-robin) VCAT members may travel along different routes in SONET/SDH network Intermediate network elements don’t need to know about VCAT (unlike contiguous concatenation that is handled by all intermediate nodes) Y(J)S Eo. S Slide 48

SDH virtually concatenated VCs VC VC-11 -Xv Capacity (Mbps) if all members in one VC 1. 600, 3. 200, … 1. 600 X in VC-3 X ≤ 28 C ≤ 44. 800 in VC-4 X ≤ 64 C ≤ 102. 400 VC-12 -Xv 2. 176, 4. 352, … 2. 176 X in VC-3 X ≤ 21 C ≤ 45. 696 in VC-4 X ≤ 63 C ≤ 137. 088 VC-2 -Xv 6. 784, 13. 568, …, 6. 784 X in VC-3 X ≤ 7 C ≤ 47. 448 in VC-4 X ≤ 21 C ≤ 142. 464 So we have many permissible rates 1. 600, 2. 176, 3. 200, 4. 352, 4. 800, 6. 400, 6. 528, 6. 784, 8. 000, … Y(J)S Eo. S Slide 49

SONET virtually concatenated VTs VT Capacity (Mbps) VT 1. 5 -Xv 1. 600, 3. 200, … 1. 600 X If all members in one STS in STS-1 X ≤ 28 C ≤ 44. 800 in STS-3 c X ≤ 64 C ≤ 102. 400 VT 2 -Xv 2. 176, 4. 352, … 2. 176 X in STS-1 X ≤ 21 C ≤ 45. 696 in STS-3 c X ≤ 63 C ≤ 137. 088 VT 3 -Xv 3. 328, 6. 656, … 3. 328 X in STS-1 X ≤ 14 C ≤ 46. 592 in STS-3 c X ≤ 42 C ≤ 139. 776 VT 6 -Xv 6. 784, 13. 568, … 6. 784 X in STS-1 X ≤ 7 C ≤ 47. 448 in STS-3 c X ≤ 21 C ≤ 142. 464 So we have many permissible rates 1. 600, 2. 176, 3. 200, 3. 328, 4. 352, 4. 800, 6. 400, 6. 528, 6. 656, 6. 784, … Y(J)S Eo. S Slide 50

Efficiency comparison rate w/o VCAT efficiency with VCAT efficiency 10 STS-1 21% VT 2 -5 v 92% VC-12 -5 v 100 STS-3 c 67% VC-4 1000 STS-48 c VC-4 -16 c STS-1 -2 v 100% VC-3 -2 v 42% STS-3 c-7 v 95% VC-4 -7 v Using VCAT increases efficiency to close to 100% ! Y(J)S Eo. S Slide 51

PDH VCAT overhead octet 1 st frame of 4 E 1 s TS 0 Recently ITU-T G. 7043 expanded VCAT to E 1, T 1, E 3, T 3 Enables bonding of up to 16 PDH signals to support higher rates Only bonding of like PDH signals allowed (e. g. can’t mix E 1 s and T 1 s) Multiframe is always per G. 704/G. 832 (e. g. T 1 – ESF 24 frames, E 1 16 frames) 1 byte per multiframe is VCAT overhead (SQ, MFI, MST, CRC) Supports LCAS (to be discussed next) each E 1 time Y(J)S Eo. S Slide 52

VCAT overhead octet PDH VCAT overhead octet frames of an E 1 … TS 0 There is one VCAT overhead octet per multiframe, so net rate is T 1: (24*24 -1=) 575 data bytes per 3 ms. multiframe = 191. 666 k. B/s E 1: (16*30 -1=) 495 data bytes per 2 ms multiframe = 247. 5 k. B/s T 3 and E 3 can also be used We will show the overhead octet format later (when using LCAS, the overhead octet is called VLI) Y(J)S Eo. S Slide 53

Delay compensation 802. 1 ad Ethernet link aggregation cheats – each identifiable flow is restricted to one link – doesn’t work if single high-BW flow VCAT is completely general – works even with a single flow VCG members may travel over completely separate paths so the VCAT mechanism must compensate for differential delay Requirement for over ½ second compensation Must compensate to the bit level but since frames have Frame Alignment Signal the VCAT mechanism only needs to identify individual frames Y(J)S Eo. S Slide 54

VCAT buffering Since VCAT components may take different paths At egress the members are no longer in the proper temporal relationship VCAT path termination function buffers members and outputs in proper order (relying on POH sequencing) (up to 512 ms of differential delay can be tolerated) VCAT defines a multiframe to enable delay compensation – length of multiframe determines delay that can be accommodated H 4 byte in member’s POH contains : n sequence indicator (identifies component) (number of bits limits X) n MFI multiframe indicator (multiframe sequencing to find differential delay) Y(J)S Eo. S Slide 55

Multiframes and superframes Here is how we compensate for 512 ms of differential delay 512 ms corresponds to a superframe is 4096 TDM frames (4096*0. 125 m=512 m) For HOS SDH VCAT and PDH VCAT (H 4 byte or PDH VCAT overhead) The basic multiframe is 16 frames So we need 256 multiframes in a superframe (256*16=4096) The Multi. Frame Indicator is divided into two parts: n n MFI 1 (4 bits) appears once per frame – and counts from 0 to 15 to sequence the multiframe MFI 2 (8 bits) appears once per multiframe – and counts from 0 to 255 For LOS SDH (bit 2 of K 4 byte) – a 32 bit frame is built and a 5 -bit MFI is dedicated – 32 multiframes of 16 ms give the needed 512 ms Y(J)S Eo. S Slide 56

LCAS Link Capacity Adjustment Scheme Y(J)S Eo. S Slide 57

LCAS is defined in G. 7042 (also numbered Y. 1305) LCAS extends VCAT by allowing dynamic BW changes LCAS is a protocol for dynamic adding/removing of VCAT members – hitless BW modification – similar to Link Aggregation Control Protocol for Ethernet links LCAS is not a “control plane” or “management” protocol – it doesn’t allocate the members – still need control protocols to perform actual allocation LCAS is a “handshake” protocol – it enables the path ends to negotiate the additional / deletion – it guarantees that there will be no loss of data during change – it can determine that a proposed member is ill suited – it allows automatic removal of faulty member Y(J)S Eo. S Slide 58

LCAS – how does it work? LCAS is unidirectional (for symmetric BW need to perform twice) LCAS functions can be initiated by source or sink J 1 B 3 C 2 G 1 F 2 H 4 F 3 K 3 N 1 POH LCAS assumes that all VCG members are error-free – LCAS messages are CRC protected LCAS messages are sent in advance – sink processes messages after differential compensation – message describes link state at time of next message – receiver can switch to new configuration in time LCAS messages are in the upper nibble of – H 4 byte for HOS SONET/SDH – K 4 byte for LOS SONET/SDH – VCAT overhead octet for PDH – VCAT and LCAS Information LCAS messages employ redundancy – messages from source to sink are member specific – messages from sink to source are replicated Y(J)S Eo. S Slide 59

LCAS control messages LCAS adds fields to the basic VCAT ones Fields in messages from source to sink: – MFI Multi. Frame Indicator – SQ Se. Quence indicator (member ID inside VCAT group) – CTRL Con. TRo. L (IDLE, being ADDed, NORMal, End of Sequence, Do Not Use) – GID Group Identification (identifies VCAT group) Fields in messages from sink to source (identical in all members): – MST Member Status (1 bit for each VCG member) – RS-Ack Re. Sequence Acknowledgement Fields in both directions – CRC Cyclic Redundancy Code The precise format depends on the VCAT type (H 4, K 4, PDH) Note: for H 4 format SQ is 8 bits, so up to 256 VCG members for PDH SQ is only 4 bits, so up to 16 VCG members Y(J)S Eo. S Slide 60

reserved fields MFI 2 bits 1 -4 MFI 2 bits 5 -8 CTRL 0 0 0 GID 0 0 CRC-8 bits 1 -4 CRC-8 bits 5 -8 MST bits more MST bits 0 0 0 RS-ACK 0 0 0 0 SQ bits 1 -4 SQ bits 5 -8 MFI 1 0 0 0 1 0 0 0 1 1 0 1 0 0 0 1 1 0 0 1 1 1 1 0 0 0 1 0 0 1 1 0 1 1 0 0 1 1 0 1 1 1 1 0 1 1 16 frame multiframe reserved fields H 4 format Y(J)S Eo. S Slide 61

H 4 format – some comments CRC-8 (when using K 4 it is CRC-3) – covers the previous 14 frames (not sync’ed on multiframe) – polynomial x 8 + x 2 + x + 1 MST – – – each VCG member carries the status of all members so we need 256 bits of member status this is done by muxing MST bits there are MST bits per multiframe and 32 multiframes in an MST multiframe no special sequencing, just MFI 2 multiframe mod 32 GID – single bit - cycles through 215 -1 LFSR sequence Y(J)S Eo. S Slide 62

reserved fields MFI 2 bits 1 -4 MFI 2 bits 5 -8 CTRL 0 0 0 GID 0 0 CRC-8 bits 1 -4 CRC-8 bits 5 -8 MST bits more MST bits 0 0 0 RS-ACK 0 0 0 0 SQ MFI 1 0 0 0 1 0 0 0 1 1 0 1 0 0 0 1 1 0 0 1 1 1 1 0 0 0 1 0 0 1 1 0 1 1 0 0 1 1 0 1 1 1 1 0 1 1 16 frame multiframe reserved fields VLI format Y(J)S Eo. S Slide 63

LCAS – adding a member (1) When more/less BW is needed, we need to add/remove VCAT members Adding/removing VCAT members first requires provisioning (management) LCAS handles member sequence numbers assignment LCAS ensures service is not disrupted Example: to add a 4 th member to group “ 1” GID=g SQ=1 CTRL=NORM Initial state: GID=g SQ=2 CTRL=NORM GID=g SQ=3 CTRL=EOS Step 1: NMS provisions new member source sends CTRL=IDLE for new member sink sends MST=FAIL for new member GID=g SQ=1 CTRL=NORM GID=g SQ=2 CTRL=NORM GID=g SQ=3 CTRL=EOS GID=g SQ=FF CTRL=IDLE Y(J)S Eo. S Slide 64

LCAS – adding a member (2) Step 2: source sends CTRL=ADD and SQ sink sends MST=OK for new member l if it has been provisioned l if receiving new member OK l if it is able to compensate for delay otherwise it will send MST=FAIL and source reports this to NMS GID=g SQ=1 CTRL=NORM GID=g SQ=2 CTRL=NORM GID=g SQ=3 CTRL=EOS GID=g SQ=4 CTRL=ADD Step 3: source sends CTRL=EOS for new member starts to carry traffic sink sends RS-ACK GID=g SQ=1 CTRL=NORM GID=g SQ=2 CTRL=NORM GID=g SQ=3 CTRL=NORM GID=g SQ=4 CTRL=EOS Note 1: several new members may be added at once Note 2: removing a member is similar Source puts CTRL=IDLE for member to be removed and stops using it All member sequence numbers must be adjusted Y(J)S Eo. S Slide 65

LCAS – service preservation To preserve service integrity if sink detects a failure of a VCAT member LCAS can temporarily remove member (if service can tolerate BW reduction) GID=g SQ=1 CTRL=NORM Example: Initial state GID=g SQ=2 CTRL=NORM GID=g SQ=3 CTRL=NORM GID=g SQ=4 CTRL=EOS Step 1: sink sends MST=FAIL for member 2 source sends CTRL=DNU (special treatment if Eo. S) and ceases to use member 2 Note: if Eo. S fails, renumber to ensure Eo. S is active GID=g SQ=1 CTRL=NORM GID=g SQ=2 CTRL=DNU GID=g SQ=3 CTRL=NORM GID=g SQ=4 CTRL=EOS Step 2: sink sends MST=OK indicating defect is cleared source returns CTRL to NORM and starts using the member again Note: if NMS decides to permanently remove the member, proceed as in previous slide Y(J)S Eo. S Slide 66

Po. S Packet over SONET Y(J)S Eo. S Slide 67

Packet over SONET Currently defined in RFC 2615 (PPP over SONET) obsoletes RFC 1619 SONET/SDH path can provide a point-to-point byte-oriented full-duplex synchronous link PPP is ideal for data transport over such a link Po. S uses PPP in HDLC framing to provide a byte-oriented interface to the SONET/SDH infrastructure SONET/SDH POH signal label (C 2) indicates Po. S as C 2=16 (C 2=CF if no scrambler) Y(J)S Eo. S Slide 68

Po. S architecture IP PPP HDLC SONET/SDH Po. S is based on PPP in HDLC framing Since SONET/SDH is byte oriented, byte stuffing is employed A special scrambler is used to protect SONET/SDH timing Po. S operates on IP packets If IP is delivered over Ethernet – the Ethernet is terminated (frame removed) – Ethernet must be reconstituted at the far end – require routers at edges of SONET/SDH network Y(J)S Eo. S Slide 69

What happened to the Ethernet ? Ethernet IP Ethernet The conventional model: n Ethernet is a LAN technology – last 100 m – 10 s of hosts n IP is a WAN technology – data transported in native IP – different L 2 technologies for last segment But modern Ethernet wants to be more Y(J)S Eo. S Slide 70

Po. S Details IP packet is encapsulated in PPP – default MTU is 1500 bytes – up to 64, 000 bytes allowed if negotiated by PPP FCS is generated and appended PPP in HDLC framing with byte stuffing 43 bit scrambler is run over the SPE byte stream is placed octet-aligned in SPE – (e. g. 149. 760 Mbps of STM-1) – HDLC frames may cross SPE boundaries Y(J)S Eo. S Slide 71

RFC 2615 vs. RFC 1619 did not have the 43 bit scrambler Malicious users could generate packets n containing frame alignment pattern – deceiving framer into mis-syncing n with low transition density – degrading clock performance n containing SONET/SDH reset scrambler pattern – causing errors So RFC 2615 added the scrambler does not reset during use hard to guess proper internal state Y(J)S Eo. S Slide 72

POS problems Po. S is BW efficient but POS has its disadvantages n BW must be predetermined n HDLC BW expansion and nondeterminacy n BW allocation is tightly constrained by SONET/SDH capacities – e. g. Gb. E requires a full OC-48 pipe n POS requires removing the Ethernet headers – So lose RPR, VLAN, 802. 1 p, multicasting, etc n POS requires IP routers Y(J)S Eo. S Slide 73

LAPS Link Access Protocol over SDH X. 85 and X. 86 Y(J)S Eo. S Slide 74

LAPS In 2001 ITU-T introduced protocols for transporting packets over SDH n X. 85 IP over SDH using LAPS n X. 86 Ethernet over LAPS Built on series of ITU “LAPx” HDLC-based protocols Use ISO HDLC format Implement connectionless byte-oriented protocols over SDH X. 85 is very close to (but not quite) IETF Po. S Y(J)S Eo. S Slide 75

X. 85 vs. X. 86 X. 85 X. 86 IP IP IP LLC LAPS LLC MAC SDH MAC IP IP IP LLC LLC MAC MAC LAPS SDH X. 85 transports IP packets if delivered over Ethernet, the Ethernet is terminated X. 86 transports Ethernet can transport all sorts of Ethernet traffic – not only IP packets Y(J)S Eo. S Slide 76

X. 85 flag address ctrl SAPI 7 E (16 b) 03 (16 b) IP Packet FCS flag (32 b) 7 E IP over SDH using LAPS address = 04 (or FF for compatibility with Po. S) SAPI = 21 for IPv 4 =57 for IPv 6 (changed to be like Po. S) Scrambler always used Can use LOP VCs, HOP VCs or STMs Y(J)S Eo. S Slide 77

MAC X. 86 reconciliation MII/GMII LAPS rate adaptation SDH Similar to X. 85 (IP over SDH using LAPS) but transports the entire Ethernet frame Provides a virtual MII/GMII interface Transparent to all Ethernet features (VLAN, P bits, RPR, etc. ) Rate adaptation by adding hex DD (after byte stuffing 7 D DD) Ammendment specifies use of Ethernet PAUSE frames for rate limiting flag address ctrl SAPI Ethernet frame FCS flag 7 E (16 b) 03 FE 01 DA SA T/L INFO PAD FCS (32 b) 7 E Y(J)S Eo. S Slide 78

LAPS drawbacks Only IP or Ethernet payloads Single bit errors (e. g. in flags) may cause misalignment Not very efficient HDLC BW expansion HDLC BW nondeterminacy Y(J)S Eo. S Slide 79

GFP Generic Framing Procedure Y(J)S Eo. S Slide 80

GFP architecture Defined in ITU-T G. 7041 (also numbered Y. 1303) originally developed in T 1 X 1 to fix ATM limitations (like ATM) uses HEC protected frames instead of HDLC GFP generically encapsulates client (e. g. IP, Ethernet) onto transport network (e. g. SONET/SDH, OTN) Ethernet IP HDLC other GFP – client specific part GFP – common part PDH SDH OTN other Client may be PDU-oriented (Ethernet MAC, IP) or block-oriented (Gb. E, fiber channel) GFP frames – are octet aligned – contain at most 65, 535 bytes – consist of a header + payload area Any idle time between GFP frames is filled with GFP idle frames Y(J)S Eo. S Slide 81

GFP frame structure Every GFP frame has a 4 -byte core header – 2 byte Payload Length Indicator PLI = 01, 2, 3 are for control frames – 2 byte core Header Error Control X 16 + X 12 + X 5 +1 core header so idle frames are B 6 AB 31 E 0 (Barker-like codes) Non-idle GFP frames – have ≥ 4 bytes in payload area – the payload has its own header – 2 payload modes : GFP-F and GFP-T – optionally protect payload with CRC-32 – payload is scrambled like Po. S c. HEC (2 B) payload header (4 -64 B) – entire core header is XOR’ed with B 6 AB 31 E 0 Idle GFP frames – have PLI=0 – have no payload area PLI (2 B) payload area payload optional payload FCS (4 B) Y(J)S Eo. S Slide 82

GFP payload header has – type (2 B) PTI (3 b) PFI EXI (4 b) – type HEC (CRC-16) UPI (8 b) – extension header (0 -60 B) either null or linear extension (payload type muxing) – extension HEC (CRC-16) type (2 B) t. HEC (2 B) extension header (0 -58 B) e. HEC (2 B) type consists of – Payload Type Identifier (3 b) l PTI=000 for client data l PTI=100 for client management (OAM d. LOS, d. LOF) – Payload FCS Indicator (1 b) l PFI=1 means there is a payload FCS – Extension Header ID (4 b) – User Payload Identifier (8 b) l values for Ethernet, IP, PPP, FC, RPR, MPLS, etc. Y(J)S Eo. S Slide 83

GFP modes GFP-F - frame mapped GFP Good for PDU-based protocols (Ethernet, IP, MPLS) or HDLC-based ones (PPP) Client PDU is placed in GFP payload field GFP-T – transparent GFP Good for protocols that exploit physical layer capabilities In particular 8 B/10 B line code used in fiber channel, Gb. E, FICON, ESCON, DVB, etc Were we to use GFP-F would lose control info, GFP-T is transparent to these codes Also, GFP-T needn’t wait for entire PDU to be received (adding delay!) Y(J)S Eo. S Slide 84

GFP-T Main application – Storage Area Networks (SAN) SANs use 8 B/10 B line code and are very delay sensitive 8 B/10 B line code maps each of the 256 values of the 8 -bit input into 1 or 2 different 10 bit words Maintains a running 0 -1 balance and when encoding an input with 2 possibilities, it chooses the one that improves the balance spare 10 b symbols are used as control codes (e. g. start/end of frame) Were we to use GFP-F would lose control info, GFP-T is transparent to these codes Also, GFP-T needn’t wait for entire PDU to be received (adding delay!) GFP-T maps 8 B/10 B line code into 64 B/65 B block code Y(J)S Eo. S Slide 85

GFP-F Client packet/frame without un-needed overhead (e. g. flags, preamble, etc) is placed in GFP payload field Interface is at link layer More BW efficient than GFP-T since idle periods are filtered out preambles, frame-start, etc are also not transported GFP-F must know the client protocol in order to detect frames Can mux different client protocols on a frame to frame basis If the client protocol has a good FCS, don’t need to use GFP’s FCS GFP-F is used for Eo. S Either IP in PPP or native Ethernet can be used Y(J)S Eo. S Slide 86

GFP advantages Supports multiple protocols (not just Ethernet and IP) For Ethernet, GFP can transparently transport entire frame Robust – single bit errors do not cause loss of alignment Constant predictable overhead Good efficiency (similar to LAPS best case) GFP-T for SAN support Can run over OTN (G. 709) as well as SONET Y(J)S Eo. S Slide 87

Alternatives Y(J)S Eo. S Slide 88

There are yet other ways … n Ethernet in the first mile (EFM) n WAN-PHY (10 GBASE-W) n Ethernet over wavelengths (Eo. W) or OTN (G. 709) n Ethernet over Resilient Packet Rings (RPR) n Ethernet pseudowires (PWs) Y(J)S Eo. S Slide 89

Ethernet in the First Mile IEEE 802. 3 ah task force produced the EFM definition Optical technologies n point to point optical fiber @ 100 Mbps 10 km – Dual fiber duplex 100 Base-LX 10 – Single fiber simplex 100 Base-BX 10 n point to point optical fiber @ 1 Gbps 10 km – Dual fiber duplex 1000 Base-LX 10 – Single fiber simplex 1000 Base-BX 10 n point to multipoint optical fiber @ 1 Gbps 10/20 km (EPON ) – Single fiber simplex 1000 Base-PX 10/20 Copper technologies n point to point copper @ 10 Mbps 750 m (short reach PHY) – VDSL 10 PASS-TS n point to point copper @ 2 Mbps 2. 7 km (long reach PHY) – SHDSL. bis 2 Base-TL – up to 45 Mbps by bonding OAM Y(J)S Eo. S Slide 90

WAN-PHY (10 Gb. E in STM-64) 10 GBASE-W 802. 3 -2005 Clause 50 G. 707 Annex F There is a special case where Ethernet and SDH bit-rates are close STM-64 is 9953. 28 Mbps Gb. E 10 GBASE-R (64 B/66 B coding) can be directly mapped into a STM-64 (with contiguous concatenation) without need for GFP MAC creates "stretched Inter. Packet Gap" to compensate for rate being < 10 G This is the fastest connection commonly used for Internet traffic Complication: SDH clock accuracy is 4. 6 ppm, Gb. E accuracy is 20 ppm 64*(270 -9) = 16704 columns J 1 63 columns of fixed stuff Y(J)S Eo. S Slide 91

Ethernet over Wavelengths Rather than muxing Ethernet flows using SONET mechanisms We can allocate a separate wavelength (lambda) per flow Wavelength Division Multiplexing (WDM) For example, each wavelength may support OC-48 (2. 5 Gbps) Up to 8 channels is called coarse CWDM More than 8 wavelengths (20 Gbps) is called dense DWDM Present DWDM technology allows about 80 channels Higher densities expected soon DWDM’s tight channel spacing requires expensive cooled laser sources Y(J)S Eo. S Slide 92

Ethernet PWs Customer Edge Pseudowire (PW): mechanism that emulates essential attributes of a native service while transporting over a PSN (CE) Customer Edge MPLS network Provider Edge (CE) (PE) Customer Edge Ethernet (CE) MPLS label stack Pseudo. Wires (PWs) PW label PWE control word (CE) Ethernet frame (with or w/o FCS) Y(J)S Eo. S Slide 93