Using Cut-Through Forwarding to Retain the Real ... - Semantic Scholar

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, PAPER NO. 13

1

Using Cut-Through Forwarding to Retain the Real-Time Properties of Profibus over Hybrid Wired/Wireless Architectures Christos Koulamas, Stavros Koubias, Member, IEEE and George Papadopoulos, Member, IEEE

Abstract— The benefits of wireless extensions in industrial networks are well recognized as long as the integration of the wireless and fieldbus domains will be capable of retaining the real-time and dependability quality of the current wired industrial networking solutions. In this paper, the architecture and the operation of a cut-through forwarding device are described, to be used in broadcasting, hybrid wired/wireless Profibus systems. Analytical models of the delay overhead introduced due to frame forwarding are presented. It is shown that the usage of cutthrough forwarding devices relaxes the bit-rate requirements in the radio segments, while it drastically improves the inherent advantages and reduces the drawbacks of hybrid transmission media architectures, which are based on a single MAC domain. Index Terms— Cut-through forwarding, real-time systems, wireless fieldbuses, Profibus

I. I NTRODUCTION

T

HE maturity of digital radio communications and wireless LAN technologies has been already proven by the recent product evolution and their high market penetration and adoption from users in the home and office environments. At the same time, the benefits of wireless communications are well recognized in other areas, such as industrial control and automation, since they have already motivated extensive research during the last decade. The most critical differences between home/office and industrial environments are either application and technology related restrictions, which deal with real-time performance and dependability aspects, or market related peculiarities, such as the longer expected life cycles of industrial installations. With respect to the latter, compatibility with existing infrastructures and engineering practices must be considered as strict requirements for any new proposed technology or architecture [1]. Regarding the technology related restrictions, research has been focused on a number of directions, such as on the evaluation and analysis of the impact of error prone communication links on existing industrial communications protocols [2] or on the proposal of enhancements for such protocols to compensate possible link Manuscript received August 30, 2003; revised May 31, 2004. This work was supported by the European Commission under the project name RFIELDBUS (IST-1999-11316). C. Koulamas is with Industrial Systems Institute (I.S.I.), 26500 Rio, Greece (phone:+30 2610 996226; fax: +30 2610 910333; e-mail: [email protected]). S. Koubias is with the Dept. of Electrical Engineering & Computer Technology, University of Patras, Greece (e-mail: [email protected]). G. Papadopoulos is with Industrial Systems Institute (I.S.I.), 26500 Rio, Greece (email: [email protected]). Digital Object Identifier 00.0000/XXXX.200x.000000

impairments [3]. Other approaches focus on the development of more robust and highly immune to noise and multipath fading radio physical layers, possibly along with accurate definitions of planning guidelines, parameter sets and limit values in order to guarantee a near-wireline communication quality and robustness [4]. Moreover, much attention has also been paid to architectural issues, resulting in an already available set of different approaches being proposed, analyzed and compared, towards the construction of multi-segment, hybrid wired/wireless fieldbus systems. These are based either on physical layer repeaters/couplers [5], [6] or on higher layer bridges/routers [7], proxies [8], gateways [9] and wireless interconnecting backbones [10]. Although all of the above mentioned architectural approaches can guarantee the basic real-time requirement of time bounded communication cycles, they do not perform equally well on the absolute values or on the jitter of the end to end transfer delays, which may be crucial when variable synchronization requirements are also assessed. Such synchronization requirements are of great importance for the stability of control loops implemented over the fieldbus, and they are usually satisfied through special services offered by each specific protocol with the assumption of the operation over a single MAC domain. This is true for example with the SYNC and FREEZE services of Profibus DP or with the synchronization variables, the variable lists and other temporal or spatial consistency services of WorldFIP, where an inter-segment provision of such services exist neither in the standards themselves nor in the existing literature which follow interconnection approaches above the MAC operation (i.e. bridges or gateways). This paper focuses on the hybrid wired/wireless architecture defined and proposed as one of the results of the IST project RFieldbus [5], especially on how the interconnection below the MAC operation and the cut-through implementation of the interconnecting devices affect the performance and the realtime characteristics of the whole fieldbus system. The paper is structured as follows: section II presents the main characteristics of the proposed system and device architectures. Section III focuses on the operational and timing characteristics of a cut-through forwarding device, providing analytical figures for the calculation of the most important, performance related, timing parameters. Section IV applies the analytical model on the basic building block of the proposed architecture, in order to show the absolute transfer delay values achieved and how these values are affected by the bit-rate characteristics of

c 200x IEEE 0000–0000/00$00.00

2

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, PAPER NO. 13

S4

S1

Data Link Layer (MAC / DLL) Forwarding Function (FF) Physical Layer 1 (PhL1)

M1 FD5

Station Manager (SM)

FD1

FD4 Cell2

Cell1 FD6

Physical Layer 2 (PhL2)

FD2 FD3 S2

S3

M3

M2

Fig. 1.

Forwarding Device (FD) Model

the interconnecting wireless medium. Section V presents the impact of such forwarding architectures on other performancerelated parameters of the fieldbus system, and provides a comparison with the best case of a bridged-based solution, whereas section VI concludes the paper. II. T HE PROPOSED SYSTEM ARCHITECTURE The two fundamental architectural characteristics of the proposed hybrid fieldbus system are the formal definition of a new wireless physical layer [11] and the specification of the inter-segment communication function as a frame forwarding function at the service interface of the interconnected physical layers [5], [11]. Consequently, the hybrid network appears with a uniform MAC protocol operation over a unified broadcasting ’hyper-medium’ consisting of a number of forwarding functions, which transparently inter-connect the different, wired or wireless, physical layer (PhL) segments. The proposed model for the new radio physical layer [11], [12] is mainly influenced by the relevant specifications of the IEC-61158-2 [13] and IEEE-802.11 [14] standards. With respect to the forwarding function (FF), a general model of the corresponding Forwarding Device (FD) in which it is hosted, is presented in Fig. 1. It must be noticed that not all protocols are equivalently suitable to such hybrid architectures, while in the context of the RFieldbus project the Type 3 (Profibus) protocol was selected as the enabling platform for the prototypes development and the pilot installations. This was mainly due to the fact that the Profibus Data Link Layer (DLL) Protocol Data Unit (PDU) encodes directly or indirectly its length in the beginning of the frame and because the standard defines time-out and inter-frame register sizes large enough to hold the extended values imposed by the proposed architecture. The selection of a particular protocol automatically defines the available wire-line physical layer combinations, bit-rates and device types. In the Profibus case, and for high speed traffic, the asynchronous transmission frame format over RS485 lines is used. In Fig. 2 an example of such a hybrid topology is presented, which also depicts the two different types of stations defined, namely the master (M) and slave (S) stations. Although both wire-line and wireless terminal stations (M or S) have been defined in the proposed architecture [5], Fig. 2 depicts a hybrid system which consists only of standard (i.e. existing) wire-line Profibus nodes connected to standard

Fig. 2.

Example of a Hybrid Profibus System

RS485 asynchronous wire-line PhL segments, whereas these wired segments are interconnected through the new radio PhL by the usage of FDs. It is foreseen that such topologies will present the initial roadmap for faster deployment of wireless extensions to existing installations, due to the long expected life time of industrial installations and due to the fact that the forwarding nodes are more general products, in contrast to complete Profibus nodes which usually include additional application related components (specific sensor, actuator, PLC etc). Nevertheless, the analysis presented in section III is equivalently applicable to any valid hybrid topology, according to [5]. III. C UT-T HROUGH FORWARDING TIMING MODEL In general, the transmission time TF of a frame with a Physical layer Service Data Unit (PSDU) size of ’N’ bytes is calculated as TF = TP OV + TP SDU = TP ST + N ×BF + TT R = TP R + THD + N ×OF ×TBIT + TP T R

(1)

where, OF equals to the number of bits used for the transmission of a byte of information, TBIT equals to the transmission time of 1 bit in the medium, TP R corresponds to the preamble transmission time possibly needed from the selected PhL technology and THD , TP T R correspond to the transmission time of additional bytes in a possible header and trailer respectively, in the Physical layer Protocol Data Unit (PPDU). Moreover, BF stands for the transmission time of one byte of information (BF = OF ×TBIT ), TP SDU stands for the transmission time of the PSDU part (TP SDU = N ×BF ), TP ST stands for the physical layer overhead prior to the PSDU part (TP ST = TP R + THD ) and TP OV stands for the overall PhL overhead over the PSDU (TP OV = TP ST + TP T R ). To define the context of the analysis presented in this section, a snapshot of a general forwarding operation is depicted in Fig. 3 [16]. At time zero, one of the ports of the FD becomes active at the medium, indicating the starting point of a frame reception. For a number of restrictions discussed below, this point in time does not necessarily define the activation time of the forwarding transmission in the output port. The time instance, in which all the relevant requirements due to such restrictions are satisfied, is defined as the triggering time instance, TT RG , when a triggering event is assumed to be

KOULAMAS et al.: USING CUT-THROUGH FORWARDING TO RETAIN THE REAL-TIME PROPERTIES OF PROFIBUS

3

T F_FWD T FWD_OVHD T F(I) Input PhL

Preamble

Header

T PR(I)

T HD(I)

.....

Len

.....

n

BF(I) T TRG

Trailer

T PTR(I)

TP D

T PC +T PD

TP S

Output PhL

Preamble

Header

T PR(O )

T HD(O)

.....

n

Last

Trailer

BF(O )

BF(O)

T PTR(O)

T F(O )

Fig. 3.

Cut-Through Forwarding Timing Model

generated inside the FD. The time overhead TF W D OV HD of the forwarding operation, qualitatively defined as the time elapsed from the end of the frame in the input port (I) to the end of the frame in the output port (O), is quantified according to Fig. 3 by the following equation: TF W D OV HD = TT RG + TP S + (TF (O) − TF (I) )

(2)

where TP S is the time elapsed from TT RG up to the activation of the output medium, due to transmission initialization processing inside the FD. It is also evident that the overall frame forwarding transaction (TF F W D ) lasts: TF

FWD

= TF (I) + TF W D OV HD = TT RG + TP S + TF (O)

(3)

The most important inherent restriction of a cut-through forwarding operation is this of the availability of each data byte when it has to be transmitted. Since no gaps are allowed between bytes in any transmission technique used in the physical layers of [13], the TT RG time instance may have to be delayed until the above requirement is satisfied for every byte of the PSDU part. According to Fig. 3 this is quantified as TP ST (I) + n×BF (I) + TP D ≤ TT RG + TP ST (O) + (n − 1)×BF (O) , ∀n, 1≤n≤N

(4)

where TP D is the sustained event transfer latency of the device, that is the time elapsed from the appearance of any input data event up to the activation of the corresponding output data event. Moreover, since the decision upon a correctly or erroneously received frame is made at the very end of any frame’s reception and there are decisions and actions that should be taken on certain events and CRC forwarding policies (i.e. pass unchanged, reconstruct etc.), there may be an additional requirement due to such a processing overhead TP C after the end of the frame reception and prior to sending the last byte in the forwarding retransmission, as this is considered the last

chance to signal a corrupted frame. Therefore, it is derived that the time offset between the end of the whole frame reception in the input port and the starting point of the transmission of the last PSDU byte in the output port must never fall below TP C + TP D . This is described as TP ST (I) + N ×BF (I) + TP T R(I) + TP D + TP C ≤ TT RG + TP S + TP ST (O) + (N − 1)×BF (O)

(5)

which, actually, is a further constraint for (4), and can be easily merged with it by adding TP T R(I) + TP C to its left part for n = N (for the last PSDU byte). From (4) & (5) it is clear that for an FD to decide its TT RG time, the size and the data part (PSDU) bitrate information of the frame must be available. According to the PhL service specifications [13], [14], this information is made available to the forward control entity either with the RX START service primitive, if it is encoded in the PhL header and assuming that a cross-layer design approach is not adopted, or after the reception of L bytes of the PSDU, assuming that the length is encoded, directly or indirectly, in the Lth byte of the DLL PDU. In the latter case, it is evident that the forwarding entity has to decode, to a certain degree, the DLL protocol transfer rules. In conclusion, it holds min TT RG = TP ST (I) + L×BF (I) + TP D

(6)

where L may be 0, to cover the first case described above. Since (6) gives the absolute minimum possible time instance of the TT RG , and since (4) & (5) solved for TT RG may give negative (i.e. unreal) results, it seems more appropriate to define the ∆TT RG , that is the possible additional delay of the TT RG time instance beyond the minTT RG value, and to identify the prerequisites for the ∆TT RG to be positive. It is defined TT RG = min TT RG + ∆TT RG = TP ST (I) + L×BF (I) + TP D + ∆TT RG

(7)

4

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, PAPER NO. 13

After merging (4) & (5) and substituting from (7), it finally holds  n×∆BF + BF (O) − P 2,     ∀n, 1 ≤ n ≤ (N −1)  N ×∆BF + BF (O) − P 2 + P 1 (8) ∆TT RG =max      0

where:

P 1 = TP T R(I) + TP C P 2 = TP S + TP ST (0) + L × BF (0) ∆BF = BF (I) − BF (O)

(9)

Equation (8) shows that, if ∆BF = 0 (i.e. BF (I) = BF (O) ) then the maximum ∆TT RG for a frame is derived for n = N , since P1 is always zero or positive number. Then, ∆TT RG ≥ 0 happens if and only if it is BF (O) + P 1 ≥ P 2. Moreover, if ∆BF ≥ 0 (i.e. BF (I) > BF (O) ) then ∆TT RG increases with ’n’. Therefore, the worst case is again for n = N , that is, defined from the total length of the frame. Then, ∆TT RG is positive if and only if N > NF W D1 , where NF W D1 =

P 2−(BF (0) +P 1) ∆BF

(10)

which simply happens for all sizes N if BF (O) + P 1 > P 2. Finally, if ∆BF < 0 (i.e. BF (I) < BF (O) ) then ∆TT RG increases when ’n’ decreases, and the actual ∆TT RG for a frame forwarding is defined by either the frame’s length N or by its first byte (i.e. n=1). The first byte’s case dominates when (∆BF +BF (O) −P 2)−(N ×∆BF +BF (O) − P 2+P 1) > 0 which leads to P1 N > NF W D2 , where NF W D2 = +1 (11) −∆BF and ∆TT RG is positive only when ∆TT RG +BF (O) > P 2 or, equivalently, BF (I) > P 2. For the latter, it may be easily proven that it holds only for L = 0. If (11) does not hold, then the second branch of (8) applies, and consequently (10) with the inequality reversed (since we divide with ∆BF which is negative) for the ∆TT RG to be positive. In conclusion, ∆TT RG is finally described by the following:  N ×∆B 1,  F + BF (O) − P 2 + P    (B > B )  F (I) F (O)  if   &(N > NF W D1 )     or       (B   F (I) < BF (O) )     &(N ≤ NF W D2 ) if       &(N < NF W D1 )        BF − P2 + P 1,  (12) ∆TT RG = (BF (I) = BF (O) = BF )   if   & (BF + P 1 > P 2)         BF (I) −P 2,      (P 2 < BF (I) < BF (O) )   if   & (N > NF W D 2 )         0,    otherwise

By distinguishing the two major cases of the ∆TT RG value - i.e. being a function of the frame size or not - substitution from (12) to (2) leads to a more expressive form for the TF W D OV HD definition, which is:  BF (O) +T    P T R(O) +TP D +TP C,   (BF (I) > BF (O) )   if   &(N > NF W D1 )     or       (B   F (I) < BF (O) )    &(N ≤ NF W D2 ) if TF W D OV HD =   (13) &(N < NF W D1 )         minTT RG + ∆TT RG + TP S     + (TP OV (O) − TP OV (I) )     + N ×(BF (O) − BF (I) ),    otherwise

Equation (13) shows that in the cases where the TT RG time instance is independent from the frame size, the overhead introduced by the FD is a function of that size, increasing with a size increase if BF (O) > BF (I) , decreasing otherwise - see the second branch of (13), which covers the last three, ∆T RG branches of (12). On the other hand, whenever the TT RG time instance is a function of the frame size, the resulting forwarding overhead is constant for a given bit-rate in the output medium. The latter constant value, given from the first branch of (13) is also the minimum achievable time overhead in any frame forwarding. It must be also noticed that (13) is a general equation which applies equivalently to cut-through and store&forward forwarders. For the latter case, the second branch of (13) is used and minTT RG + ∆TT RG is substituted by the time instance in which the whole input frame is received and checked, that is TF (IN ) + TP D + TP C . In this case (i.e. store&forward), (13) becomes TF W D OV HD = TP S + TP D + TP C + TF (OU T )

(14)

which is the rather obvious result that in a store&forward operation the overhead is the total duration of the retransmission of the frame in the output port plus any processing and data transfer overhead delay inside the forwarder. IV. F ORWARDING UNDER P ROFIBUS In Profibus PhL Type 1 (RS485) the parameters are TP R =THD =TP ST =TP T R =0 and OF =11. For the length parameter (L), different options exist, since there are different frame formats defined. For the fixed size frame formats, the length is indirectly encoded in the frame type which is defined by the first byte of the frame, therefore it holds L=BF , whereas for the variable size frame format (SD2), it holds L=2×BF . In the proposed and implemented radio PhL [6], the related parameters are TP R =54usec, THD =9×BF (since the PhL frame header consists of 9 bytes transmitted with the same bit-rate with the PSDU), OF =8 and L=0 (since the length of the frame is encoded in the PhL header). Moreover, the TP T R parameter is zero for all frames except the short acknowledgement (SC) and the token telegram (SD4) which

KOULAMAS et al.: USING CUT-THROUGH FORWARDING TO RETAIN THE REAL-TIME PROPERTIES OF PROFIBUS

140

5

350 R = 1.091 Mpbs 300

Forwarding Delay (usec)

Forwarding Delay (usec)

120 100 R = 1.5 Mpbs

80

R = 1.168 Mpbs

60 40 20 R = 2 Mpbs

N = 255 250 200 150 100

N = 10 N = 50

50

N = 100 0

0

50

100

150

200

0

250

1

1.2

Profibus Telegram Size (Octets) Fig. 4.

1.4

1.6

1.8

Wire-line (1.5Mbps) to Radio Forwarding Delays

900

900 R = 2 Mpbs

N = 255 800

700 600 R = 1.5 Mpbs 500 400 300 200

R = 1.168 Mpbs

100

Forwarding Delay (usec)

800

Forwarding Delay (usec)

2

Radio PhL Bitrate (Mbps)

700 600 500 400 N = 100 300 200

N = 50

100

N = 10

R = 1.091 Mpbs 0

0

50

100

150

200

250

Profibus Telegram Size (Octets) Fig. 5.

0 0.8

1

1.2

1.4

1.6

1.8

2

Radio PhL Bitrate (Mbps)

Radio to Wire-line (1.5Mbps) Forwarding Delays

are not protected by the Frame Check Sequence (FCS) and the End Delimiter (ED) bytes. Since the parity protection of Profibus PhL Type 1 is lost in the 8-bit byte frame of the wireless transmission, these two frame types have TP T R =2×BF (W IRELESS) in order to cover the two CRC bytes of the selected frame format, which is defined from the same protocol (Profibus) but from the synchronous PhL (Type 2) specification. Fig. 4 & 5 depict the values of TF W D OV HD for a variable size Profibus frame (i.e. an SD2 telegram) with respect to its size (N) and the wireless medium bitrate (R) for a constant standard wire-line Profibus bitrate of 1.5Mbps, and for both directions, namely from wire-line to wireless (TF W D OV HD(W 2R) ) and from wireless to wire-line (TF W D OV HD(R2W ) ). In these figures, the special wireless bi-

trates of 1.091 Mbps and 1.1680Mbps correspond to the points where BW F =BRF (1.091Mbps) and NF W D 1 =maxN =255 (1.1680Mbps), respectively. The plots correspond to an ideal forwarding device with TP S =TP C =TP D =0, in order to point out the ultimate performance of such a forwarding mechanism due to only its restrictions and not due to certain implementation characteristics. What is derived from Fig. 4 & 5 is the fact that for a given standard wire-line bitrate, even though an increase in the wireless bitrate improves performance in the wire-line to wireless direction, it results in a performance degradation in the opposite direction. This is of much importance for topologies equivalent to this of Fig.2 where the frame has to pass through more than one FD and more than one wire-line segments with standard and equal bit-rates. The general form

6

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, PAPER NO. 13

900

900 N = 255

R = 2 Mpbs 800

700 600 R = 1.5 Mpbs 500 400 R = 1 Mpbs

300 200 100

Forwarding Delay (usec)

Forwarding Delay (usec)

800

700 600 500 400 N = 100 300 N = 50

200

0

50

100

150

200

250

N = 10

100 0.8

1

Profibus Telegram Size (Octets) Fig. 6.

1.2

1.4

1.6

1.8

2

Radio PhL Bitrate (Mbps)

Path (2 FD) End to End (Wire-line 1.5Mbps) Forwarding Delays

for a path forwarding delay overhead calculation is given by TF W D OV HD(P AT H) =

N FD X

TF W D OV HD(i)

(15)

i=1

N FD X

(TT RG(i) + TP S(i) )

M1_REQ

S2_RESP

M1_REQ

where NF D is assumed to be the number of the forwarding devices in the path. For the special case of a path between two wire-line stations in systems with one bitrate for the wire-line segments and another one for the radio segments, substitution of (2) in (15) gives the alternative TF W D OV HD(P AT H) =

TID

TSDR

(16)

i=1

Fig. 6 presents the plot of (16) with respect to the bitrate of the intermediate radio PhL and for NF D =2. It clearly shows that a range of wireless bit-rates exists, in which the system appears with optimal characteristics. This range depends on the maximum size of telegrams in the network, where it is noticed that by reducing this maximum size the wireless bitrate may be further increased up to a limit, above which the performance degrades significantly. Moreover, in the ideal case of FD delays, and in this optimal bitrate, the end-to-end delay overhead due to frame forwarding from one wire-line segment to another through a wireless link is constant and independent from the frame size. The actual delay values for the prototypes developed are TP S =17usec, TP C =25usec, TP D =2usec for wire-line to wireless forwarding, and TP S =32usec, TP C =35usec, TP D =1usec for wireless to wire-line forwarding, while the respective plots are nearly equivalent in quality with these given in Fig. 4, 5 & 6. More precisely, they give for Nmax=255 an optimal wireless bitrate of 1.1665 Mbps and a nearly constant delay of 183-188 usec (275 TBIT in 1.5Mbps Profibus instead of 205 TBIT of the ideal case), or, for Nmax=108, an optimal 1.2707 Mbps and an absolutely constant delay of 177 usec (265 TBIT ), for all variable size frames in a path of 2 forwarding devices.

minTSDR

M1_REQ

S2_RESP

M1_REQ

S2_RESP

TID

S1_RESP

M1_REQ

M1_REQ

M1_REQ

S1_RESP

S1_RESP

S1_RESP

M1_REQ

S1_RESP

M1_REQ

S1_RESP

minTSDR

Fig. 7.

Impact of forwarding

V. E VALUATION OF NETWORK PERFORMANCE The impact of the introduction of forwarding devices, additional physical layer segments and additional delays in a fieldbus network is that a number of parameters which coordinate each specific protocol operation have to be reconsidered. This especially applies to timeouts, either between a request and a response frame or in physical medium idle times, which have to be prolonged for the normal operation of the hybrid network (refer to the message cycle between M1 and S2 in Fig. 7, with respect to the topology of Fig. 2). Furthermore, some kind of message rate control has to be applied in cases that two or more stations communicate through the same physical medium of a hybrid system or that a single station is continuously transmitting unacknowledged frames, as it is also proposed in [15], [16] (refer to the propagation of message cycles between M1 and S1 in Fig. 7). The advantages and drawbacks of broadcasting hybrid architectures based on a single MAC operation over the whole set of PhL segments in the system, as the one described in Section II, have been discussed in [7], [8]. Among the most crucial drawbacks of such architectures seem to be the lower network utilization achieved, due to higher medium idle times while waiting for responses (i.e. the maxTSDR parameter in Profibus) and the system responsiveness to failures, as this is governed by timeout values (the TSLOT parameter, regarding

KOULAMAS et al.: USING CUT-THROUGH FORWARDING TO RETAIN THE REAL-TIME PROPERTIES OF PROFIBUS

7

(a) 2 Mbps Radio PhL

(b) 1.168 Mbps Radio PhL

Fig. 8.

Impact of the wireless PhL bit-rate in the slave response time

the Profibus protocol, which is also ruled by maxTSDR ). Considering the snapshot depicted in Fig. 7, the TSDR of S2 in response to a request from M1 (first message cycle) is maxTSDR =

maxTFREQ W D OV HD(P AT H)

+ minTSDR + max

NP FD i=1

T (t)

k l maxTF W D OV HD(kP AT H) + minTIT (O) ≤ TIT (I) k P T (t+1) + min (TT RG + TP S )i , ∀k:1≤k≤NF D i=1

(17)

(TT RG + TP S )RESP i

which, if the assumptions for (16) hold, becomes

⇒

 T (t)  maxTF W D OV HD(kP AT H)    k NF D P T (t+1) l TIT − min (TT RG +TP S )i (I) ≥ max k=1   i=1   + minT k IT (O)

maxTSDR = maxTFREQ W D OV HD(P AT H) + minTSDR

Thus, for all intermediate path lengths of k FDs, in a total path of NF D FDs (i.e. k ≤ NF D ), it must hold

(18)

+ maxTFRESP W D OV HD(P AT H) Moreover, considering the case of continuous message cycles between stations connected in the same PhL segment, or continuous broadcastings from a simple node, there is a need to extend the inter-transaction time in order to avoid queuing delays inside the FDs [15], [16]. Let T(t) stands for the tth transaction, that is, the transmission of a simple or of a specific k k number of frames in a medium, and let also TIT (I) and TIT (O) define the inter-transaction times in the two sides of the k th FD. Then, for an FD to hold at most one transaction, the output medium has to be available at the time instance in which the first bit of the first frame of a transaction is to be transmitted. This means that the forwarding of the previous transaction has to be completed, including the needed inter-transaction time.

        

(19)

k where minTIT stands for the minimum inter(O) transaction time allowed in the output medium of FD(k), T (t) TF W D OV HD(kP AT H) stands for the total forwarding delay of transaction T(t) through a path of k FDs, and T (t+1) (TT RG + TP S )i stands for the time instance of the output medium activation due to the first (or the only) frame of the following transaction forwarding in the ith FD. The importance of the cut-through operation and of the intermediate wireless medium bit-rate in the above mentioned parameters is presented in Fig. 8 and 9. With respect to the topology of Fig. 2, Fig. 8 depicts a maximum length requestresponse message cycle between master M1 and slave S2, whereas Fig. 9 presents a sequence of broadcasted, unacknowledged transactions of variable lengths (SDN Profibus telegrams) transmitted from M1. As it is shown, the reduction of the TF W D DELAY according to the graphs presented in Fig.6 results in the reduction of the total TSDR and the queue sizes in the intermediate FDs. Moreover, the independency of the TF W D OV HD over the frame size at the optimal

8

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, PAPER NO. 13

(a) 4 Mbps Radio PhL

(b) 1.168 Mbps Radio PhL

Fig. 9.

Impact of the wireless PhL bit-rate in the intermediate queuing delays

intermediate wireless bit-rate introduced in section IV, results in an exact copy of the initial traffic pattern in every PhL segment, and thus simplifies (19) as it is enough to be applied only for the basic building block of the 2 FD sub-system. On the other hand, among the most important advantages of single MAC hybrid networks are: a) the straightforward and simplified deployment, since they are not different in principle from the standard, existing wire-line solutions, b) the simplicity of the forwarding devices, since they do not have to implement the whole MAC protocol, as well as c) the minimization of the skew in the global, system-wide time, which is important to synchronization needs of control applications. In complex real-time applications where a number of variable synchronization requirements may exist inside a single scan cycle in addition to simple requirements of fast time-bounded scan cycles, the importance of the last argument usually overrules the importance of the best possible network utilization. As a similar situation may be considered this of the penetration of the Ethernet in the automation domain, where the gain from the usage of a common and well-known networking technology seems to be larger than the fact of the necessary low utilization of the medium, in order to keep it real-time friendly. Furthermore, the resulting lower network utilization may be completely canceled in a number of cases, if the advantageous side effect of the elimination of the bitrate / distance trade-offs by the existence of wireless forwarding devices is taken into account. That is, in industrial installations which use a lower bitrate than the maximum available, due to medium distance restrictions, the introduction of interconnecting wireless segments in proper places allows for the utilization of larger bitrates (3, 6 or 12 Mbps instead

of the popular 1.5 Mbps in Profibus DP), thus improving the capacity of the fieldbus system, with the minimum needed system re-planning. In bridged architectures, system-wide synchronization commands (e.g. input FREEZE in Profibus DP) capture or drive the environment conditions with a worst-case time skew equal to the sum of the maximum latencies in the intermediate segments’ MAC operation, which for multi-master Profibus segments are in the order of several msec (target rotation times). The best possible case in the bridging approach although hardly achievable, if not artificial - may be considered this of zero waiting time for medium access in the output port of the forwarding device. Even then, the FD would ideally function as a store&forward repeater, due to both theoretical and practical reasons. Theoretically, the available service primitives at the MAC service interface require the whole MAC service data unit (MSDU) to be passed at once, thus all data have to be totally received from the input port. Practically, most of the available implementations are based in protocol specific integrated circuits (ASICs) which follow the atomically interchanged double buffer approach for network variable transfers. That is, there is a separate data buffer available to the DLL or MAC user entity in order to put or get the corresponding data, and a separate buffer holding the variable data which may be under transmission or reception. In contrast, in single MAC, multi-physical segment hybrid networks the worst-case time-skew equals to the maximum TF W D OV HD delay over the longer path in the system, which may be in the order of usec due to the cut-through forwarding approach and since it is calculated for a single message cycle. In order to point out the differences between these two

KOULAMAS et al.: USING CUT-THROUGH FORWARDING TO RETAIN THE REAL-TIME PROPERTIES OF PROFIBUS

30 N = 255

SF / CT Overhead Ratio

25

20

15 N = 100 10 N = 50 5 N = 10 0 0.8

1

1.2

1.4

1.6

1.8

2

Radio PhL Bitrate (Mbps) Fig. 10. Comparison of Store&Forward (SF) and Cut-Through (CT) forwarding overhead delays

approaches, Fig. 10 depicts the ratio of the end-to-end delay in a store&forward FD (TF W D OV HD(SF ) ) to the delay in a cut-through FD (TF W D OV HD(CT ) ), which correspond to the 2-FD path scenario described by Fig.6. It clearly shows that, in the optimal radio PhL bitrate and for the maximum size Profibus telegram, the overhead is more than 25 times less when using cut-through forwarding under a single MAC domain, compared to a store&forward approach - which is actually the best case of a bridge-based interconnection over multiple MAC domains.

9

[5] RFieldbus project IST-1999-11316, D1.3: General System Architecture of the R-Fieldbus. [6] Koulamas C., Koubias S., Papadopoulos G., Real-Time Performance of Cut-Through Forwarding in Hybrid Wired/Wireless Industrial Networks, in Proceedings of the 10th International IEEE Conference on Emerging Technologies and Factory Automation, ETFA 2003, Lisbon, Portugal, Vol 1, pp 23-30. [7] Luis Ferreira, Mario Alves, Eduardo Tovar, Hybrid Wired/Wireless PROFIBUS Networks Supported by Bridges/Routers, 4th IEEE International Workshop on Factory Communication Systems, Sweden, August 2002, pp 193-202. [8] Aslanis S., Koulamas C., Koubias S., Papadopoulos G., Architectures for an integrated hybrid (wired/wireless) fieldbus, ETF Journal of Electrical Engineering, Vol 9-10 Oct. 2001, pp.5-13. [9] Suk Lee, Kyung Chang Lee, Man Hyung Lee, Fumio Harashima, Integration of Mobile Vehicles for Automated Material Handling Using Profibus and IEEE 802.11 Networks, IEEE Transactions on Industrial Electronics, Vol.49(3), pp. 693-701, June 2002. [10] S. Cavalieri, D. Panno, A novel solution to interconnect Fieldbus systems using IEEE wireless LAN technology, Computer Standards & Interfaces 20, Elsevier Science 1998. [11] RFieldbus project IST-1999-11316, D2.1.1: Physical Layer Specification. [12] Koulamas C, Lekkas A. Kalivas G., Koubias S., Papadopoulos G, Physical Layer for Industrial Radio Fieldbus Networks, Proceedings of the IST Mobile & Wireless Telecommunications Summit, Greece, 2002, pp.643-648. [13] IEC 61158-2, IEC Standard 1158-2, Fieldbus Standard for Use in Industrial Control Systems - Part2 Physical Layer Specification and Service Definition. [14] International Standard ISO/IEC 8802-11 ANSI/IEEE Std 802.11, First edition, ISBN 0-7381-1658-0, 1999. [15] Alves M., Tovar E., Vasques F., On the Adaptation of Broadcast Transactions in Token-Passing Fieldbus Networks with Heterogeneous Transmission Media, 4th IFAC Conference on Fieldbus Systems and their Applications, - FET ’01, November 2001, pp. 278-284. [16] RFieldbus project IST-1999-11316, D2.3: Real-Time Performance of the Lower Layer Subsystem.

VI. C ONCLUSION In this paper, the architecture and operation of a cut-through forwarding device and a corresponding network architecture of broadcasting, hybrid wired/wireless Profibus systems were described. Analytical models of the delay introduced during a frame forwarding and propagation were also presented. It was pointed out that there are certain bitrate relationships between the various physical layer segments for which the real-time communication characteristics are optimal. Moreover, it was shown how the usage of cut-through forwarding in single MAC, multiple PhL architectures: a) lowers the bitrate requirements in the interconnecting radio segments, b) retains the quality of the real-time synchronization services offered by the fieldbus in a more efficient and straightforward way compared to bridged solutions and c) reduces the medium idle times thus increases network capacity compared to store&forward alternatives. R EFERENCES [1] RFieldbus project IST-1999-11316, D1.1, Requirements for the RFieldbus System, Apr. 2000. [2] Andreas Willig, Adam Wolisz, Ring Stability of the PROFIBUS Token Passing Protocol over Error Prone Links, IEEE Transactions on Industrial Electronics, 48(5), pp. 1025-1033, Oct. 2001. [3] Andreas Willig, The Adaptive-Intervals MAC Protocol for a wireless PROFIBUS, in Proc. IEEE International Symposium on Industrial Electronics, 2002. [4] RFieldbus project IST-1999-11316, D2.1.2: Radio Interface Implementation

Christos Koulamas received the diploma degree in Computer and Informatics Engineering from the University of Patras, Greece, in 1994, and is expected to receive the PhD degree in Electrical and Computer Engineering from the same university, in 2004. He is currently a Researcher at the Industrial Systems Institute in Patras, Greece. Since 1994 he has participated in several national and EU projects in the areas of data communications, real-time embedded systems and software architectures, as a research engineer at the Computer Technology Institute (CTI), the Industrial Systems Institute (ISI) and the Applied Electronics Laboratory (Dept. of Electrical and Computer Engineering, University of Patras, Greece). His research interests include real-time communications, wireline / wireless fieldbus systems, realtime distributed software architectures and machine vision systems. He is a member of the Technical Chamber of Greece (TEE).

10

Stavros Koubias received the Diploma in Electrical Engineering and the Ph.D degree from the University of Patras, Greece, in 1976 and 1982 respectively. Since 2004 he has been a Professor in the Department of Electrical Engineering of the University of Patras, Greece. From 1976 to 2004 he was a Research Associate, Lecturer, Assistant and Associated Professor at the Department of Electrical Engineering, University of Patras, Greece. From March 2001 he is the Director of the Network Operation Center (NOC) of the University of Patras. From June 1998 he is member of the Industrial Systems Institute (ISI) of Patras. His main current research interests include real-time distributed systems, communication protocols, advanced wired and wireless industrial networking systems, advanced interoperable industrial networking architectures, interworking of heterogeneous control systems, distributed enterprise systems, advanced building networking systems, embedded systems, advanced microprocessor architectures, as well as industrial machine vision systems.He has participated as co-ordinator and partner in many big Greek and European R&D programs. He has published over 100 papers in miscellaneous international journals and conference proceedings. He has also written books and course notes in industrial networks and microprocessor systems. Prof. Koubias is a member of IEEE and Technical Chamber of Greece (TEE).

George Papadopoulos received his BEE degree at the City College of New York in 1963 and his Master’s and Ph.D. degrees at M.I.T. in 1964 and 1970 respectively, all in electrical engineering. He is Professor in the Department of Electrical and Computer Engineering at the University of Patras, where he directs the Applied Electronics Laboratory. In the period 1998-2003 he was director of the Industrial Systems Institute (independent institute under the auspices of the General Secretariat of Research and Technology in the Ministry of Development). Also, in the period 1997-2001 he was chairman of the Scientific Board of the Greek National Research Center DEMOKRITOS. He has served as Chairman of the ECE Department (1991-1993) and Head of the Electronics and Computers Division (1994-96), Visiting Professor in the ECE Department of the University of MASS in Amherst (1984-85), and Director of the Office for Naval Research and Technology in the Department of National Defense (1987-88). His current research interests include the following areas: networked embedded systems, wireless communication networks, RF electronics, methodologies for integrated systems design and real-time embedded systems. Over the last 20 years, 25 Ph.D. theses have been completed under his supervision and more than 150 diploma thesis. He has participated as coordinator, partner or subcontractor in many EU funded projects. In parallel, he was the responsible person for research projects funded by Greek agencies.

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, PAPER NO. 13