OPTIMAL GUARANTEED SERVICES TIMED TOKEN (OGSTT) MEDIA ACCESS CONTROL (MAC) PROTOCOL FOR NETWORKS THAT

Transcription

1 OPTIMAL GUARANTEED SERVICES TIMED TOKEN (OGSTT) MEDIA ACCESS CONTROL () PROTOCOL FOR NETWORKS THAT SUPPORT HARD R AND NON R S. Ozuomba 1, C. O. Amaefule 2, J. J. Afolayan 3 1, 3DEPT. OF ELECTRICAL/ELECTRONIC AND COMPUTER ENGINEERING, UNIVERSITY OF UYO, AKWA IBOM, NIGERIA 2DEPARTMENT OF ELECTRICAL ENGINEERING, IMO STATE UNIVERSITY, OWERRI, NIGERIA E-mals addresses: 1 smeonoz@yahoo.com, 2 lttlekr222@yahoo.com, 3 afolayan.jmoh@yahoo.com Abstract In networks that support real-tme traffc and non-rea real-tme traffc over the same physcal nfrastructure, the challenge to the Meda Access Control () protocol of such network s the ablty to support the dfferent traffc wthout compromsng qualty of servce (QoS) for any of them. Generally, tmed-token token protocols group the dverse real-tme traffc nto one category and then dedcate certan porton of the avalable bandwdth to them. At the same tme, some bandwdth are left unassgned but avalable to the non real-tme traffc. The unassgned bandwdth, and n some cases, the unused bandwdth left by the real-tme traffc are assgned to the non-real real-tme traffc on best effort bass. In ths paper, Optmal Guaranteed Servces Tmed Token (OGSTT) protocol s developed and analyzed. In order to provde better support for both real-tme traffc and non-real real-tme on the same local area network, OGSTT employs the tmed-token token mechansms n the Tmely-Token Token protocol along wth that of Budget Sharng Token (BuST) protocol. Some bounds on the behavor of OGSTT protocol are dscussed along wth the ablty of OGSTT protocol to support real-tme and non-real tme traffc. In partcular, the paper demonstrated that the performance acheved by OGSTT s better than the Tmely- Token and BuST. Furthermore, OGSTT protocol can be ncorporated nto the Ethernet network to provde real- tme performance guarantees to multmeda applcatons and hard and soft real-tme traffc. Keywords: Tmed-Token, Protocol, Ethernet, Tmely-Token, multservce, Networks, Real-Tme, Non-Real- Tme Traffc, Meda, Access Control, qualty of servce. 1. Introducton Nowadays, there s a rapd advent and advancements of many new and ectng applcatons: mage processng and transmsson, multmeda communcatons, offce and factory automaton, embedded real-tme dstrbuted systems, space vehcle systems, and the ntegraton of epert systems nto avoncs and ndustral process controls. The stuaton has placed an ncreasng demand for effectve and effcent multservces local area networks. Such networks protocols must deal wth dfferent traffc patterns and must provde not only bounded message transmsson tme, as requred by the hard real-tme tasks, but also hgh throughput, as demanded by soft real-tme and other non-real-tme tasks [1]. An attractve approach for ntegratng such traffc s the tmed-token protocol. Consequently, the tmedtoken protocol has been ncorporated nto several hgh-bandwdth network standards [2], ncludng IEEE802.4 Token Bus [3], FDDI [4-8], SAFENET [9], Manufacturng Automaton Protocol (MAP) [10], Ngeran Journal of Technology (NIJOTECH) Vol. 32. No. 3. November 2013, pp Copyrght Faculty of Engneerng, Unversty of Ngera, Nsukka, ISSN: Hgh-Speed Rng Bus [11] and n PROFIBUS whch s a Feldbus network standard [12]. The dea behnd the Tmed-Token protocol s, frst, separate the messages generated n the system at run tme nto two classes, namely, real-tme and nonreal-tme messages. Real-tme messages are transmtted perodcally and have a deadlne, whle non-real-tme messages are transmtted on a besteffort bass. Durng ntalzaton, the Target Token Rotaton Tme (TTRT) s selected. TTRT represents the epected tme needed by the token to complete an entre round-trp of the network. Each node s allocated h tme unts (bandwdth) whch s a porton of the TTRT; whenever a node receves the token, t can transmt ts real-tme messages for a tme not greater than h. It can then transmt ts nonreal-tme messages f the tme elapsed snce the prevous token departure from the same node s less than the value of TTRT, that s, only f the token arrves earler than epected. FDDI tmed-token s one of the earlest tmed-token passng protocol. In FDDI, the token rotaton tme may reach 2(TTRT) [6]. Due to ths token lateness, Ngeran Journal of Technology Vol. 32, No. 3, November

2 AND NON an FDDI network can use at most half of ts bandwdth to transmt real-tme traffc [5, 13, 14]. To allevate ths defcency, Shn et al. proposed the FDDI-M token protocol [5]. In FDDI-M, the token s never late. Ths allows FDDI-M to double FDDI s ablty to support real-tme traffc. However, FDDI-M has one major weakness; starvaton of non-real-tme traffc. Ths means that n some cases, FDDI-M may not be able to transmt non real-tme traffc. Budget Sharng Token (BuST) protocol [14, 15] and Tmely- Token [13] protocols are tmed-token protocols recently ntroduced to mprove the communcaton servces provded by FDDI and FDDI-M networks. The BuST and Tmely-Token solved the problems of token-lateness n FDDI and the starvaton of nonreal-tme traffc n FDDI-M. The network and message models are presented n Secton 2. Tmely-Token and BuST protocols, along wth ther weaknesses, are descrbed n Secton 3. The OGSTT protocol s descrbed n Secton 4. Also, the performance bounds of the OGSTT protocol s presented n secton 4. Secton 5 compares the OGSTT aganst Tmely-Token and BuST protocols. Also, n secton 5, sample numercal eample and dscusson of results are presented. Fnally, concluson and recommendatons for further studes are gven n Secton Revew of Relevant Lterature 2.1 Network Model The tmed-token protocols n ths paper operate on a token rng network consstng of N nodes. Each node has a unque number n the range 0, 1, 2 N-1. For each node, the net node s node (+1) or more approprately node (+1) mod N. The token frame crculates around the rng from node to nodes + 1, + 2, untl node + (N-1), then to nodes, + 1, +2,, etc. Let w denote the latency or walk-tme between a node and ts upstream neghbor node ( + 1). The sum of all such latences n the rng s known as the rng latency or the token walk-tme, W, where W = N 1 w length, C, s the amount of tme needed to transmt a mamum sze message. Perod length, P s the mnmum nter-arrval perod for the real-tme message stream at node. Message deadlne; D s the mamum amount of tme that can elapse between a message arrval and the completon of ts transmsson. Thus, f a message stream arrves at tme t, then t must be transmtted by tme t + D. Smlar to the Tmely-Token n [13], t wll be assumed that D P. Furthermore, n the followng dscusson, t s assumed that the network s free from hardware or software falures. 2.3 Operaton of the Estng Tmed-Token Token Protocols ols Generally, n the tmed-token protocols, durng the ntalzaton, each node declares a Target Token Rotaton Tme, TTRT. The mnmum declared value s selected as the rng's TTRT. Each node s then assgned a porton h of the TTRT to transmt ts realtme traffc. When a node receves the token, t can transmt ts real-tme traffc for a tme not greater than h tme unts. However, to ntalze the tmers, no packets are transmtted durng the frst token rotaton. The man dfference among the varous tmed-token protocols concerns the non-real-tme message servce. Let H be defned as, h where h s the sum of the tme reserved for the real-tme traffc n all the nodes n every cycle. Let T = H +W, where T s the total tme allocated per cycle to the real-tme traffc and walk-tme. The value of TTRT s denoted asτ. In the tmed-token protocols, there are two categores of bandwdths that can be used by the non-real-tme traffc, namely; Category I: (τ-t) whch s the total bandwdth that s not allocated to the real-tme traffc and rng latency. τ -T bandwdth (tme unts) s avalable to the nonreal-tme traffc n every cycle. Let A* = τ -T Category II: (U) ( whch s the bandwdth that s allocated to the real-tme traffc but not used by the real-tme traffc n the prevous cycle. The dfferent tmed-token protocols dffer n the way they allocate the two categores of avalable bandwdth to the non-real-tme traffc. As shown n [13-15], [17], [18], the tmely-token protocol [13] and the BuST protocol [15] mproved on the ablty of the FDDI and FDDI-M tmed-token protocols to support real-tme and non-real-tme traffc. 2.2 Message Model Messages generated n the system at run tme may be classfed as ether real-tme messages or nonreal-tme messages. Agrawal et al. [16] showed how a token-rng network havng multple real-tme streams per staton could be transformed nto a logcally-equvalent network wth one real-tme message stream per staton. Therefore, wthout loss 2.4 Non real-tme Traffc Transmsson Mechansm n of generalty, a sngle real-tme message stream per the Tmely-Token Token Protocol staton s assumed. The real-tme message stream of In FDDI and FDDI-M protocols, problems occurred staton s denoted by the trple (P, D, C ). Message because a staton cannot dstngush between unused real-tme bandwdth and unused non real- Ngeran Journal of Technology Vol. 32, No. 3, November

3 AND NON tme bandwdth. To overcome ths, an nteger U s added to the token, where U represents the sum of unused real-tme traffc bandwdth of all statons durng the prevous cycle [13]. When the token arrves n staton, U should also nclude the unused real-tme bandwdth of staton n the prevous cycle. In the Tmely-Token, when the token arrves at a node, the node can transmt non real-tme traffc for a tme not greater than the Token Holdng Tme, THT where THT s derved from the Tmely-Token algorthm [13] [18] as; THT =TTRT U TRT for TRT < TTRT U otherwse, THT =0, where TRT s the Token Rotaton Tme. TRT measures the tme between token arrvals at node. Drawbacks of Tmely-Token Protocol In the Tmely-Token, non-real-tme traffc makes use of only Category I avalable bandwdth. The Tmely- Token does not permt the non-real-tme traffc to use the spare bandwdth (.e. U) left over by the realtme traffc. As such, the throughput of the Tmely- Token decreases when U > Non real-tme Traffc Transmsson Mechansm n The BuST Protocol In the BuST, a node can delver non real-tme traffc each tme t gets the token, early or not, usng the spare bandwdth (.e. U) left by the real-tme traffc. If s s the tme unts consumed by node to delver real-tme traffc, then t can send non real-tme traffc for a tme not greater than h - s tme unts even f the token s not early. Drawbacks of BuST Protocol In BuST protocol, the non-real-tme traffc makes use of only Category II avalable bandwdth. As such, when the load level of the real-tme traffc s heavy, s = h, then, no bandwdth wll be left for the nonreal-tme traffc. In that case, non-real-tme traffc wll be starved. Besdes, Category II bandwdth s not allocated n such a way that the unused bandwdth n a node can be used by the non-real-tme traffc n another node. So, whle some nodes wth lght load of real-tme and non-real-tme traffc may have spare bandwdth left over, the other nodes wth heavy load of real-tme traffc wll stll starve ther non-realtme traffc as they cannot use the spare bandwdth from other nodes. 3. Methodology In ths paper, Optmal Guaranteed Servces Tmed Token (OGSTT) protocol was proposed for networks that support hard real-tme and non-realtme traffc. The analytcal modelng approach s adopted to develop and evaluate the proposed protocol. Mathematcal epressons are developed for varous performance parameters for the proposed protocol. Then, the performance of the protocol s evaluated wth MatLab software. Specfcally, wth the help of the MatLab software, the mathematcal epressons derved for each of the performance parameters s used to compute ts value under varous network and traffc confguratons. The parameters for the performance analyss nclude; Average Bandwdth Used by the Real-tme Traffc Per Cycle, Average Bandwdth Used by the Non Real-tme Traffc Per Cycle, and Average Cycle Length. These performance parameters are then used to compare the performance of the proposed protocol wth those of some estng tmed-token passng protocols. The comparson s based on the ablty of each of the protocol to support the non-real-tme traffc for any gven load level of the real-tme traffc. 4. Outlne of the OGSTT Protocol (a) Durng the rng ntalzaton phase, each node declares a TTRT. The mnmum declared value s selected as the rng's TTRT. Each node s then assgned a porton h of the TTRT to transmt ts real-tme traffc n every cycle. Durng each token rotaton, staton can transmt real-tme packets for at most h tme unts. (b) Each staton has a token-rotaton tmer, TRT for measurng the tme between token arrvals. (c) Each staton has a non-real-tme -lmt varable, A. In ths varable, staton stores the amount of tme t may transmt non real-tme messages. In addton, staton mantans a varable φ, where t stores the porton of h, the reserved real-tme bandwdth t used n transmttng real-tme traffc n the prevous token-rotaton. Also, another varable, b s defned, where staton stores the porton of h, the reserved real-tme bandwdth staton used n transmttng non real-tme traffc n the prevous token rotaton. Also, staton mantans a varable s, where t stores the total tme unts used out of h, n the prevous token-rotaton, where s = φ, + b. (d) To ntalze the token-rotaton tmers, no packets are transmtted durng the frst token rotaton. In addton, s s set to zero for all, and U = h = H. The nteger U s added to the token, where U represents the sum of unused real-tme bandwdth of all statons durng the prevous token-rotaton. When the token arrves at staton, U should also nclude the unused real-tme bandwdth of staton n the prevous token-rotaton. Ngeran Journal of Technology Vol. 32, No. 3, November

4 AND NON When staton receves the token, t performs the followng steps: 1. A := (TTRT - U -TRT ) + (1) 2. TRT := 0 (2) 3. U := U - (h - s ) (3) 4. If node has real-tme packets, t transmts them untl TRT counts up to h, or untl all the real-tme traffc s sent, whchever comes frst. 5. φ s assgned the number of tme unts of realtme transmsson used n step If TRT < h then f node has a real-tme packets, t transmts them untl TRT counts up to h, or untl all the non-real-tme traffc s sent, whchever comes frst. 7. b s assgned the number of tme unts of nonreal-tme transmsson used n step s s assgned the total number of tme unts of real-tme and non-real-tme transmssons used n step 4 and step U := U + (h - s ) 10. If staton has non real-tme packets, t transmts them for a tme perod of up to A tme unts, or untl all ts non-real-tme packets are transmtted, whchever occurs frst. 11. Staton passes the token to staton ( + 1) mod N. 4.1 Performance Bounds OGSTT Protocol In prncple, OGSTT operates lke a heavly loaded Tmely-Token protocol. The dfference les n how OGSTT and the Tmely-Token handle U, the drop n load of real-tme traffc. In OGSTT protocol, Category I (.e A*) avalable bandwdths are allocated to the non-real-tme traffc just lke n the Tmely-Token protocol. At the same tme, Category II (U) spare bandwdths left over by the real-tme traffc are allocated to the non-real-tme traffc just lke n the BuST protocol. Consequently, mamum throughput s mantaned by OGSTT even n the face of drop or varaton n the load level of the real-tme traffc. Techncally, the dfference between OGSTT and Tmely-Token s that n the Tmely-Token s = φ whereas n the OGSTT s = φ, + b. As such, analyss of the OGSTT protocol s smply the analyss of the heavly loaded Tmely-Token system where s s composed of φ, and b, the bandwdths used by the real-tme and non-real-tme traffc respectvely. Hence, n the analyss, the approach employed for the heavly loaded Tmely-Token n [13] s adopted. There s however one slght dfference n the assumpton made here. In [13], the system s assumed to be heavly loaded wth real-tme and non-real-tme traffc. In ths paper, the system s loaded wth lght load of real-tme traffc but wth heavy load of non-real-tme traffc. As such, n ths paper, n every token recept, the real-tme traffc may not use all the tme unts reserved for t n the node. However, the unused portons of the reserved real-tme tme unts are used by the non-real-tme traffc n every node. In ths way, the system stll behaves lke a heavly loaded system snce all the tme unts for data transmsson are used up n every node n every token recept. In order to reason about values that change over tme, the notatons used for the analyss are enhanced to nclude the followng terms presented below. 4.2 Defnton of terms T U,V : round m of staton,.e., tme nterval [t, W X ], where t s the tme when staton receves the token for the mth tme, and W X, s the tme when staton receves the token for the (m + 1)th tme. Y U,V Z :value assgned to A j durng T U,V. In partcular, Y U U,V s the value assgned to A when the token s receved at the begnnng of T U,V. [ Z U,V : duraton of non-real-tme transmsson of staton j durng T U,V. Note that [ Z U,V Y Z U,V [13]. hj :duraton of tme unts reserved for real-tme transmsson of staton j n every round. \ Z U,V : the porton of the hj tme unts actually used for real-tme transmsson n staton j durng T U,V. Note that [13]: \ Z U,V ] Z U,V h^ (4[) _ U,V Z : the porton of the hj tme unts actually used for non-real-tme transmsson n staton j durng T U,V. Note that: _ U,V Z ] U,V Z h^ (4_) ] U,V Z : the total of the portons of the h j tme unts actually used for real-tme and non-real-tme transmssons n staton j durng T U,V. Note that ] U,V Z h j [13]. Also, ] Z U,V = \ Z U,V + _ Z U,V h^ (5) `T`ZU,V : value of TRT j when staton j receves the token durng T U,V. In partcular, `T`UU,V s the value of TRT when the token s receved at the begnnng oft U,V [13]. `T`UU,V = ] Z + [ Z + c (6) 4.3 Theorem 1 (The Token s never late). For every staton, upon token arrval, TRT TTRT. The proof for Theorem 1 s gven n [13]. The same apples to OGSTT protocol. It was shown n [13] that Ngeran Journal of Technology Vol. 32, No. 3, November

5 AND NON for the heavly loaded Tmely-Token protocol, the followng epressons hold; U [ Z A U = h U U ] Z U [h ] Z U h Z (7) From the dscusson n ths paper, t can be seen that for the Tmely-Token protocol [13], j] Z U = j\ Z U (8) whereas, for the OGSTT protocol, U ] Z = U \ Z + U _ Z U h U (9) Then, `T`ZU,V = ] Z + [ Z + c (10) `T`Z U,V = U \ Z + U _ Z + [ Z + c (11) `T`ZU,V U h U + Y + c TTRT (12) So, the token s never late snce the Token Rotaton Tme, TRT does not eceed TTRT. 5. Comparson of OGSTT aganst a the Tmely-Token Token and BuST Protocols In ths secton, the BuST protocol s compared aganst the Tmely-Token and BuST Protocols. The comparson focuses on the ablty of these protocols to support real-tme and non-real-tme traffc. The comparson s based on the epresson for the upper bound on the average cycle length (Ĉ) for these protocols, because the epressons drectly reflect the ablty of the protocol to provde servces to the real-tme and non-real-tme traffc. 5.1 Epresson For The Upper Bound On The Average Cycle Length (Ĉ) FDDI Protocol In FDDI tmed-token protocol [6],[7],[19],[20], each node has two tmers, the token holdng tmer (THT ) and the token-rotaton-tmer (TRT ). The TRT counter always ncreases, whereas the THT only ncreases when the node s delverng non real-tme traffc. When TRT reaches TTRT, t s reset to 0 and the token s consdered as late by ncrementng the node's late count, Lc by one. The actual token cycle tme, denoted n ths paper as TRT # l s gven as; # TRT l = TTRT l + L nl (TTRT l ) (13) The token s consdered to arrve early at node f Lc = 0 otherwse the token s late (n ths case, Lc 1). When the token arrves at a node, the node can transmt non real-tme traffc for a tme no greater than THT where THT s gven as; THT =TTRT TRT # # l for TRT l < ``T` otherwse THT = 0; (14) where TRT # s the tme spent n the last round-trp of the token. Then, for the FDDI, A = ma(0, TTRT- TRT #). Joseph and Fouad has shown n [19] that for FDDI protocol, the upper bound on the average cycle length (Ĉ) for a heavly load system s gven as N Ĉ p q(τ T)+ T (15) N+1 Then, the upper bound on the average bandwdth allocated to the non-real-tme traffc (Ấ) s gven as Ấ = s t v(τ T) (16) tub Smlarly, Ozuomba and Chukwudebe showed n [20] that for FDDI protocol, Ĉ and Ấ for a system wth lght load of real-tme traffc but wth heavy load of non-real-tme traffc, are defned as follows; N Ĉ p N+1 q(τ T)+p N qu+ (H U) N+1 +W (17) Ấ = s t t v(τ T) + s vu (18) tub tub where U s the unused real-tme transmsson tme n the last round-trp of the token. The assumpton made n [20] s that U s constant for at least the N+ 1 consecutve cycle where the average s taken Tmely-Token Protocol The dfference between the FDDI and the Tmely- Token s n the use of TTRT n the FDDI and TTRT* n the Tmely-Token protocol, where TTRT* = TTRT - U. For the Tmely-Token, U= ( h s ) and s = ϕ then, τ can be replaced wth τ - U n the epressons for Ĉ n Eq 17 and Ấ n Eq 18 to obtan Ĉ T and Ấ T for the Tmely-Token, where U \ U = Uwab Uwy \ U and lm v= then, ub ub U { s N Ĉ p N+1 q(τ T)+ j\ U + W (19) U N Ấ =p q (τ T) (20) N BuST Protocol In the BuST protocol, Category I avalable bandwdth (.e.(τ-t) s not used by any traffc. The non-realtme traffc makes use of only the Category II,whch s the U spare bandwdth left over by the real-tme traffc. So, THT = 0 for and A = U. Now U= ( h s ) and s =ϕ, thus, Ĉ B and Ấ B for the BuST protocol are gven as follows; Ĉ } U + U \ U + W (21a) Ĉ } ~H+ U \ U + U \ U + W (21b) Ấ } = U = H U \ U (22) Ngeran Journal of Technology Vol. 32, No. 3, November

6 AND NON OGSTT Protocol For the OGSTT protocol, TTRT* = TTRT-U, then U = U (h U + U ) (23[) and U = \ U +_ U (23_). Snce a system wth heavy load of non-real-tme traffc s beng consdered, then, H= _ U U + U \ U (24) τ can be replaced wth τ -U n the epressons for Ĉ n Eq 17 and Ấ n Eq 18 to obtan Ĉ G and Ấ G as follows; Ĉ s t v T)+ _ tub U U + U \ U u (25) Ĉ s t v (τ T)+~H \ tub U U + U \ U u (26) Ấ s t v (τ T)+ _ tub U U (27[) Ấ s t v (τ T)+~H \ tub U U (27_) 5.2 Smulaton Results Consder a rng network wth four nodes (.e. N = 4) where τ = 100, W = 4 and h = 20 for all the nodes. It wll be assumed that the network s heavly loaded wth non-real-tme traffc but wth a varable load of the real-tme traffc. The real-tme traffc load, ϕ can vary from 0 to h. The values of Ĉ and Ấ for the varous load levels of the real-tme traffc are computed for the Tmely-Token, BuST and OGSTT protocols. The results are presented n Table 1 and Table 2. It s mportant to note that the unt for the performance parameters (ϕ, Â,and Ĉ ) dscussed n ths paper s tme unt, t can be second, mllsecond or mcrosecond dependng on the partcular network speed. Ths apples to the data n Table 1, Table Dscusson of results A System Wth Heavy Load Of Real-tme and Non real-tme Traffc When there s heavy load of real-tme traffc, that s U \ U = H = 80, U = H - U \ U =0 then (a) the BuST wll not allocate bandwdth to the non real-tme traffc, that s  = 0 (Table1) and Ĉ = 84 (Table2) (b) the Tmely-Token wll allocate a constant average bandwdth ( ub (Y )) to the non realtme traffc, where A* = 16, N = 4, so ub (Y ) =12.8. So  = 12.8 (Table1) and Ĉ = 96.8 (Table 2) (c) the OGSTT wll allocate an average bandwdth ((H U \ U ) + ub.y ) to the non real-tme traffc, where (H U \ U )=0, A* = 16, N = 4, so ub (Y ) =12.8). So,  = 12.8 (Table1) and Ĉ = 96.8 (Table2) So, n the case of a system wth heavy load of realtme and non-real-tme traffc, the Tmely-Token and the OGSTT have the same throughput whch s hgher than the BuST throughput. In partcular, BuST wll not allocate any bandwdth to the non-real-tme traffc, n ths case ( = 0). Table 1 Comparson of Average Bandwdth for Realtme Traffc Per Cycle,  for BuST, Tmely- Token and OGSTT. BuST Tmely-Token OGSTT ϕ ϕ    h = 20, τ=100, N = 4, W = 4 Table 2 Comparson of the computed values of Average Cycle length, Ĉ for BuST, Tmely-Token and OGSTT. BuST Tmely-Token OGSTT ϕ ϕ Ĉ Ĉ Ĉ h = 20, τ=100, N = 4, W = A System Wth No Real-tme Traffc But Wth Heavy Load Of Non real-tme Traffc When there s no real-tme traffc, that s ϕ = 0; H = 80, U = 80 Ngeran Journal of Technology Vol. 32, No. 3, November

7 AND NON (a) the BuST wll allocate all the (U) spare bandwdths to the non-real-tme traffc, that s  = 80 (Table1) and Ĉ = 84 (Table 2) (b) the Tmely-Token wll allocate the same constant average bandwdth ( ub (Y ) to the non-realtme traffc, where A* = 16, N = 4, so ub (Y ) =12.8. So  = 12.8 (Table1) and Ĉ = 16.8 (Table 2) (c) the OGSTT wll allocate an average bandwdth ((H U \ U ) + ub (Y )) to the non-real-tme traffc, where (H U \ U ) = 80, A* = 16, N = 4, so, ub (Y ) =12.8. Then,  = 92.8 (Table1) and Ĉ = 96.8 (Table 2) So, n the case of a system wth no real-tme traffc but wth heavy load of non-real-tme traffc, the Tmely-Token wll allocate the least amount of bandwdth to the non-real-tme traffc whle the OGSTT wll allocate the hghest. The BuST wll allocate all the spare bandwdth (U = H - U \ U = 80) left unused by the real-tme traffc to the non-realtme traffc ( = H = 80) A System wth Varable Load Level of Real-tme Traffc but wth Heavy Load Of non-real-tme Traffc Generally, f there s heavy load of non-real-tme traffc, then, as the load of the real-tme traffc ncreases from zero (no real-tme traffc) to the heavy load state, then, (a) the Tmely-Token allocates the same amount of average bandwdth ( ub (Y )) to the non-realtme traffc. The overall throughput of the system ncreases but t s less than the achevable mamum value of H + ub (Y ). (b) the BuST allocates all thespare bandwdths(u = H U \ U ) left over by real-tme traffc to the nonreal-tme traffc. The overall throughput of the system remans the same as H but t s less than the achevable mamum value of H + ub (Y ) (c) the OGSTT allocates ub (Y )plus the spare bandwdth (U = H - U \ U ) left over by real-tme traffc to the non-real-tme traffc. The overall throughput of the system remans the same as the achevable mamum value of H + ub (Y ). So, n the case of a system wth heavy load of non real-tme traffc but wth varable load of real-tme traffc, the OGSTT protocol mantans hgher throughput than the Tmely-Token and BuST protocols as long as U \ U < H. 6. Concluson and Recommendatons 6.1 Concluson Ths paper presented the Guaranteed Servces Token protocol (OGSTT) whch mproved the performance of estng tmed-token protocols, ncludng the Tmely-Token and BuST protocols. BuST and Tmely-Token protocols are tmed-token protocols recently ntroduced to mprove the communcaton servces provded by FDDI and FDDI-M networks. However, OGSTT mantaned hgher throughput than BuST and Tmely-Token protocols n the face of varatons n the load level of the real-tme traffc. At the same tme, OGSTT delvered guaranteed servces as requred by the hard and soft real-tme applcatons. Consequently, OGSTT s more sutable for mult-servces network snce t can effcently support dfferent traffc patterns and also provde not only bounded message transmsson tme as requred by the hard real-tme tasks, but also hgh throughput, as demanded by soft real-tme and non-real-tme tasks. 6.2 Recommendatons OGSTT can be ncorporated nto Ethernet and Profbus networks to mprove on the performance of those networks. The approach to be adopted and the mplementaton ssues are areas of further research. References 1. Regner, P. and Lma, G. Determnstc ntegraton of hard and soft real-tme communcaton over sharedethernet. In Proc. of Workshop of Tempo Real, Curtíba, Brazl, Ncholas Malcolm, We Zhao, "The tmed-token protocol for real-tme communcatons," Computer, vol. 27, no. 1, pp , January, IEEE, Token-passng bus access method and physcal layer specfcatons, Amerca natonal Standards ANSI/IEEE std Bao C. and We Z., Propertes of the Tmed Token Protocol, Department of Computer Scence Teas A&M Unversty College Staton, TX Oct., 1992 Techncal Report Shn K. G. and Zheng Q., FDDI-M: A scheme to double FDDI s ablty of supportng real-tme traffc, IEEE Trans. on Parallel and Dstrbuted Systems, Vol. 6, No. 11, pp ,Nov Kenneth C. Sevck and Marjory J. Johnson, "Cycle tme propertes of the FDDI token rng protocol," IEEE Trans. Software. Eng., vol. SE-13, pp , Mar Grow R., A tmed token protocol for local area networks, Proc. Electro 82, Token Access Protocols, Paper 17/3, May Chan E., Chen D., Cao J. and Lee C. H., The tme Propertes of the FDDI-M Medum Access Control Protocol The Computer Journal, Vol. 82, No. 1, pp , Jan Ngeran Journal of Technology Vol. 32, No. 3, November

8 AND NON 9. Survvable Adaptable Fbre-Optc Embedded Networks, MIL- STD-2004, US Dept. of Defense, Washngton D.C., Sept, Mcguffn L.J., Red L.O, and Sparks S.R., MAP/TOP n CIM Dstrbuted Computng, IEEE Network, vol.2 no. 3 May 1988, pp Uhlhorn R.W., The fbre-optc hgh-speed data bus for a new generaton of a mltary arcraft, IEEE LCS, Vol.2 No.1, pp 36 43, Feb Tovar E., Vasques F., Settng Target Rotaton Tme n Profbus Based Real-Tme Dstrbuted Applcatons, Proc. of the 15th IFAC Workshop on Dstrbuted Computer Control Systems, Jorge A. C., Maohua L., The Tmely-Token protocol, Computer Communcatons Vol.27 no.7, pp May, Franchno G., Buttazzo G. C., and Facchnett T., "Tme Propertes of the BuST Protocol under the NPA Budget Allocaton Scheme. In Proc. Of the Conference on Desgn, Automaton and Test n Europe, Munch, Germany, 10th -14th March Franchno G., Buttazzo G. C., and Facchnett T., BuST: Budget Sharng Protocol for Hard Real-Tme Communcaton. In Proc. of the 12th IEEE Internatonal Conference on Emergng Technologes and Factory Automaton (ETFA, Sept Agrawal G., Chen B., Davar W. S. Guaranteeng realtme message Deadlnes wth the Tmed Token Access Control Protocol, IEEE. Transacton on Computers, Vol 43. No.3, March 1994 pp Cobb J., Ln M., The On-Tme tmed-token protocol, Proc. IEEE GLOBECOM, Kalua C., Ozuomba S., Onoh G.N. Dynamc-Threshold- Lmted Tmed-Token (DTLTT) Protocol. Ngeran Journal of Technology, Vol. 32, No. 1, March, 2013, pp Joseph W. M. and Fouad A. T. Throughput Analyss of a Tmer-Controlled Token Passng Protocol under Heavy Load, IEEE Trans. Comm, Vol 37, No7 pp , July Ozuomba S. and Chukwudebe G.A, An Improved Algorthm For Channel Capacty Allocaton In Tmer Controlled Token Passng Protocols, Internatonal Journal Of Ngeran Computer Socety (NCS), Vol. 9 No 1, pp , June Ngeran Journal of Technology Vol. 32, No. 3, November