Copyright 10 IFAC Distributed Computer Control Systems,
`Seoul, Korea, 1997
`
`AN EXTENDED TCP /IP PROTOCOL OVER THE
`LOCAL AREA NETWORK FOR DCCS
`
`Jaehyun Park, Youngchan Yoon, and Sangchul Lee
`
`Department of Industrial Automation,
`Jnha University,
`Incheon 402-751, Korea
`email:jhyun@rcsl.inha.ac.kr
`
`Abstract: This paper proposes an extended TCP /IP protocol over local area
`networks, that can be used for soft real-time systems including distributed computer
`control systems and manufacturing automation systems. Because the proposed
`protocol extends the standard TCP /IP, all the existing application softwares can
`be used with the proposed protocol without modification. The proposed TCP /IP
`also provides the periodic transmission mode(PTM) , which provides a very efficient
`transmission method of the periodical data for real-time monitoring and control
`systems. With PTM, the periodical data collection and updating the control signals
`are possible with relatively small traffic overhead. This paper includes the computer
`simulation results, and the prototype system has been applied and evaluated to the
`large scale distributed control systems and human-like robot systems.
`
`Keywords: Ethernet, real-time communication
`
`1. INTRODUCTION
`
`Ethernet and TCP /IP pair is one of the most
`popular protocols for the MAC(Medium Access
`Control)(Metcalf and Boggs, 1976) and the mid(cid:173)
`dle range protocol layers in industry. Because
`TCP /IP is ported and used on hundreds of op(cid:173)
`erating systems and hardwares, there are lots of
`system-independent applications developed on the
`top of TCP /IP. Although TCP /IP on the top
`of Ethernet works well for the ordinary network
`applications such as remote login, file transfer, and
`remote procedure call, this is hardly used for the
`real-time applications such as manufacturing au(cid:173)
`tomation and distributed computer control appli(cid:173)
`cations. The major difference between the general
`internet applications and real-time applications
`is the time constraints in the message delivery.
`The usual internet applications permits relatively
`large amount of variances in the message delivery
`time and throughput. The real-time applications,
`however, have strict or very tight requirements in
`
`the message delivery time. Failure to meet these
`requirements may cause a severe trouble in the
`overall system. Although a sophisticated real-time
`network protocols can be used to meet these strict
`real-time constraints, general softwares can not
`be used on the top of these new real-time net(cid:173)
`work protocols without modification. This paper
`focuses an approach to use the popular TCP /IP -
`Ethernet protocols for the soft real-time systems
`with keeping compatibility.
`
`There are many researches to use Ethernet for
`the real-time applications. Because the packet
`collision of Ethernet is one of the major reasons
`why it can hardly used for real-time application,
`most of the researches concentrate on solving
`the packet collision problem (Tanenbaum, 1989).
`BRAM(Broadcast Recognizing Access Method)
`(Chlamtac et al., 1979) and MBRAM(Modified
`BRAM) (Signorile, 1988) are the protocols that
`resolve the collision problem in Ethernet network.
`Although, these protocols are collision-free proto-
`
`97
`
`IPR2022-00833
`CommScope, Inc. Exhibit 1022
`Page 1 of 8
`
`

`

`cols and provide the deterministic latency, there
`are several reasons to prevent these collision-free
`MAC protocols from being widely used.
`
`• No standard contention-free MAC protocol
`is fixed yet. The manufacturer of control de(cid:173)
`vices and softwares are not willing to support
`non-standard protocols. They may want to
`support general solution based on world wide
`standard such as CSMA/CD, token ring, to(cid:173)
`ken bus, TCP /IP, and etc. Backward com(cid:173)
`patibility must be supported so that the ex(cid:173)
`isting software should be reusable.
`• Reprogramming the 1/0 drivers is required
`for all the existing Ethernet devices. In
`case of the embedded MAC drivers, such as
`MC68EN360 CPU, it is impossible to modify
`MAC layer without redesign the chip.
`• All nodes connected to physical media should
`support the same scheduling algorithm. Even
`single node that use standard MAC proto(cid:173)
`col may cause unpredictable congestion prob(cid:173)
`lems. Even though a special conflict-free pro(cid:173)
`tocol such as PCSMA (Yavatkar, 1994) is
`developed, most of the MAC protocols can
`not coexist with each other.
`
`These reasons make it difficult to use modified
`MAC protocol for the real-time applications.
`
`Another approach to use TCP /IP - Ethernet
`protocols for the real-time applications is to im(cid:173)
`prove the performance of TCP /IP layer. TCP /IP
`was originally designed to be used in the inter(cid:173)
`networking environment where long latency and
`relatively unreliable data transmission are ex(cid:173)
`pected. To cope with this unreliable connection
`due to inter-networking, multiple steps of error
`controls are provided. To maintain optimal uti(cid:173)
`lization of the low bandwidth channel, standard
`TCP /IP uses the best-effort paradigm for packet
`scheduling, buffer management, feedback, and end
`adjustment(Lefelhocz et al., 1996). Recently, a se(cid:173)
`ries of algorithms and implementation techniques
`improving the performance of the transmission
`control protocol(TCP) have been proposed: TCP
`for the transactions(ISI, 1994; ISI, 1992), receiving
`overhead reduction technique(Clark et al. , 1989),
`and header prediction algorithm(Jacobson, 1990).
`While these researches are useful for the conven(cid:173)
`tional internet applications, they are not efficient
`for the real-time applications over Ethernet-only
`environment that most of the real-time systems
`are used in.
`
`This paper proposes a modified transmission con(cid:173)
`trol protocol(TCP) to reduce communication la(cid:173)
`tency over Ethernet link without modification
`of MAC layer. For this, this paper introduces
`another type of service called least-effort. Even
`though least-effort is not able to be applicable
`to the low bandwidth link, internet, it shows a
`
`relatively high performance for a special purpose
`network such as manufacturing networks based
`on Ethernet that has a higher bandwidth than
`ordinary internet. This paper also proposes a
`special service called PTM(Periodic Transmission
`Mode). This mode increases utilization and results
`in lower transmission delay.
`
`Next section describes the key features of the
`proposed extended TCP protocol. Implementa(cid:173)
`tion issues and evaluation results follows in the
`subsequent sections.
`
`2. KEY FEATURES OF LAN-TCP
`
`This paper extends the standard TCP protocol for
`the Ethernet-only environment, and the proposed
`protocol is called LAN-TCP(TCP for local area
`network only). The LAN-TCP has three major
`features : least-effort strategy, structure oriented
`protocol, and periodic transmission mode. In this
`section, these features are described in detail.
`
`2.1 Least-effort strategy
`
`The standard TCP follows the best-effort strategy
`to transmit a packet over a physical network link
`that is usually slow. TCP sliding window is a
`good example of explai.ning best-effort delivery
`(Comer, 1995; Stevens, 1994). It makes stream
`transmission efficient in wide area network, and it
`is an important feature of TCP for internetwork
`between high speed network and lower one. The
`best-effort strategy, however, increases the pos(cid:173)
`sibility of packet collision when used with high
`speed link like Ethernet. This packet collision
`is one of the major reasons of unexpected la(cid:173)
`tency of packet delivery over Ethernet. To re(cid:173)
`duce this packet collision without modification
`of MAC layer, the least-effort strategy, contrary
`to the best-effort strategy, is proposed in this
`paper. The main idea of the least-effort proto(cid:173)
`col is that a certain limitation of the network
`bandwidth is reserved to single node, and the
`operating system is responsible for managing this
`limitation. The bandwidth limitation reserved to
`each node depends on the system architecture,
`the operating systems, and the transfer rate of
`media. The minimum bandwidth can be set to
`default or required amount requested by applica(cid:173)
`tions when applications open or bind SAP(Service
`Access Point) of transport layer or socket layer
`in BSD socket compatible environment. To max(cid:173)
`imize the effect of the least-effort strategy, LAN(cid:173)
`TCP does not use the sliding window algorithm
`that is derived from the best-effort paradigm and
`developed for the inter-networking purpose. In
`the LAN-TCP, to disable the sliding window, the
`window size is limited to MSS(Maximum Segment
`
`98
`
`IPR2022-00833
`CommScope, Inc. Exhibit 1022
`Page 2 of 8
`
`

`

`Size:(Comer, 1995; Stevens, 1994)). If one of two
`machine has a LAN-TCP facility, bandwidth lim(cid:173)
`itation can be achieved by pseudo window size
`which is described in the next section.
`
`Least-effort strategy is applied only to TCP or
`socket layer in LAN-TCP. There are four reasons
`for this restriction.
`
`• Port de-multiplexing is applied only above
`the IP layer and the output data of IP
`layer are transmitted or queued to IP buffer
`without further de-multiplexing. Generally
`IP fragmentation is not required in local area
`network because TCP process already knows
`the size of MTU(Maximum Transfer Unit). If
`the bandwidth is not managed by TCP pro(cid:173)
`cess, de-multiplexed TCP segments stored at
`IP buffer must be re-arranged to distribute
`transmission rate of each application pro(cid:173)
`cesses. This rearrangement will be another
`overhead of protocol layering and make the
`buffer management complex.
`• Application process should be pended until
`sending system call is terminated if multiple
`segments are sent, and TCP process passes
`segments to the IP process at intervals for
`pending duration.
`• LAN-TCP should be compatible and coex(cid:173)
`istable with TCP. If least-effort delivery is
`not required, the standard TCP connection
`should be explicitly used. IP can not nego(cid:173)
`tiate connection type with remote machine
`due to its connection-less feature. Compati(cid:173)
`bility is an important feature of LAN-TCP.
`If LAN-TCP can not support conventional
`TCP protocol, the problem of difficulty of
`implementation remains same as collision(cid:173)
`free MAC protocol.
`• IP or TCP process must know which ap(cid:173)
`plication process is sending packet and how
`much bandwidth is reserved. IP process can
`be modified to deal with these informa(cid:173)
`tion, but TCP process already has these fa(cid:173)
`cility known as TCB(Transmission Control
`Block) (Comer and Stevens, 1994; Wright
`and Stevens, 1995). TCB can be easily mod(cid:173)
`ified to implement least-effort mechanism.
`
`tured datagram. The destination machine passes
`to the receiver exactly the same sequence and size
`of blocks that the sender passes to it on the source
`machine. This feature makes buffer management
`of LAN-TCP easy and structural.
`
`2.3 Periodic transmission mode
`
`The periodic data transmission is one of the fun(cid:173)
`damental communication method for the real(cid:173)
`time systems including distributed computer con(cid:173)
`trol systems. When a packet is transmitted peri(cid:173)
`odically, the receiver is able to predict next arrival
`time in local area network. LAN-TCP supports a
`special transmission mode for the periodic trans(cid:173)
`mission called PTM(periodic transmission mode) .
`To identify the type of segment data, TCP header
`contains 6--bit fields: URG, ACK, PSH, RST,
`SYN, and FIN. Each bit is used to initiate transi(cid:173)
`tion of TCP state machine( Comer, 1995; Stevens,
`1994). Among them, ACK bit is important for the
`PTM mode. To the TCP acknowledgment scheme,
`ACK is sent by destination machine after segment
`from source machine was received. Although des(cid:173)
`tination machine can generate only last ACK for
`acknowledgment of multiple segments in sliding
`window algorithm, the receiver always generate
`ACK when segment is transmitted periodically
`because each segments are transmitted every the
`fixed interval. Generation of series of ACK can
`be avoided in PTM mode. ACK packets are not
`sent after PTM connection is established. ACK
`packet is used only for negative acknowledgment
`when destination machine didn't receive segment
`until next arrival time. From the queuing analysis
`of CSMA/CD, PTM can reduce the media traffic
`up to half or less(Rom and Sidi, 1990; Tobagi and
`Hunt, 1980).
`
`3. IMPLEMENTATION
`
`This section describes major issues to implement
`the LAN-TCP. The various techniques for opti(cid:173)
`mizing TCP is not implemented in the prototype
`LAN-TCP yet.
`
`2.2 Structure oriented protocol
`
`3.1 Connection management
`
`TCP is a stream oriented protocol, and TCP's
`stream is unstructured. This unstructured stream
`makes programming inefficient. On the contrary,
`UDP uses structured datagram delivery, but UDP
`doesn't have transmission control facility. Appli(cid:173)
`cation process using UDP is responsible for con(cid:173)
`trolling transmission. LAN-TCP is datagram and
`reliable stream oriented protocol. In the LAN(cid:173)
`TCP, stream is not divided into bytes but struc-
`
`Only a slight modification of standard TCP is re(cid:173)
`quired for implementing the LAN-TCP. The con(cid:173)
`nection establishing and closing scheme of LAN(cid:173)
`TCP is almost same as that of the standard TCP
`except the PTM mode. Generally, TCP options
`follow only SYN segment while the connection
`is established. Standard TCP header contains 6
`reserved bits for future use. Two unused bits
`are used for negotiating connection of LAN-TCP.
`
`99
`
`IPR2022-00833
`CommScope, Inc. Exhibit 1022
`Page 3 of 8
`
`

`

`.......... """ ......
`;···-, n.-.......,__.....,,
`......... ,-:1mmm1m1:1
`.._
`c~:~~~iif ~ ~
`·-----
`I
`
`U l6
`
`,~-'°"'...to
`J2•...-aa11icr
`
`I
`
`16--MI TCPcti..b.a
`
`.,
`.
`
`~Gl-,1
`
`- -~ •)')
`
`'recv' system call
`
`)I
`
`',
`
`16-.flit ........
`
`16-N lfP flCIHla
`
`Awaken by kernel
`
`Suspended
`
`Delay/SI
`
`complete
`
`Delay/Sleep
`
`LAN-TCP
`
`SendACK
`
`Reset PTM timer
`
`'recv' processing
`
`Tenninate system call
`
`Fig. 2. LAN-TCP recv processing
`
`set to the system clock. With this resolution, the
`maximum bandwidth depends on system clock.
`For example, if system clock is 60Hz and media
`MTU is 1500, then burst transfer rate of single
`process can be up to 90 kbytes.
`
`3.3 Periodic transmission mode
`
`The PTM connection is not bi-directional. When
`LAN-TCP connection is established, only sender
`set ' PTM' bit in SYN packet, thus uni-directional
`PTM connection is established. Transmission in(cid:173)
`terval is automatically detected by receiver side,
`and it decreases from initial value to mean value
`gradually.
`
`In PTM mode, PTM timer notifies that next
`packet is not arrived yet, and ACK, which means
`negative acknowledgment in PTM mode, must
`be sent(see Fig. 2). If sender side receives ACK,
`sequence number of ACK is checked, and corre(cid:173)
`sponding segment must be sent again. Standard
`TCP process already has a facility, retransmission
`buffer, enough to solve this problem. Mostly ring(cid:173)
`style buffer is used with retransmission buffer. In
`typical TCP process, acknowledged segment in
`retransmission buffer is removed, but it must be
`left in PTM mode. Segment in ring buffer is not
`removed but destroyed by head pointer of ring
`buffer step by step. If receiver side can not process
`
`Fig. l. LAN-TCP header
`
`One is for LAN-TCP connection and the other is
`for PTM connection. The position of two bits is
`shown in Fig. 1, enclosed by dashed box. 'LAN' is
`used for LAN-TCP connection and 'PTM' is used
`for PTM connection. According to IAB internet
`standard, these bits must be zero(ISI, 1981).
`
`There are some congestion control mechanism for
`best-effort delivery such as slow start algorithm,
`round-trip time measurement, Nagle algorithm.
`These do not need to be implemented in LAN(cid:173)
`TCP because LAN-TCP is only for local area
`network especially inside subnet.
`
`3.2 Additional timers for LAN- TCP
`
`TCP process requires various timers to control
`transmission: retransmission timer, persist timer,
`and keep alive timer(Wright and Stevens, 1995).
`Some of them are disabled while LAN-TCP con(cid:173)
`nection is established. Besides these standard
`timers, two additional timers are added for LAN(cid:173)
`TCP: in-band timer and PTM timer. In-band
`timer is used by sender side for bandwidth lim(cid:173)
`itation, and PTM timer is used by receiver side
`to send ACK for negative acknowledgment while
`PTM connection is established. Before a TCP
`segment is sent, TCP process must check whether
`in-band timer is expired or not. If in-band timer
`is already expired, segment is sent immediately,
`otherwise, TCP process is delayed until the timer
`is expired. After a segment was sent, in-band
`timer is initiated. The operation of two timers are
`depicted in Fig. 2 and Fig. 3. If PTM connection
`was established, all of TCP timer must be disabled
`in PTM connection requester except keep alive
`timer, in-band timer, and PTM timer. In most
`operating systems, the resolution of TCP timer
`is around several tens of mili-second unit. This
`resolution is sufficient to manage control mecha(cid:173)
`nism of the standard TCP, but higher resolution
`is required for LAN-TCP. It is somewhat easy
`to obtain higher resolution in RTOS(Real Time
`Operating System). POSIX standard supports a
`function 'nanosleep' which has higher resolution,
`but real resolution is bounded up system clock and
`depends on operating system and processor. To fix
`this variation, the resolution of LAN-TCP timer is
`
`100
`
`IPR2022-00833
`CommScope, Inc. Exhibit 1022
`Page 4 of 8
`
`

`

`ID0000
`
`700000
`
`f=
`
`~ SOXIOO
`fi
`i'"""°°
`
`0
`
`(b) PTM mode
`
`300<00
`
`2CD000
`
`//! /
`
`/
`!
`:
`
`:_./'
`:
`...
`
`---_ ... -.::.--.-:~--~::_J_·:~L32~--------'
`
`-(cid:173)~ -(cid:173)"'""-
`
`. _.. .... --
`
`ACI(-
`
`(a) Nonnal mode
`
`·--
`
`1100,-,-----,-,--,--,-----,-,- ~ -- ~ - - -
`100 •• ··
`3:1:) . . ..
`
`,eoo ~ ,- - - - - - - -·- - - - ------ -J:::;: .
`io...-•··
`
`I :- ,,/,.iJ3:=:i=,=~,~
`: .__.·:~:~:_·"'_ .. ----------
`
`.,•
`
`...... -·······
`
`200 ....._,,:...___,__...._ ___ .____.__....__.....___._ _
`
`_.____J
`
`Fig. 4. Evaluated Performance of CSMA/CD with
`20 nodes
`,..,,.,--,----.--~-~~~--r- ~ - - , - -
`100 ·• ·
`!00 . • ..
`too •O ••
`1100 - 11· •
`
`~ -
`
`I
`
`900000
`
`Fig. 3. finite state machine used for LAN-TCP
`send processing
`periodic packets in time by unexpected distur(cid:173)
`bance, then series of segments can be dropped.
`If dropped packets are out of ring buffer, these
`packets can not be retransmitted. In this case, the
`receiver must send source quench ICMP message,
`or the connection will be aborted. ICMP is only
`for internet protocol, but it is the only way to
`notify alert status.
`
`3.4 Backward compatibility
`
`Because the LAN-TCP is compatible with the
`standard TCP, the connection between the LAN(cid:173)
`TCP nodes and the ordinary TCP nodes can be
`established, and two protocols work well without
`conflict at all. TCP has a facility known as window
`size advertisement to manage sliding windows.
`Window size in TCP header(see Fig. 1) is the
`size of free TCP buffer in remote machine's TCP
`process. Local machine sends full size of window
`which can be received and stored in remote ma(cid:173)
`chine. To prevent remote machine from sending
`more than one segment of window, pseudo window
`size is used in LAN-TCP. If one of two machine
`adopts LAN-TCP, bandwidth limitation can be
`achieved easily by pseudo window size. LAN-TCP
`uses pseudo window size rather than actual win(cid:173)
`dow size when the remote machine doesn't sup(cid:173)
`port bandwidth limitation. The maximum pseudo
`
`101
`
`u,,aoo
`___ _ __.__..,.:;.__,,L.-_.....___._ _ _._ _ _.____J
`200
`G
`l00
`1000
`1200
`1400
`1500
`1SX>
`&00
`20'JO
`~ -]
`
`Fig. 5. Delay on LLC with 20 nodes
`
`window size will be limited to MSS(Ma.ximum
`Segment Size).
`
`If applications use PTM, they need to use spe(cid:173)
`cial socket system call rather than standard
`socket system call. These applications pass pa(cid:173)
`rameters different from normal LAN-TCP con(cid:173)
`nection when create the end point of connection.
`For example, 'socket' system call can be called
`with new parameter 'SOCK.PSTREAM' rather
`than 'SOCK..STREAM', if PTM connection is
`required. If remote machine doesn't support LAN(cid:173)
`TCP, PTM connection can not be established,
`thus the connection will be aborted. In this case,
`normal LAN-TCP connection is established and
`works well without conflict with conventional
`TCP.
`
`4. PERFORMANCE EVALUATION
`
`To evaluate the performance of the proposed pro(cid:173)
`tocol, a computer simulation package is devel(cid:173)
`oped. In the CSMA/CD protocol, if two stations
`detect an idle state and begin transmission at
`the same time, a collision will occur. Any station
`detecting a collision stops its transmission and
`generates jam signal, and then waits a random
`
`IPR2022-00833
`CommScope, Inc. Exhibit 1022
`Page 5 of 8
`
`

`

`period of time before retransmission. To avoid
`overloading the channel, the range of retrans(cid:173)
`mission interval increases using a binary expo(cid:173)
`nential back off algorithm. The complete anal(cid:173)
`ysis of binary exponential back off algorithm is
`very complex. Even though the simplified models
`were proposed(Tobagi and Hunt, 1980; Boggs et
`al., 1988), the precise model with complete binary
`exponential back off algorithm should be used to
`evaluate the performance of the proposed proto(cid:173)
`col. Additionally, to evaluate the effect of physical
`environment, simulation model should contain a
`facility of handling propagation delay.
`
`The first simulation is evaluating the basic uti(cid:173)
`lization under heavy load to validate the com(cid:173)
`puter simulation package itself. Utilization of the
`CSMA/CD shown in Fig. 4 is almost same as the
`measured utilization by Shoch and Hupp(Shoch
`and Hupp, 1980). Evaluated utilization is up
`to 95% with 20 nodes and ll00bytes of packet
`length. Average transmission delay from network
`layer to the destination node is shown in Fig. 5.
`
`4.1 Simulation model
`
`The network configuration simulated consists of
`four components(two periodic servers, one burst
`server, and one burst client) , and two groups of
`periodic sources. Periodic server receives packets
`from the periodic data sources and sends acknowl(cid:173)
`edgment packet to them in standard TCP model,
`but in the LAN-TCP model, acknowledgment
`packet is not sent.
`
`Burst server and burst client exchange series of
`segments and acknowledgment packets as sliding
`window scheme. Burst client in this simulator
`does not generate packets by probability function
`but follows actual transmission scheme which ex(cid:173)
`changes one acknowledgment packet per several
`segments. Burst client does its best to transmit
`packets as soon as possible in conventional TCP
`model, but in the LAN-TCP model, transmission
`interval is never exceeded a preset limitation re(cid:173)
`served at each node.
`
`4.2 Simulation results
`
`Fig. 6 through Fig. 8 show a delay, average load,
`and number of collisions using the proposed pro(cid:173)
`tocol and standard TCP for the same network
`configuration. Fig. 6 shows the transmission delay
`observed at the network layer under the various
`load condition generated by periodic sources with
`10 % of variance of transmission interval. The effi(cid:173)
`ciency of PTM connection can be measured in this
`figure. There is significantly higher delay under
`conventional TCP than proposed protocol when
`
`, I.AH TCP ·• ·
`10 ,-bk ~
`:, pencxlc ~ · LN' TcP ·• ·
`10 p,4lltodc ~ • 1'CP •
`2'~IOUIIW-TCP •
`
`,00 .
`
`...
`
`300
`
`I
`
`I
`a
`·• •••
`
`t
`
`20
`
`25
`
`, __ .. _ - - - i - i "'
`
`30
`
`35
`
`Fig. 6. Delay on LLC under loads generated by
`periodic sources (packet length is 258bytes)
`
`, . .,,.
`
`·•.
`
`~
`······~ .. ~=~:~:········- --. --~: .. · .. ::-.......... =.~---···· .. -;-....... .
`____ ···_··-_· ........... :·····--... _ ....... ....................... ~..., .
`.. .
`
`100000 -~-
`
`~ .
`~ 10C,,0
`~
`&
`Ct 1000
`
`Fig. 7. Transmission delay caused by transient
`burst transmission
`burst transmission is performed(Fig. 7). In a TCP
`burst transmission, average load can not exceed by
`unity because the transmission is based on packet
`exchange(Fig. 9). Even though this type of traffic
`is not continued long, it can cause critical problem
`in industrial network, especially very long delay,
`even transmission failure on MAC layer. From
`the simulation result, many transmission failures
`caused by 16 contiguous collisions are occurred
`as the number of collisions is increased (Fig. 8).
`The reason of huge delay is best-effort transmis(cid:173)
`sion which makes the average load around unity
`abruptly. Fig. 9 shows the effect of best-effort
`transmission under low load. For example, offered
`load is 348kbytes/sec with 25 mili-second of trans(cid:173)
`fer interval by 24 TCP periodic sources. The load
`is increased to 1044kbytes/sec after burst trans(cid:173)
`mission is started. This load results in significant
`transmission delay(Fig 7) and collisions(Fig 8). By
`the simulation result, 20~220 packets are dropped
`by transmission failure.
`
`Of course, this number of collisions is not practi(cid:173)
`cal because burst transmission is never continued
`long during tens second. The number of colli(cid:173)
`sions and transmission delay are varied by the
`start time of periodic transmission. Consider the
`
`102
`
`IPR2022-00833
`CommScope, Inc. Exhibit 1022
`Page 6 of 8
`
`

`

`uoo
`
`'"'°
`
`...,
`
`,00
`
`~
`
`~
`
`~
`
`~
`
`' ......
`.... ······::·o::.::::•-.•.~:··:-.:::•::: ·:.:·:·:·
`............. -······-
`o '---,__-== =-=--=-~L.;;..=a::..:::=a::.=~
`u
`u
`~
`~
`Tf'WIIII.-WWVIIOl,-klClc:~,..J
`
`..
`
`teristics, the proposed TCP /Il' includes periodic
`transmission mode(PTM), which provides a very
`efficient transmission method of the periodical
`data for real-time monitoring and control systems.
`With PTM, periodical data collection and up(cid:173)
`dating the control signal are possible with rela(cid:173)
`tively small traffic overhead. Computer simulation
`shows the proposed TCP /Il' has relative higher
`performance compare to the ordinary TCP /Il'
`protocol.
`
`The prototype system has been applied and evalu(cid:173)
`ated to the large scale distributed control systems
`and human-like robot systems. Optimization of
`the proposed protocol is required to be used for a
`large scale distributed system and fine calibration
`may be necessary to maximize real-time perfor(cid:173)
`mance to the base operating system.
`
`6. REFERENCES
`
`Boggs, D.R., J .C. Mogul and C.A. Kent (1988).
`Measured capacity of an ethernet: Myths and
`reality. Proceeding of ACM SIGCOMM '88
`pp. 222-234 .
`Chiamtac, Imrich, William R. Franta and K. Dan
`Levin (1979). BRAM: The broadcast recog(cid:173)
`nizing access method. IEEE Transactions on
`Communications 27(8), 1183-1189 .
`Clark, David D., Van Jacobson, John Romkey and
`Howard Salwen (1989). An analysis of tcp
`processing overhead. IEEE Communication
`Magazine pp. 23- 29.
`Comer, Douglase (1995). Internetworking with
`TCP /JP Volume I : Principles,Protocols, and
`Architecture. Prentice Hall, Inc. Englewood
`Cliffs, New Jersey.
`Comer, Douglase and David L. Stevens (1994). In(cid:173)
`ternetworking with TCP /IP Volume II : De(cid:173)
`sign,Implementation, and Internals. Prentice
`Hall, Inc. Englewood Cliffs, New Jersey.
`(1981). Transmission control protocol.
`RFC793.
`ISI (1992). Extending TCP for transactions -
`concepts. RFCJ 379.
`ISI (1994). T /TCP - TCP extensions for transac(cid:173)
`tions functional specification. RFC1644-
`Jacobson, V. (1990). 4BSD TCP header predic(cid:173)
`tion. Computer Communication Review.
`Lefeihocz, Christopher, Bryan Lyles, Scott
`Shenker and Lucia Zhang (1996). Congestion
`control for best-effort service: Why we need a
`new paradigm. IEEE Network pp. 10-19.
`Yletcalf, R.M. and D.R. Boggs (1976). Ethernet:
`Distributed packet switching for local comp(cid:173)
`tuter networks. Communication of the A CM
`19(7), 395-404.
`Rom, Raphael and Moshe Sidi (1990). Multiple
`Access Protocols. Springer-Verlag. New York.
`
`ISI
`
`Fig. 8. Number of collisions under burst transmis(cid:173)
`sion
`
`1,00
`
`i s
`i 1()00
`g ...
`...
`5 ...,
`
`i
`1
`R
`1
`
`,00
`
`0 •
`
`•. 111,
`
`,::::~:::~::"'''':!::•:-· -·--• --
`.... ··:·:_•.:.---=~.:::::·~:~: : ...... _ ... :·::~:-:::
`
`Fig. 9. A variation of channel utilization
`
`worst case, if all nodes start periodic transmission
`and the transmission interval is not varied, then
`extreme transmission delay is occurred without
`burst transmission. On the other hand, if the
`start time is well distributed, then no collision is
`detected at all. So, periodic transmission which is
`based on system clock has potential problem.
`
`If the simulation results(Fig 6, Fig 7, and Fig 8)
`are acceptable in real world, 20 periodic sources
`generating 200bytes of control data at each 10
`mili-second and some LAN-TCP hosts can be
`feasible configuration.
`
`5. CONCLUSION
`
`This paper describes an effort to use industry
`standard TCP /IP - Ethernet for soft real-time
`systems including distributed computer control
`systems and manufacturing automation systems.
`To maintain the backward compatibility, the stan(cid:173)
`dard TCP header is extended to reduce traffic
`overhead and include periodic transmission mode.
`Because the proposed protocol extends the stan(cid:173)
`dard TCP /Il', all the existing application soft(cid:173)
`wares the proposed TCP /Il' can be used with(cid:173)
`out modification. Besides soft real-time charac-
`
`103
`
`IPR2022-00833
`CommScope, Inc. Exhibit 1022
`Page 7 of 8
`
`

`

`Shoch, John F . and Jon A. Hupp (1980). Measured
`Performance of an Ethernet Local Network.
`Xerox Corporation. California.
`Signorile, Robert P. (1988) . MBRAM-A priority
`protocol for PC based local area networks.
`IEEE Network 2(4), 55-59.
`Stevens, W. Richard (1994). TCP/IP fllustrated,
`Volume 1. Addison-Wesley Publishing Com(cid:173)
`pany. Massachusetts.
`Tanenbaum, Andrew S. (1989). Computer Net(cid:173)
`works. Prentice Hall, Inc .. Englewood Cliffs,
`New Jersey.
`Tobagi, F.A. and V.B . Hunt (1980). Performance
`analysis of carrier sense multiple access with
`collision detection. Comput. Networks 4, 245-
`259.
`Wright, Gary R. and W.Richard Stevens (1995).
`TCP /IP fllustrated, Volume 2 : The Imple(cid:173)
`mentation. Addison-Wesley Publishing Com(cid:173)
`pany. Massachusetts.
`Yavatkar, Rajendra (1994). A reservation-based
`CSMA protocol for intergrated manufac(cid:173)
`IEEE TI-ans. on Sys(cid:173)
`turing networks.
`tems,MAN, and Cybernetics 24(8) , 1247-
`1258.
`
`104
`
`IPR2022-00833
`CommScope, Inc. Exhibit 1022
`Page 8 of 8
`
`