Comparison of Type 2 and Type 4 Services for MIL-STD-188-220B Dr. David J. Thuente 1 Timothy E. Borchelt Department of Computer Science Indiana University Purdue University Fort Wayne Fort Wayne, Indiana 46815 email: thuente@ipfw.edu Raytheon Systems Company Fort Wayne, Indiana, 46808 email: teborc@ftw.rsc.raytheon.com Abstract - Numerous papers at recent MILCOM conferences have proposed and examined different media access control (MAC) algorithms. Many of these have been for combat net radios or for MIL-STD-188-220B. The two most visible and perhaps viable MAC algorithms from MIL-STD-188-220B are deterministic adaptable priority network access delay (DAP-NAD) and radio embedded network access delay (RE-NAD). Analyses of these algorithms and their variations have been published. These comparative analyses of parts of MILSTD-188-220B have not addressed an equally fundamental aspect of the datalink layer protocol: the type of service supported. MIL-STD-188-220B has defined four types of service: Types 1, 2, 3, and 4. Type 2 and 4 are generally used to reliably transfer large amounts of data because they allow concatenation and acknowledgments. Type 2 services are connection oriented with decoupled acknowledgments based on HDLC protocols. Type 4 services are connectionless with individual decoupled acknowledgments for every packet received. Type 1 and 3 are connectionless operations with Type 1 being unacknowledged and Type 3 requiring an immediate acknowledgment (also called coupled acknowledged Type 1 in MIL-STD-188-220B). There has been no published analysis of Type 2 and 4 services and yet this is a fundamental choice for most reliable transmissions on networks that will use MIL-STD-188220B. (FS) system have been used to drive the model. Most of the comparative studies for Type 2 and Type 4 were done using DAP-NAD. We show, for our example networks, that if they are lightly or moderately loaded in an errorfree environment, then the performance of the Type 2 and Type 4 services for average message latency and for the completion of fire missions are relatively close to each other. In general, Type 2 slightly outperforms Type 4 under these conditions. When the load on the network becomes heavier and when errors are introduced, Type 2 significantly outperforms Type 4 in all measures of network performance. The results also show that the maximum window size of 20 for Type 4 services can be a significant constraint on the performance of the Type 4 protocol. Comparative results for Type 2 and Type 4 using RE-NAD showed that Type 2 outperformed Type 4 much more completely for RE-NAD than for DAP-NAD. The results in this paper show that DAP-NAD is significantly better than RE-NAD with both Type 2 and Type 4 services. All of these results have direct and immediate applications to networks that are to use MILSTD-188-220B. As an outgrowth of the work begun on the Advanced Field Artillery Tactical Data System (AFATDS) program, a detailed model for the analysis and evaluation of many aspects of MIL-STD-188-220B has been built. This model has been used to develop many important results about the performance of MIL-STD-188-220B for various network configurations and message types and loads. The primary emphasis in this paper is on the comparative results for Type 2 and Type 4 services. The message behavior (threads, sizes, frequency, dependencies, sequences, etc.) of an actual fire support 1 This work was performed, in part, while the author was a consultant at Raytheon Systems Company C3I Segment, Tactical Systems division and where he continues as a consultant on communications protocols. I. INTRODUCTION Efficient utilization of the available bandwidth is critical to the use of combat net radios. Two of the key issues for this efficient utilization are the MAC algorithms and the types of service provided. Some results presented at previous MILCOM conferences have been on the performance, improvements, or variations of RE-NAD and on extensions of the RE-NAD protocol to better transmit compressed voice. See for example, [1], [6], [10], [11], [13], [14] and references within those papers. All of the extensions and improvements still rely on the type of service for the reliable delivery of packets. The results presented here are the first time, to our knowledge, any study has shown the conditions for and the magnitude of the benefits of Type 2 services. There have not been any published studies comparing the efficiency of the types of service for MIL-STD-188-220B. The authors have presented results at MILCOM conferences comparing RE-NAD and DAP-NAD for Types 1 and 2 and have discussed an OPNET model for other MILSTD-188-220B features. Earlier results presented at the 1 0-7803-5538-5/99/$10.00 (c) 1999 IEEE Tactical Communications Conference [9] developed some improvements and optimizations for DAP-NAD as part of the AFATDS modeling work. Extensions to the original AFATDS model were accomplished in conjunction with its application to two other projects as well as some Raytheon sponsored IR&D. The work presented in this paper is an extension of that original AFATDS OPNET work but is not part of the AFATDS development. The original model of the AFATDS communication system was verified and validated and the description of that effort was described in [8], [9]. The primary emphasis in this paper is on the comparative results for Type 2 and Type 4 services for various network configurations and MAC algorithms. II. NETWORK AND SCENARIO CONFIGURATIONS Two different sized networks are used for the simulation scenarios. Similar network configurations are expected to be used in actual FS situations. The smaller network, called Net_A has seven nodes: six fire units, FU_As, and an operations center, OC_A. The OC_A generates all of the fire missions. There are two types of fire missions generated: thread 1 has a Subsequent Adjust and consists of a sequence of six messages with corresponding ACKs plus two additional messages. Thread 2 does not have a Subsequent Adjust message and consists of a sequence of four messages with corresponding ACKs. The average message size is 75 bytes with a range of 20 to 400 bytes. The Type 2 acknowledgments (when needed) and the Type 4 acknowledgments are eight bytes each. The second network, called Net_AB, has 14 nodes: two operations centers, OC_A and OC_B, with each operations centers controlling its corresponding set of six fire units called FU_As and FU_Bs. All fire missions are generated exponentially at the OC_A or OC_B and randomly distributed to their respective fire units. A load of 600 fire missions on Net_A means that 300 fire missions of Thread 1 and 300 fire missions of Thread 2 are generated at OC_A. A load of 600 fire missions on Net_AB means that 150 fire missions of Thread 1 and 150 fire missions of Thread 2 are generated at OC_A and 150 fire missions of Thread 1 and 150 fire missions of Thread 2 are generated at OC_B. The fire mission load is distributed randomly across the two operation centers and subsequently across their fire units. In addition, situational awareness (SA) type messages of 30 bytes each are generated exponentially with mean of 60 seconds at every node. This data is transmitted to all other nodes using unacknowledged Type 1 services. The OPNET model developed includes modified pipeline stages that provide an accurate simulation of the ReedSolomon (32,12) error correction algorithm. The ReedSolomon (RS) model incorporates 32-hop barberpole interleaving for an input RS data block size of 1440 bits and RS transmission block size of 4608 bits. Each transmission unit includes a frequency hoping synchronization block of 2304 bits and the first RS block of each transmission unit includes a message indicator and header information. See [2], [3], [4], [7] for a discussion of RS and structure of the RS error blocks. RS error correction is not part of MIL-STD188-220B but 24/12 Golay error correction is part of the standard. Earlier papers [10], [11] that compared DAP-NAD and RE-NAD were based on MIL-STD-188-220B using Golay error correction. However, RS error correction is generally more efficient and is an error correction technology used in some of the radios currently used on tactical networks. The error correction block size for 24/12 Golay with time dispersal coding (TDC) is 384 bits versus 4608 bits for RS. A careful analysis of message transmission blocks shows the effect of the Type 4 acknowledgments would be more pronounced for Golay than it is for RS. The reason is that for most cases several 64 bit Type 4 acknowledgment packets can be included in the 4608 bits of the RS error block with no additional bits transmitted since the RS error block must be padded with fill bits to the 4608 bits. Consequently, there is usually no additional cost for the Type 4 acknowledgment packet. However, the additional bits will frequently require an additional 384 bit TDC Golay error block. Each of the FS threads has delay time for human/machine interaction or support built into the threads. Threads 1 and 2 have about 105 and 65 seconds, respectively, of delay incorporated into their execution. The results reported give the total execution time for the various threads. Many of the graphs for the completion time of fire threads are scaled to account for this delay time. The reporting of the packet hop times or message latency is not scaled in this fashion since these values represent the actual times. Message precedence can be set to urgent, priority or routine. According to MIL-STD-188-220B, Type 4 acknowledgments have the same priority as the messages that they are acknowledging. Type 2 messages piggyback their acknowledgments on any precedence message but can set their acknowledgments to any precedence level if a separate ACK packet is generated. The FS messages were determined to have priority precedence and hence, for Type 4, so did their acknowledgments. The effect on the network of priority packets for Type 4 acknowledgments was that the network stayed in the priority precedence for the entire simulation unless the network was extremely lightly loaded. For all of our Type 4 simulation runs, the precedence of the network remained at priority. In order to fairly compare Type 2 and 4, the precedence of all messages for Type 2 was therefore set at priority. Simulation has shown that the precedence effects for DAP-NAD and RE-NAD can be significant but much work needs to be done before any definitive results can be presented. (A detailed study of message and network precedence and their effect on the performance of the 2 0-7803-5538-5/99/$10.00 (c) 1999 IEEE packets being retransmitted while Type 4 retransmits packets when communication system used on tactical networks is warranted.) All simulation runs were executed for 13,000 seconds with the SA type messages being generated by all fire units for the entire simulation run. The FS threads were generated from the beginning until 10,000 seconds. This allowed all FS threads to run to completion unless an intermediate message in the thread was not able to be received by the necessary node. The thread then becomes suspended and does not complete. The simulation runs include several sets where significant random noise is introduced on the transmission medium to test how the networks respond to errors. The random noise is of sufficient magnitude that it partially disrupts the communication channel and the RS error correction is not adequate to correct the subsequent errors. The noise level was set high enough so that RS could not correct many of the resulting errors but sufficiently low so that most FS threads could run to completion. It also needed to be high enough to test the retransmission differences between Type 2 and Type 4. The effect of the channel noise is that approximately 30% of the packets are not received correctly. Noise was used to introduce errors only for the DAP-NAD transmissions since DAP-NAD was the primary MAC algorithm used. All results in this paper use DAP-NAD unless explicitly stated otherwise. Noise was not used with RE-NAD because transmission failures resulting from collisions (primarily soft collisions) cause more than sufficient retransmissions. III. PROTOCOL CONFIGURATIONS The Type 2, Type 4, DAP-NAD, and RE-NAD protocols will be discussed only briefly and the reader is referred to MIL-STD-188-220B for a complete description. There are obviously too many details to be included in a paper of this length. Type 2 services are connection oriented with decoupled acknowledgments and are based on the HDLC (high-level datalink control) protocol. Type 2 services provide high reliability because of the connection state but do not require individual acknowledgments. In most heavily used networks, Type 2 acknowledgments are included in other packet headers via piggy backing and incur little or no additional overhead. Type 4 uses decoupled acknowledgments (separate ACK packet) for every individual Type 4 packet. For reliable transfer of large amounts of data, Types 2 and 4 are generally used because of the decoupled nature of the acknowledgments facilitates concatenation. Type 2 is considered more efficient since it does not require separate acknowledgments for each packet. For both Type 2 and Type 4, the acknowledgment packets are only 64 bits and so the additional overhead for Type 4 may not be that significant. However, Type 4 differs substantially from Type 2 when there are network transmission errors. Type 2 implements selective retransmission with only lost TABLE I. Type 2 and Type 4 Parameters Type 4 Parameters Type 2 ACK_timer 38.0 38.0 P_Bit_timer 19.0 Reject_timer 38.0 Response_Delay_timer 7.0 Max_Retransmissions 3 3 K_Window 64 20/100 K2_Threshhold 32 K3_Threshhold 64 the packet or its acknowledgment is lost. Type 4 will retransmit the packets when its ACK timer expires whereas Type 2 will issue a request to the receiver for an acknowledgment. The Type 2 and Type 4 parameters are given in Table I. For Type 2, the K_Window parameter represents the maximum number of outstanding (unacknowledged) information frames allowed on a connection and it did not constrain the network performance in any way. The need for a large Type 2 K_Window is further limited by the response delay timer and the K2 threshold. For Type 4, the K_Window parameter represents the maximum number of unacknowledged frames allowed for a transmitting station or node. The range of K_Window parameter for Type 4 is 5 to 20 according to MIL-STD-188-220B [5, Appendix E]. Several sets of simulations were executed with the K_Window parameter set to 20 to show the limitation it has on the performance of the networks. Most of the simulations were run with this parameter set to 100 so as to test other performance aspects of Type 2 versus Type 4. For almost all simulation runs, K_Window = 100 for Type 4 was adequate to not impact network performance. DAP-NAD is a method of generating network access delays, which provides every subscriber an equal opportunity to use the network. It is a dynamic protocol and is deterministic only in that each subscriber can determine the maximum amount of time before its next network access opportunity. From any node’s point of view, if no other subscriber starts transmitting before that node’s networkaccess-time occurs then that node will transmit. The formulas for computing the access slots are given in [5], [8]. The RE-NAD algorithm dynamically controls network access based on network load, network topology, and load factors. RE-NAD uses two levels of algorithms for media access control: (1) modem to radio; the continuous scheduler, defined in the standard, and (2) radio to radio; the radio embedded portion, not defined in the standard. 3 0-7803-5538-5/99/$10.00 (c) 1999 IEEE RE-NAD uses a continuous scheduler interval computed as sum of fixed part and a random part; range is 1.0 to 30.0 seconds. The second level of the algorithm uses contention slots to access the medium. Both levels of RE-NAD’s MAC are, in part, random. The contention slot period of RE-NAD is not part of MILSTD-188-220B. The contention period using the CSMA contention slots follows the protocol given by Whitehall [12]. The radio embedded part of the medium access control scheme is based on voice, urgent, priority, and routine messages each being assigned a group of slots in which they may be transmitted. Various slot configurations were used during tests and tuning of RE-NAD. The slot configurations used for the simulation were the most favorable for this simulation application of any slot configuration tested. The slot groups may overlap and the slots are numbered starting at zero. The slot duration was 0.018 seconds and there are nine priority slots beginning at slot 12. The net_busy_detect_time = 0.1 seconds was used for all RE-NAD and DAP-NAD simulation runs. This time is appropriate for radio configurations that have a squelch pin, which includes most current production radios used on tactical networks. It is important for network performance that the squelch pin be used and configured into the network operation. For the simulation, the timers used for DAPNAD were TP_time = 0.11 and the DTETURN_time = 0.01 seconds. Larger values of the net_busy_detect_time significantly decrease the performance for Type 2 and 4 and for RE-NAD and DAP-NAD. The raw bit capacity of the networks was assumed to be 16,000 bps. After RS encoding, that capacity is reduced to approximately 5000 bps. This does not include any overhead for the synchronization or COMSEC headers. The simulation accounted for all such headers. present the following results. All simulation results presented are the average of a number of simulation runs with different random number seeds. The metrics of particular interest for this paper are the time to completion of the FS threads and the packet-hop-time (latency) for Type 2 and Type 4 packets. The latency time is the time from when the packet is queued for transmission at the transmitting node until it is received at the datalink layer by the receiving node. This represents the individual packet latency on a single link. The completion times of the FS threads are important for FS applications. Generally, these two measures of performance are consistent and may vary at different rates but this is not always the case. SA type messages are generated by every fire unit at an average rate of one every 60 seconds and their latency is reported. This is not a key performance parameter since the average latency from the fire units is generally reasonable. The performance benchmarks given in Figs. 1 and 2 are for error free networks with the K_Window = 100 for Type 4 service. The scenario for Figs. 1, 2, and 3 is the most favorable to the optimum performance of the networks and the protocols. Ne t_ A A v e r a g e Fir e M is s io n C o m p le t io n ( DA P- NA D) 300 T im e ( se c ) 250 Th r e a d 1 Ty p e 4 200 Th r e a d 1 Ty p e 2 Th r e a d 2 Ty p e 4 150 Th r e a d 2 Ty p e 2 100 50 200 300 400 500 600 700 800 900 The message precedence assumptions of priority were used for the simulation in order to make a fair comparison between Type 2 and Type 4 services. As indicated earlier, Type 4 acknowledgments are at the same priority as the original message. We assigned all messages the priority precedence which forced the network to stay in priority precedence. A great deal more should be done with the effect of precedence on network performance for the various protocols. The effect could be quite significant and a separate study should be initiated to do a careful analysis. It is not addressed in this paper. L o a d ( m is s io n s p e r h o u r ) Fig. 1. Comparison of Types 2 and 4 Fire Mission Completion Times on Net_A IV. SIMULATION RESULTS Benchmarks of Type 2 and Type 4 performance for Net_A and Net_AB are presented in Figs. 1, 2 and 3. These results are also used to make comparisons with subsequent simulation runs. Fig. 1 graphs the completion time of the two FS thread types on the seven node network, Net_A. Type 4 services clearly reach a knee in the time-to-completion curve around a load of 800 fire missions per hour. Recall that these fire mission threads have short message sequences. Type 2 services provide better performance throughout the whole loading range and superior performance at the heaviest loads. There are many performance metrics of interest and more than 48 metrics are collected per node as well as numerous global statistics for each simulation run. More than a thousand simulation runs have been made and analyzed to The completion times of the FS threads for the 14 node network, Net_AB, are given in Fig. 2. The performance knee for the seven node network is much sharper than the knee in the 14 node network. The reason is that the traffic generated 4 0-7803-5538-5/99/$10.00 (c) 1999 IEEE by the 800 or more fire missions and their extra ACK packets for Type 4 on the single operations center is too much for the single node to handle and messages must wait there longer before transmission. Recall that for the 14 node network, the fire thread load is split between two operation centers. The time to completion increases more steeply for the larger network but not as dramatically at the knee of the curve. This reflects the slight decrease in performance as the network size increases. For the lighter loads, Type 2 and Type 4 performance is very close but Type 2 consistently performs better than Type 4. Ne t _ A B A v e r a g e Fir e M is s io n C o m p le t io n ( D A P- N A D ) 300 of unacknowledged packets to all destinations from any individual transmitting node to 20 packets. Most of the simulations were run with a maximum window size of 100 at each node. Several simulations for Type 4 with K_Window size equal to 20 were run to determine the effect this has on the performance of the system under various configurations. When Net_AB was simulated using RE-NAD, Type 4, load 500, and K_Window = 100, the average and maximum number of unacknowledged messages averaged over the two operations centers was approximately 28 and 70 respectively. Similarly, on Net_A with errors using DAP-NAD, the average and maximums were 41 and 70 respectively. For Net_A with 30% errors, the effect on thread completion time is given in Table II. Obviously, the 500 fire mission load is beyond the capacity of the network with K_Window = 20. T im e ( s e 250 TABLE II. Comparison of Thread Completion Times for Type 2 and Type 4 with a Load of 500 on Net_A with Errors (30%) with 100 and 20 Window Size Th r e a d 1 Ty p e 4 200 Th r e a d 1 Ty p e 2 Th r e a d 2 Ty p e 4 150 Th r e a d 2 Ty p e 2 Type 2 64 Window 100 50 200 300 400 500 600 700 800 900 Thread 1 Thread 2 L o a d ( m is s io n s p e r h o u r ) Fig. 2. Comparison of Type 2 and Type 4 Fire Mission Completion Times on Net_AB. Ne t _ A a n d Ne t _ A B L a t e n c y ( DA P- NA D) 60 T im e ( s e c 50 40 Ne t_ A B Ty p e 4 Ne t_ A B Ty p e 2 30 Ne t_ A Ty p e 4 Ne t_ A Ty p e 2 20 10 0 200 300 400 500 600 700 800 900 L o a d ( m is s io n s p e r h o u r ) Fig. 3. Comparison of Type 2 and Type 4 Latency on Net_A and Net_AB The packet latency times for Type 2 and Type 4 packets for both Net_A and Net_AB are given in Fig. 3. The sharp increase in the packet latency for Net_A at 800 fire mission again reflects the difficulty of having a single node with a very heavy traffic load to transmit. For the Net_AB, this load is split between the two operations centers and it makes a more balanced network and a more efficient utilization of network resources. MIL-STD-188-220B stipulates that Type 4 (but not Type 2) services have a maximum of 20 packets (range 5 – 20) in the message window (K_Window). This limits the number 269.2 sec. 174.0 sec. Type 4 100 Window 554.1 sec. 365.3 sec. Type 4 20 Window 3165.7 sec. 2350.2 sec. Fig. 4 gives the graph for the average latency times for the same Net_A with errors. The window restriction of 20 for Type 4 makes these packet latencies become unacceptable and go off the chart. Fig. 4 also gives the packet latency for the SA (Type 1) packets. SA type data continues to be delivered in a timely manner even under the most adverse conditions. The average delivery times for the Type 1 packets varies from about 2 seconds for the 200 fire mission load to about 4 seconds for the 500 fire mission load. As can be seen from the bottom two lines in the graph of Fig. 4, the effect on Type 1 message latency of using Type 2 or Type 4 for FS messages was not as significant as many other message latencies in this study. The reason for this is that Type 1 messages are generated by all nodes at approximately the same rate instead of having just two nodes generate the initial messages for the FS threads. Again, the performance of Type 1 messages consistently did better using Type 2 rather than Type 4 for the FS messages. The effect of Type 4 (K_Window of 20 and 100) versus Type 2 on the completion time of fire threads is shown in Fig. 5. A comparison of Fig. 5 and Fig. 1 shows that the performance of the network has dramatically decreased due to errors for the Type 4 services. Type 2 continued to perform quite well. It should be noted that the fire mission load for Figs. 4 and 5 varies from 200 to 500 fire missions per hour whereas the load for Figs. 1, 2, and 3 goes up to 900 fire missions per hour. 5 0-7803-5538-5/99/$10.00 (c) 1999 IEEE was used since that would be less advantageous to DAPNAD (generally considered better on smaller nets) for any possible comparisons between MAC algorithms. The network for RE-NAD had no external noise. Fig. 6 gives the packet latency. The latency for Type 1 packets was acceptable across the entire loading of FS threads. Again, Type 2 for the FS messages allowed better performance for the Type 1 messages than did Type 4. However, the message latency for Type 4 messages was clearly unacceptable for even modestly loaded networks. Type 2 performed well for the all fire mission loads using RE-NAD. However, comparison of Fig. 6 (RE-NAD) and Fig. 3 (DAP-NAD) shows the message latency was much lower for DAP-NAD with an average of approximately 5 seconds for DAP-NAD versus 36 seconds for RE-NAD at the 500 fire mission per hour load. Similar observations can be made for the thread completion times using Fig. 7 for RE-NAD and Fig. 2 for DAP-NAD. 200 180 160 140 120 100 80 60 40 20 0 Type 4 (w indow 100) Type 2 Type 1 (w ith Type 4 w indow 100) Type 1 (w ith Type 2) Type 4 (w indow 20) 200 300 400 500 Load (m is s ions pe r hour ) Fig. 4. Effect of Errors on Type 2 and Type 4 (100 and 20 Window) on Latency on Net_A Ne t_ A T h r e a d Co m p le tio n w ith 3 0 % Er r o r s (DA P-NA D) 800 600 500 Th r e a d 2 Ty p e 4 ( w in d o w 2 0 ) 80 Th r e a d 1 Ty p e 4 70 Ty p e 4 60 400 T im e ( s e c ) Th r e a d 1 Ty p e 2 Th r e a d 2 Ty p e 4 20 Th r e a d 2 Ty p e 2 T y p e 1 ( w ith T y p e 2) 10 L o a d ( m is s io n s p e r h o u r ) Fig. 5. Effect of Errors on Type 2 and Type 4 Mission Completion Times on Net_A 0 0 0 0 50 45 40 10 0 0 0 500 0 400 30 300 0 200 T y p e 1 ( w ith T y p e 4) 30 25 100 40 0 200 Ty p e 2 50 20 300 15 T im e ( s e c ) N e t_ A B L a te n c y (R E -N AD ) Th r e a d 1 Ty p e 4 ( w in d o w 2 0 ) 700 35 Time (sec) Ne t_AB Late ncy w ith 30% Errors (DAP-NAD) L o a d ( m is s io n s p e r h o u r ) DAP-NAD was used as the MAC algorithm for all of the previous results. DAP-NAD is a deterministic high performance MAC algorithm and was ideal for testing Type 2 versus Type 4. The other most important MAC algorithm in MIL-STD-188-220B is RE-NAD. Type 2 and 4 services needed to be tested with it as well. The larger net, Net_AB, Fig. 6. Type 2 and Type 4 Latency for RE-NAD on Net_AB Ne t_A B Th rea d Co m plet ion (R E -N AD ) 1 00 0 9 00 8 00 T im e ( sec ) Network utilization, viewed as raw bits seen on the network, for the Net_A with errors (the networks for Figs. 4 and 5) is 6% higher for Type 4 than for Type 2 for the 500 load with the 100 window. The reason is that Type 4 retransmits packets when the sender has its ACK_timer expire even if the packets were correctly delivered and only the ACK was destroyed. Type 2 issues only an acknowledgment request. The network utilization with DAPNAD gets as high as 85% with many transmission units at the user configurable maximum size of 57600 bits and averaged over 25,000 bits. Each Type 4 and Type 2 transmission unit had on average 8.9 and 4.2, respectively, packets. Half of the Type 4 packets must be datalink layer acknowledgments. The Type 4 retransmissions do allow slightly shorter fire thread completion times for the very lightly loaded networks. This is the only known case where Type 4 outperformed Type 2. 7 00 Threa d 1 Ty pe 4 6 00 Threa d 1 Ty pe 2 5 00 Threa d 2 Ty pe 4 4 00 Threa d 2 Ty pe 2 3 00 2 00 1 00 10 0 1 50 2 00 2 50 3 00 3 50 4 00 45 0 5 00 L oa d (m i ss io ns p e r h ou r ) Fig. 7. Type 2 and Type 4 Thread Completion Times RE-NAD on Net_AB 6 0-7803-5538-5/99/$10.00 (c) 1999 IEEE V. CONCLUSIONS The results presented here and the simulations showed that Type 2 services are, in general, more efficient than Type 4 services for FS type networks in all measures of network performance. Type 2 is dramatically more effective than Type 4 when any of the following hold: (1) the network load approaches maximum capacity, or (2) when the network has a reasonable number of packets received in error and the load was at least moderate, or (3) when using RE-NAD instead of DAP-NAD. These results were obtained using FS threads and SA type messages to load the network and were consistent for both the seven and 14 node network. The large size of the RS error block mitigates some of the effects of the Type 4 acknowledgments on error free lightly loaded networks. The MIL-STD-188-220B Type 4 parameter (K_Window) for the maximum number of frames outstanding from a node has the range of 5 to 20 packets. This is clearly a severe limit on the performance of the network under error or under heavy load conditions. For even moderately loaded error free networks using RE-NAD, the maximum window of 20 for Type 4 was an additional limit on the performance of RENAD. It is strongly recommended that this parameter value be changed if Type 4 services are going to continue to be used. The K_Window size of 64 for Type 2, which is half of its maximum size allowed by MIL-STD-188-220B, was never a constraint for Type 2 network performance and was never reached. As was mentioned, message and network precedence are very important for the performance of these protocols. A thorough study needs to be made if the MILSTD-188-220B protocols are going to be used on tactical networks. RE-NAD did not perform as well as DAP-NAD in general. This was shown by using Figs. 6 and 3 to compare message latency and Figs. 7 and 2 to compare thread completion times for both Type 2 and Type 4. DAP-NAD with either Type 2 or Type 4 performed very well and much better than RE-NAD. Figs. 2, 3, 6 and 7 show that Type 4 with RE-NAD performed significantly less well than any other protocol combination. These results have significant implications for tactical networks used for FS. The results presented here for the seven and 14 node networks indicate that the results will scale to larger networks with only modest decreases in the performance. For modest sized tactical internets, Type 2 and DAP-NAD appears to be the choice for performance. REFERENCES [1]. Bowman, et al., “Improved Performance for Integrated Voice and Data in a Tactical Packet Radio Network,” MILCOM’98 Conference Proceedings, Oct. 18-21, 1998, pp. 447-451. [2]. Hamilton, Bradley J., “SINCGARS System Improvement Program (SIP) Specific Radio Improvements,” Proceedings of the 1996 Tactical Communication Conference, April 30-May 2, 1996, pp. 395-406. [3]. Field Demonstration and Test Report for the SINCGARS Enhanced Point-to-Point FEC Data Evaluation, ITTA/CD NO. 31310/FEC-03-3, 24 September 1993. [4]. System Engineering Document for Ground Radio SINCGARS System Improvement Program (SIP) document (ITT-A/CD NO. A3214186, March 22, 1996). [5]. MIL-STD-188-220B "Interoperability Standard for Digital Message Device Transfer Subsystems", 20 January 1998. [6]. Nicholas, Jonathan M. and Albert Futernik, “Queuing Structures in SINCGARS Radio Networks,” Oct. 18-21, 1998, MILCOM’98 Conference Proceedings, pp. 211-217. [7]. Pursley, Michael B. “Reed-Solomon Codes in Frequency-Hop Communications” in Reed-Solomon Codes and Their Applications, Stephen B. Wicker and Vijay K. Bhargave (eds.). 1994, IEEE Press. [8]. Thuente, David J., Craig Brown, Tim Borchelt, Ed Hill, “AFATDS Network Simulation Report,” Magnavox Electronic Systems Co., Fort Wayne, IN, Oct. 13, 1995. [9]. Thuente, David J., Craig Brown, Tim Borchelt, Ed Hill, “The Design and Analysis of the AFATDS Communication Networks using Simulation,” Proceedings of the 1996 Tactical Communication Conference, April 30-May 2, 1996, pp. 267-280. [10]. Thuente, David J., and Tim Borchelt, “Simulation Studies of MAC Algorithms for Combat Net Radio,” MILCOM’97 Conference Proceedings, Nov. 2-5, 1997, pp. 193-199. [11]. Thuente, David J., and Tim Borchelt, “Simulation Model and Studies of MIL-STD-188-220B,” MILCOM’98 Conference Proceedings, Oct. 18-21, 1998, pp. 198-204. [12]. Whitehill, Eric “Use of End-to-End Acknowledgments in MIL-STD-188-220A Extended Networks,” Proceedings 1996 Tactical Communication Conference, April 30-May 2, 1996, pp. 495-502. [13]. Yoon, C. J. and C. Chiu Chan, “Efficient Net Access Scheduler in MIL-STD-188-220A,” MILCOM’98 Conference Proceedings, Oct. 18-21, 1998, pp. 253-257. [14]. Yoon, C. J. and C. Chiu Chan, “MIL-STD-188-220A Parameter Optimization for Tactical Internet,” MILCOM’98 Conference Proceedings, Oct. 18-21, 1998, pp. 960-965. 7 0-7803-5538-5/99/$10.00 (c) 1999 IEEE