A SURVEY TO REAL-TIME MESSAGE-ROUTING NETWORK SYSTEM WITH KLA MODELLING

International Journal in Foundations of Computer Science & Technology (IJFCST) Vol.7, No.3/4, July 2017
DOI:10.5121/ijfcst.2017.7401 1
A SURVEY TO REAL-TIME MESSAGE-ROUTING
NETWORK SYSTEM WITH KLA MODELLING
Paul T. Chiou, Yu Sun and G S. Young
Department of Computer Science, California State Polytechnic University, Pomona,
California USA
ABSTRACT
Messages routing over a network is one of the most fundamental concept in communication which requires
simultaneous transmission of messages from a source to a destination. In terms of Real-Time Routing, it
refers to the addition of a timing constraint in which messages should be received within a specified time
delay. This study involves Scheduling, Algorithm Design and Graph Theory which are essential parts of
the Computer Science (CS) discipline. Our goal is to investigate an innovative and efficient way to present
these concepts in the context of CS Education. In this paper, we will explore the fundamental modelling of
routing real-time messages on networks. We study whether it is possible to have an optimal on-line
algorithm for the Arbitrary Directed Graph network topology. In addition, we will examine the message
routing’s algorithmic complexity by breaking down the complex mathematical proofs into concrete, visual
examples. Next, we explore the Unidirectional Ring topology in finding the transmission’s
“makespan”.Lastly, we propose the same network modelling through the technique of Kinesthetic Learning
Activity (KLA). We will analyse the data collected and present the results in a case study to evaluate the
effectiveness of the KLA approach compared to the traditional teaching method.
KEYWORDS
Real Time Routing, Routing Messages, Network Routing, Graph Theory, Scheduling, Arbitrary Directed
Graph, Unidirectional Ring, Kinesthetic Learning Activity, KLA, CS Education
1. INTRODUCTION
In the age of multimedia today, almost everything is transferred through the Internet. As the
quality of multimedia rapidly increases over the past decade, the demand for a faster network
bandwidth is also growing at a significant rate. There are many applications relying on real-time
routing services. For instance, in a live messaging system, each participant interacts with the
others by sending and receiving messages in the form of texts, pictures, audio sequences, or video
frames [1]. Each message, originating from a participant, must be delivered to all the others in a
real-time manner. All these real-time systems are constantly in need to optimize their Quality of
Service provided to the users [2].The transfer speed is bounded by the network infrastructure
itself in the hardware level. However, it is essential to theoretically understand how these routing
take place so we can attempt to optimize for a more efficient algorithm. This is indeed an area of
research for CS students that crosses multiple domains such as Scheduling, Algorithm Design,
Complexity Analysis, NP Theory, Graph Theory, etc. According to the ACM and IEEE, CS
Education is a pathway to innovation and creativity that requires an understanding of the
fundamentals problem-solving methodology of computational thinking [3]. In our research, we
will introduce the model of a real-time system and present its method for scheduling
messages.We shall then propose and incorporate our own modelling technique using the
innovative non-traditional KLA to attempt a more effective learning approach for CS Education.

2
2. INTRODUCTION TO REAL-TIME MESSAGE ROUTING
As mentioned, in a distributed real-time system, the problem of determining whether a set of
messages can be sent on time becomes an important issue. In the last two decades, real-time
message routing has been studied extensively on various types of networks [4]. An on-line
routing algorithm is one that routes messages without any knowledge of future arrivals of
messages. Based on the different level of constraints, the complexities can vary from easy (being
solvable in polynomial time) while others are NP-hard [5]. Next, we shall first introduce the
fundamental notion of various restrictions of the four parameters- origin node, destination node,
release time, and deadline.
2.1. The Formal Model and Notation
Wewill follow the scheduling standard and represent a network by a directed graph𝐺 = (𝑉, 𝐸)
shown in Figure 1, where each vertex in the set𝑉 represents a node of the network and each
directed edge in set𝐸 represents a communication link. If (𝑢, 𝑣) ∈ 𝐸, then there is a transmitter in
node 𝑢 and a receiver in node 𝑣 dedicated to the communication link (𝑢, 𝑣) in Figure 2. We
assume that a node can simultaneously send and receive several messages if they are transmitted
on different communication links [6].
Figure 1. Network as a Directed Graph (Left),Communication Link (Right)
In our rounding problem, a set of 𝑛 messages 𝑀 = (𝑀1, … , 𝑀𝑛) as shown in Figure 3 needs to be
routed through the network. Each message 𝑀𝑖 consists of its own characteristic properties
represented by a quintuple (𝑠𝑖, 𝑒𝑖, 𝑙𝑖, 𝑟𝑖, 𝑑𝑖). In this set, the 𝑠𝑖 denotes the origin node(i.e., 𝑀𝑖
originates from node 𝑠𝑖), 𝑒𝑖 denotes the destination node (i.e., 𝑀𝑖 is to be sent to node 𝑒𝑖), 𝑙𝑖
denotes the length (i.e., 𝑀𝑖 consists of 𝑙𝑖 packets of information), 𝑟𝑖 denotes the release time (i.e.,
𝑀𝑖 originates from 𝑠𝑖 at time 𝑟𝑖), and 𝑑𝑖 denotes the deadline (i.e., 𝑀𝑖 must reach 𝑒𝑖 by time 𝑑𝑖).
These features are the constraints in which we will follow to vary the message routing
complexity. The examples of these message routing problems consist of a network 𝐺 and a set of
messages 𝑀. In our model, we further assume that it takes one unit of time to send a packet
through a communication link. This implies that it takes length 𝑙𝑖 time units for message 𝑀𝑖 to
traverse on any communication link. Also, there exists a central controller which has complete
information on the topology of the network and the characteristics of each messages in advance.
Our goal is to determine if the messages in 𝑀 can be routed through the network 𝐺 such that each
message 𝑀𝑖 is sent from node 𝑠𝑖 to node 𝑒𝑖 satisfying the given time interval of [𝑟𝑖, 𝑑𝑖] [5], [6].
2.2. Message-Routing Network System
A message-routing network system is a given system that consists of the network itself and the
messages to be transmitted across it. In notation, we let 𝑀𝑅𝑁𝑆 = (𝐺, 𝑀) be the “Message-

3
Routing Network System”, where 𝐺 = (𝑉, 𝐸) is a network and 𝑀 = (𝑀1, 𝑀2, … , 𝑀𝑛) is a set of 𝑛
messages to be routed through 𝐺. [5], [6].
A transmission 𝑆 for 𝑀𝑅𝑁𝑆 is said to be feasible if the deadlines of all the messages are met; i.e.,
for all 𝑞 = (𝑢, 𝑣, 𝑀𝑖, 𝑡1, 𝑡2), 𝑡2 ≤ 𝑑𝑖. Here, 𝑡2 is the finishing time of the communication link. A
non-preemptive transmission, denoted by 𝑆𝑁𝑃, is a transmission such that for each
configuration(𝑢, 𝑣, 𝑀𝑖, 𝑡1, 𝑡2) ∈ 𝑆𝑁𝑃, 𝑡2 − 𝑡1 = 𝑙𝑖. A preemptive transmission, denoted by 𝑆𝑃, is a
transmission defined exactly as before. A 𝑀𝑅𝑁𝑆 is feasible with respect to non-preemptive
(preemptive) transmission if there is a feasible non-preemptive (preemptive) transmission for
𝑀𝑅𝑁𝑆 [6].
3. REAL TIME ROUTING COMPLEXITY ANALYSIS
3.1. Arbitrary Directed Graph
In this section, we present a Message-Routing Network System 𝑀𝑅𝑁𝑆with an arbitrary directed
graph as shown in the Figure 2. We let 𝑀𝑅𝑁𝑆 = (𝐺, 𝑀), where 𝐺 is a network given and 𝑀 =
{𝑀1, 𝑀2, 𝑀3, 𝑀4}. The four messages to be transferred are defined with the following constraints:
𝑀1 = (1, 4, 3, 0, 6), 𝑀2 = (1, 4, 2, 2, 6), 𝑀3 = (3, 5, 2, 0, 6), and 𝑀4 = (4, 5, 4, 0, 4). The table in
Figure 9 shows an organized view. Recall the quintuple defined (𝑠𝑖, 𝑒𝑖, 𝑙𝑖, 𝑟𝑖, 𝑑𝑖) which represents
the origin node, destination node, length, release time, and deadline [5, 6].
Figure 2. Arbitrary Directed Graph Network
𝑀𝑖 𝑠𝑖
origin node
𝑒𝑖
destination node
𝑙𝑖
length
𝑟𝑖
release time
𝑑𝑖
deadline
𝑀1 1 4 3 0 6
𝑀2 1 4 2 2 6
𝑀3 3 5 2 0 6
𝑀4 4 5 4 0 4
Table 1. Message Constraints of 𝑀 for the Arbitrary Directed Graph𝑀𝑅𝑁𝑆.
3.1.1. Non-Preemptive Transmission
First, we will examine a non-preemptive instance of the 𝑀𝑅𝑁𝑆. In non-preemptive transmission,
once a message is transmitted on a communication link (𝑢, 𝑣), it must continue until the entire
message is received by node 𝑣. We use the subscript 𝑁𝑃 to denote the transmission 𝑆 is non-
preemptive. With the message constraints in Table 1, we can construct the following
transmission
𝑆𝑁𝑃 = {(1, 2, 𝑀1, 0, 3), (2, 4, 𝑀1, 3, 6), (1, 3, 𝑀2, 2, 4), (3, 4, 𝑀2, 4, 6), (3, 4, 𝑀3, 0, 2),
(4, 5, 𝑀3, 4, 6), (4, 5, 𝑀4, 0, 4)}, as shown in Figure 3 [6].

4
Figure 3. Feasible Non-Preemptive Transmission for 𝑀𝑅𝑁𝑆.
𝑀1, 𝑀2 are departing from their origin node 1 to reach destination node 4. For this transmission,
𝑀1 will take route 1 → 2 → 4and 𝑀2 will take route 1 → 3 → 4 since this will be more efficient
having both messages transferred simultaneously during time 2 to 3, 3 to 4, and 4 to 6. Note that
at time 0, messages 𝑀1, 𝑀3, 𝑀4 are released from their origin nodes 1, 3, 4 respectively. Each of
the corresponding message lengths are 3, 2, 4. Therefore, messages 𝑀1, 𝑀3, 𝑀4 arrives at node
2, 4, 5 at times 3, 2, 4. Since 𝑀4’s destination node is 5, its transfer is then complete at time 4.
While 𝑀1, 𝑀3, 𝑀4 are transferring at time 2, 𝑀2 is ready to be released and it takes 2 units (its
length) time for it arrive at node 3. At time 3, when 𝑀1 finishes arriving at node 2, it can begin
its transfer to its destination node 4. At time 4, when 𝑀2 finishes arriving at node 3, it can begin
its transfer to its destination node 4. Note that the communication link (4, 5) is being occupied
by 𝑀4 so 𝑀3 must wait until it finishes (considering this is a non-preemptive transmission).
Finally, 𝑀3 begin its transfer to its destination node 5 along with 𝑀1, 𝑀2 [6].
Clearly, 𝑆𝑁𝑃 is a feasible non-preemptive transmission because all messages met their deadlines.
Thus, with this example, we can conclude this 𝑀𝑅𝑁𝑆 is feasible with respect to non-preemptive
transmission.
3.1.2. Preemptive Transmission
Unlike non-preemptive where all messages must finish its transfer once it starts, preemptive
transmission can be interrupted and resume later. This provides more flexibility and can be more
efficient although it adds complexity to the problem. In fact, if a 𝑀𝑅𝑁𝑆 is feasible with non-
preemptive transmission, it is also feasible with preemptive transmission [6].
In the following example, we will slightly modify the previous instance to demonstrate the
preemptive transmission case. Consider a new system 𝑀𝑅𝑁𝑆′
obtained from 𝑀𝑅𝑁𝑆 by
modifying 𝑀4 to be (4, 5, 2, 3, 5). Now 𝑀4 has a shorter message length of 2 and later deadline
of 5. However, its released time is postponed to 3. Table 2shows the change in grey.
𝑀𝑖 𝑠𝑖
origin node
𝑒𝑖
destination node
𝑙𝑖
length
𝑟𝑖
release time
𝑑𝑖
deadline
𝑀1 1 4 3 0 6
𝑀2 1 4 2 2 6
𝑀3 3 5 2 0 6
𝑀4 4 5 2 3 5
Table 2. Modified Message Constraints of 𝑀 for 𝑀𝑅𝑁𝑆′
This time in 𝑀𝑅𝑁𝑆′
, messages 𝑀1, 𝑀2 remains unchanged as 𝑀4 does not affect them.
However, 𝑀4 is constrained and cannot be released until time 3. When 𝑀3 finishes transferring
to node 4 at time 2, it can begin its transfer to its destination node 5without being idling between
time 2 to 3. Now at time 3, 𝑀4 is released and since this is preemptive transmission, 𝑀3 can be
interrupted to continue later. Hence, 𝑀4 begins and finishes arriving at its destination node 5

5
after 2 time units(its length). Finally, the interrupted 𝑀3 can now resume its transfer, completing
at destination node 5attime6.Thetransmission
𝑆′
𝑃{(1, 2, 𝑀1, 0, 3), (2, 4, 𝑀1, 3, 6), (1, 3, 𝑀2, 2, 4), (3, 4, 𝑀2, 4, 6), (3, 4, 𝑀3, 0, 2), (4, 5, 𝑀3, 2, 3),
(4, 5, 𝑀3, 5, 6), (4, 5, 𝑀4, 3, 5)}, is a feasible preemptive transmission. However, the modification
made it not feasible with respect to non-preemptive transmission. By examining Figure 4, we see
the reason is due to 𝑀3. At time 3 when 𝑀4 is ready to be released, without preemption, 𝑀3 must
keep transferring until it finishes at time 4 occupying communication link (4, 5) and 𝑀4 must
wait. When 𝑀4 is finally able to begin transfer at time 4, its message length of 2 will result a
finishing time of 6, therefore missing its deadline at time 5. Thus 𝑀𝑅𝑁𝑆′
is feasible with respect
to preemptive transmission, but not non-preemptive transmission [5], [6].
Figure 4. Non-feasible Preemptive Transmission for 𝑀𝑅𝑁𝑆′
3.1.3. Theorem and Proof
We now show that if the network is an arbitrary directed graph, the problem of determining
whether there is a feasible transmission (both preemptive or non-preemptive) for a message-
routing network system is NP-complete. The hardness is not caused by the message constraints
but the nature of the network topology because it is NP-complete even when all four parameters
are fixed [6]. Consider the following theorem:
3.1.3.1 Theorem 37.1
Given 𝑀𝑅𝑁𝑆 = (𝐺, 𝑀), where 𝐺 is an arbitrary directed graph and 𝑀 is a set of messages with
identical origin nodes, destination nodes, release times, and deadlines, the problem of
determining whether MRNS is feasible with respect to preemptive transmission is NP-complete
[6].
The theorem can be proved by the reduction from the 3-Partition problem, which is known to be
strongly NP-complete. Before showing the proof, we will explore the 3-Partition problem.The 3-
partition problem is a problem where one should partition 3𝑧 numbers (allowing duplicates) into
𝑧 groups of 3, such that each group has the same sum. If all 3𝑧 numbers sum to 𝑁, this means
that every group should have a sum of 𝑁/𝑧. The 3-partition problem is a well-known NP-
complete problem. Next, we use formal notation to precisely define the problem [6, 7].
Given a list, 𝐴 = (𝑎1, 𝑎2, … , 𝑎3𝑧) of 3𝑧 positive integers such that∑ 𝑎𝑖
3𝑧
𝑖=1 = 𝑧𝐵and each 𝑎𝑖
satisfying
𝐵
4
< 𝑎𝑖 <
𝐵
2
𝑓𝑜𝑟 𝑒𝑎𝑐ℎ 1 ≤ 𝑖 ≤ 3𝑧, can 𝐼 = {1, 2, … , 3𝑧} be partitioned into 𝑧 groups of
3, 𝐼1, 𝐼2, … , 𝐼𝑧 such that ∑ 𝑎𝑖
𝑖∈𝐼𝑗
= 𝐵 𝑓𝑜𝑟 𝑒𝑎𝑐ℎ 1 ≤ 𝑗 ≤ 𝑧 (each group sums to exactly 𝐵).
The 3-partition problem is similar to the partition problem, which in turn is related to the subset
sum problem. In the partition problem, the goal is to partition 𝑆 into two subsets with equal sum.
In 3-partition the goal is to partition 𝑆 into 𝑧 subsets, not just two subsets, with equal sum [7].
To show that the decision problem is in NP, notice that preemption is necessary only when a new
message arrives at a node. Thus, the number of preemptions is bounded by a polynomial function

6
of the size of the input, and hence we can guess a preemptive transmission in polynomial time
and verify that the transmission is feasible [6]. To complete the proof, the 3-Partition problem
reduction can be defined as follows:
Let 𝐴 = (𝑎1, 𝑎2, … , 𝑎3𝑧) be an instance of the 3-Partition problem. We construct an instance of
the message routing problem as follows: 𝑀𝑅𝑁𝑆 = (𝐺, 𝑀), where
𝐺 = (
{0, 1, 2, … , 𝑧 + 1},
{
(0, 1), (0, 2), … , (0, 𝑧),
(1, 𝑧 + 1), (2, 𝑧 + 1), … , (𝑧, 𝑧 + 1)
}
)
and 𝑀 = (𝑀1, 𝑀2, … , 𝑀4𝑧), and 𝑀𝑖 = (0, 𝑧 + 1, 𝑎𝑖, 0, 5𝐵) for each 1 ≤ 𝑖 ≤ 3𝑧 and 𝑀𝑖 =
(0, 𝑧 + 1, 2𝐵, 0, 5𝐵) for each 3𝑧 + 1 ≤ 𝑖 ≤ 4𝑧. We call the first 3𝑧 messages the partition
messages and the remaining 𝑧 messages the enforcer messages [5]. This is visually represented in
Figure 5
Figure 5. Visual Representation of 𝐺 and 𝑀 for 3-Partition 𝑀𝑅𝑁𝑆
Clearly, based on the definition, all messages in 𝑀 have identical origin nodes, destination nodes,
release times, and deadlines. Thus, construction can be done in polynomial time. Since the hard
3-partition is given by the oracle, the hard-part is done. The polynomial construction simply
involves calculating the extended message length and mapping the nodes. Adding the enforcers
is also be done in polynomial time [6].
For the reduced instance, there are 𝑧 distinct paths from vertex 0 to vertex 𝑧 + 1 in 𝐺. In any
feasible transmission for 𝑀𝑅𝑁𝑆, each enforcer message must be routed along a distinct path from
vertex 0 to vertex 𝑧 + 1. The remaining partition messages must be distributed among these 𝑧
paths [6]. This is intuitively correct because the enforcer messages essentially “jams up” each of
the route, reducing the complexity and possibilities of combinatorically assigning the partition
messages.
The above 𝑀𝑅𝑁𝑆 is feasible with respect to preemptive transmission if and only if 𝐴 has a 3-
Partition. We now show the case with a substantial instance of the reduction process. The
following is a “yes”’ instance of the 3-partition of 3𝑧 messages with 𝑧 = 3. Notice that our 𝐵 =
100, which is the sum of each partition with the elements satisfying size
100
4
< 𝑎𝑖 <
100
2
.
Otherwise, it would be impossible to have 3 parts each [7].
𝐴 = {𝑎1, 𝑎2, 𝑎3, 𝑎4, 𝑎5, 𝑎6, 𝑎7, 𝑎8, 𝑎9} = {26, 29, 33, 33, 33, 34, 35, 36, 41}
𝐼1 = {𝑎1, 𝑎3, 𝑎9} = {26, 33, 41}
𝐼2 = {𝑎2, 𝑎8, 𝑎7} = {29, 36, 35}
𝐼3 = {𝑎4, 𝑎5, 𝑎6} = {33, 33, 34}
These 3𝑧 elements maps to the corresponding partition messages. We then add 𝑧 more messages
𝑀10, 𝑀11, 𝑀12 as the enforcer messages. The set 𝑀 now consists of 4𝑧 messages:

7
𝑀 = {𝑀1, 𝑀2, 𝑀3, 𝑀4, 𝑀5, 𝑀6, 𝑀7, 𝑀8, 𝑀9, 𝑀10, 𝑀11, 𝑀12}
Figure 6. Message Constraints of 𝑀 for 3-Partition𝑀𝑅𝑁𝑆
Suppose 𝐼1, 𝐼2 … , 𝐼𝑧 is a solution to the 3-Partition problem. We can construct a feasible non-
preemptive (and hence preemptive) transmission as follows [6].
 For each 1 ≤ 𝑗 ≤ 𝑧, the messages 𝑀3𝑧+𝑗and 𝑀𝑖, where 𝑖 ∈ 𝐼𝑗, are transmitted from node
0 to node 𝑧 + 1 through the edges (0, 𝑗)and (𝑗, 𝑧 + 1). Shown in Figure 7.
 𝑀3𝑧+𝑗is transmitted through the edges (0, 𝑗)and (𝑗, 𝑧 + 1) in the time intervals [0, 2𝐵]
and [2𝐵, 4𝐵] respectively. Shown in Figure 8 (left).
 The messages 𝑀𝑖, where 𝑖 ∈ 𝐼𝑗, are nonpreemptively transmitted through the edges
(0, 𝑗)and (𝑗, 𝑧 + 1) in the time intervals [2𝐵, 3𝐵] and [4𝐵, 5𝐵], respectively. Shown in
Figure 8 (right).
Using the above algorithm, we created a feasible non-preemptive transmission using the
properties of our 3-partition instance as follows
Figure 7. Message Categorization of the Partitions.
 For 𝑗 = 1, 𝑀10, 𝑀1, 𝑀3, 𝑀9 are transmitted from node 0 to node 4 through the edges
(0, 1) and (1, 4) with 𝑀10 in the time intervals [0, 200] and [200, 400] and 𝑀1, 𝑀3, 𝑀9
non-preemptively transmitted in the time intervals [200, 300] and [400, 500]

8
Figure 8. Messages’ Distinct Paths (Left), and Transmission Times (Right)
Path distinctness in the proposed 𝑀𝑅𝑁𝑆 can be show by supposing 𝑆 is a feasible preemptive
transmission for 𝑀𝑅𝑁𝑆. Observe on the left of Figure 8 that there are exactly 𝑧 distinct paths
from node 0 to node 𝑧 + 1; namely, 𝑃1 = (0, 1, 𝑧 + 1), 𝑃2 = (0, 2, 𝑧 + 1), …, and 𝑃𝑧 =
(0, 3, 𝑧 + 1) [6]. In our case, there are exactly 3 distinct paths from node 0 to node 4 which are
𝑃1 = (0, 1, 4), 𝑃2 = (0, 2, 4), and 𝑃3 = (0, 3, 4).
With these distinct paths, each message must be transmitted through one of these 𝑧 paths and if
two enforcer messages are transmitted through the same path in 𝑆, then one of them will arrive
node 𝑧 + 1 at time 6𝐵 or later, and hence it will miss its deadline. Thus, we may assume that no
two enforcer messages are transmitted through the same path in 𝑆 [5, 6]. In our case, notice if
two enforcers are transmitted through the same path, one of them will arrive node 4after the
deadline600. This can be generalized for each path1 ≤ 𝑗 ≤ 𝑧, we let 𝐼𝑗 = {𝑖 | 1 ≤ 𝑖 ≤
3𝑧 and 𝑀𝑖 is transmitted through 𝑃𝑗 in 𝑆}.
Thus, for 𝑗 = 1,𝐼1 = {𝑎1, 𝑎3, 𝑎9} and 𝑀1, 𝑀3, 𝑀9 are transmitted through 𝑃1; for
𝑗 = 2, 𝐼2 = {𝑎2, 𝑎8, 𝑎7} and 𝑀2, 𝑀8, 𝑀7 are transmitted through 𝑃2 ;for
𝑗 = 3, 𝐼3 = {𝑎4, 𝑎5, 𝑎6} and 𝑀4, 𝑀5, 𝑀6 are transmitted through 𝑃3. Clearly, this is a feasible
non-preemptive transmission as shown on the right of Figure 8. Therefore, 𝑀𝑅𝑁𝑆 is feasible with
respect to preemptive transmission.
3.1.3.1.1.Proof by Contradiction
We have shown the theorem works for one case. To show that it holds true for all cases, we
provide a generic proof. We claim that 𝐼1, 𝐼2, … , 𝐼𝑧 is a solution to the 3-Partition problem. If the
claim is not true, then there is an index 𝑘 such that ∑ 𝑎𝑖
𝑖∈𝐼𝑘
> 𝐵. In other words, 𝐼𝑘 is the 3 set
whose sum exceeds 𝐵 = 100.
Let 𝑋 ∪ {𝑀′} be the set of messages transmitted through the path 𝑃𝑘 in 𝑆, where 𝑋 = {𝑀𝑖 | 𝑖 ∈ 𝐼𝑘}
and 𝑀′
is the enforcer message. Let 𝑡′
be the time instant at which 𝑀′
is completely transmitted
to node 𝑘, and let 𝑋1 and 𝑋2 be a partition of 𝑋 such that 𝑋1 and 𝑋2 contain all the messages in 𝑋
that are completely transmitted to node 𝑘 before and after time 𝑡′
, respectively [6].
Clearly, the total time taken to transmit all the messages in 𝑋 ∪ {𝑀′} is defend in the relationship:
𝑡′
+ (2𝐵 + ∑ 𝑙𝑖
𝑀𝑖∈𝑋2
) ≥ ( ∑ 𝑙𝑖
𝑀𝑖∈𝑋1
+ 2𝐵) + (2𝐵 + ∑ 𝑙𝑖
𝑀𝑖∈𝑋2
) = 4𝐵 + ∑ 𝑙𝑖
𝑀𝑖∈𝑋
> 5𝐵
We can interpret the above inequity by dividing the terms into 3 groups. The first group denoted
by 𝑡′
+ (2𝐵 + ∑ 𝑙𝑖
𝑀𝑖∈𝑋2
) is the time (𝑡′
) the enforcer message 𝑀′
spent on the first link (0, 𝑘) to

9
completely arrive at the intermediate node 𝑘. This is summed with 2𝐵 (the length of 𝑀′
) and the
total time (length) of the partition messages to completely arrive to the destination (∑ 𝑙𝑖
𝑀𝑖∈𝑋2
) on
the next link (𝑘, 𝑒). The second group denoted by (∑ 𝑙𝑖
𝑀𝑖∈𝑋1
+ 2𝐵) + (2𝐵 + ∑ 𝑙𝑖
𝑀𝑖∈𝑋2
) is to
account for the scenario if there are partition messages that arrives to intermediate node 𝑘 before
the enforcer message 𝑀′
does at 𝑡′
. This portion spent on the first link (0, 𝑘) is ∑ 𝑙𝑖
𝑀𝑖∈𝑋1
plus the
𝑀′
itself. This (∑ 𝑙𝑖
𝑀𝑖∈𝑋1
+ 2𝐵) is then summed with (2𝐵 + ∑ 𝑙𝑖
𝑀𝑖∈𝑋2
) all the time span on the
next link (𝑘, 𝑒) as previously discussed. The third group is just algebraic manipulation but is
equally expressed as the total time 𝑀′
spent (2𝐵 + 2𝐵) on both links (0, 𝑘) and (𝑘, 𝑒) plus the
total time all other message spent denoted by ∑ 𝑙𝑖
𝑀𝑖∈𝑋 [6].
In our first non-preemptive example, analysis is trivial because all partition messages always
arrive after the enforcer message finish arriving node 𝑘 at time 𝑡′
so set 𝑋1 is empty. This can be
seen in Figure 9. Hence, 200 + (2𝐵 + 100) ≥ (0 + 2𝐵) + (2𝐵 + 100) = 4𝐵 + 100 > 5𝐵.
Here, we have 𝑋1 = ∅ and 𝑀1, 𝑀3, 𝑀9 ∈ 𝑋2 for path 𝑃1 = (0, 1, 4); 𝑋1 = ∅ and 𝑀2, 𝑀8, 𝑀7 ∈ 𝑋2
for path 𝑃2 = (0, 2, 4); and 𝑋1 = ∅ and 𝑀4, 𝑀5, 𝑀6 ∈ 𝑋2 for path 𝑃3 = (0, 3, 4);
Figure 9. Non-Preemptive Transmission I (Enforcer First)
The following transmission in Figure 10 is another non-preemptive instance where some partition
messages arrive before the enforcer message finish arriving node 𝑘 at time 𝑡′
. We have 𝑀1 ∈ 𝑋1
and 𝑀3, 𝑀9 ∈ 𝑋2 for path 𝑃1 = (0, 1, 4); 𝑀2, 𝑀8, 𝑀7 ∈ 𝑋1 and 𝑋2 = ∅ for path 𝑃2 = (0, 2, 4); and
𝑀5, 𝑀4 ∈ 𝑋1 and 𝑀6 ∈ 𝑋2 for path 𝑃3 = (0, 3, 4);
Figure 10. Non-Preemptive Transmission II (Mixed Order)
According to the definition, transmitting preemptively works as well. The following Figure 11
shows a preemptive instance. Notice for the first link (0, 2) of route 𝑃2, 𝑀8 was chopped and
began transfer later after the enforcer 𝑀11 (which was also chopped several times).

10
Figure 11. Preemptive Transmission I (Feasible)
These feasible transmissions are essentially best-optimal solutions. However, the whole purpose
of such proof is to show that it is possible to have a non-optimal solution which exceeds 5𝐵. The
next example Figure 12 in take the previous in Figure 11 and slightly delays 𝑀6. Noticed now
𝑀6 on link (3, 4) must also be delayed because it cannot start until 𝑀6 is fully received on node
3. This then delays the enforcer 𝑀12, ultimately causing 𝑀5 to miss the deadline (shown in red).
Figure 12. Preemptive Transmission II (Non-Feasible)
Thus, at least one of the messages in 𝑋 ∪ {𝑀′} has its deadline missed, contradicting our
assumption that 𝑆 is a feasible transmission. We showed it with a counter-example and since
𝑀𝑅𝑁𝑆is feasible with respect to preemptive transmission if and only if 𝐴has a 3-Partition, if 𝑆 is
not feasible, therefore it is at least as hard as 3-partition [6].
3.1.3.2. Corollary 37.1
Given 𝑀𝑅𝑁𝑆 = (𝐺, 𝑀), where 𝐺 is an arbitrary directed graph and 𝑀 is a set of messages with
identical origin nodes, destination nodes, release times, and deadlines, the problem of
determining whether 𝑀𝑅𝑁𝑆 is feasible with respect to non-preemptive transmission is NP-
complete [6].
Based on the previous examples, we clearly see that the Theorem 37.1 also works with non-
preemptive transmission. Thus, Corollary 37.1. In fact, later in Section 4.2.2, we will show that
all non-preemptive transmissions are literally preemptive transmissions.
3.2. Unidirectional Rings
In the previous section, the Theorem and Corollary have shown the message routing problem a
computationally hard problem if the network is allowed to be arbitrary (Arbitrary Directed
Graph). We now relax we our focus to the simpler network topology class of Unidirectional
Rings in order to find the minimal makespan of the routing system. This is the simplest method
in demonstrating scheduling routing in any given 𝑀𝑅𝑁𝑆. The following Figure 13 shows the
network complexity hierarchy [5, 6].

11
Figure 13. Unidirectional Ring Network (Left), Network Complexity Hierarchy (Right)
First, we will introduce the following additional notations and terminologies:
A simple network of “Unidirectional Ring” 𝐺 is defined to be 𝐺 = (𝑉, 𝐸), where 𝑉 =
{1, 2, … , 𝑚} for some integer 𝑚 > 1 and 𝐸 = {(1, 2), (2, 3), … , (𝑚 − 1, 𝑚), (𝑚, 1)} with 𝑚 nodes
[6].
Let 𝑆 be a transmission for 𝑀𝑅𝑁𝑆 = (𝐺, 𝑀), where 𝐺 is a Unidirectional Ring with 𝑚 nodes and
𝑀 is a set of 𝑛 messages. We use 𝑓(𝑀𝑖, 𝑆) to denote the finishing time of 𝑀𝑖 in 𝑆. For example,
recall the configuration 𝑞 = (𝑢, 𝑣, 𝑀𝑖, 𝑡1, 𝑡2), we have 𝑡2 as the finishing time. In this
case,𝑓(𝑀𝑖, 𝑆) = max(𝑢,𝑣,𝑀𝑖,𝑡1,𝑡2)∈𝑆𝑡2, the finding time of the message 𝑀𝑖. The makespan of 𝑆,
denoted by 𝑀𝑆(𝑆) is defined to be max1≤𝑖≤𝑛𝑓(𝑀𝑖, 𝑆)across all messages [6].
In our network example as shown on the left ofFigure 13, we have 𝑉 = {1, 2, 3, 4} and 𝐸 =
{(1, 2), (2, 3), (3, 4), (4, 1)} with 4 nodes.
3.2.1. Preemptive Transmission
3.2.1.1. Lemma 37.1
Let 𝑆𝑃 be a feasible preemptive transmission for a set of messages with the same origin node,
destination node, release time, and deadline. If the common origin node and destination node of
all the messages are 1 and 𝑒, respectively, then we have 𝑀𝑆(𝑆𝑝) ≥ ∑ 𝑙𝑖
𝑛
𝑖=1 + (𝑒 −
2)(𝑚𝑎𝑥1≤𝑖≤𝑛 𝑙𝑖) [6].
For a set of messages with identical origin nodes, destination nodes, release times, and deadlines.
Lemma 37.1 gives a lower bound for the makespan of any feasible transmission (both preemptive
or non-preemptive). Before generalization, we will first provide an instance to better visualize
the 𝑀𝑅𝑁𝑆. Consider the following
𝑀𝑖 𝑠𝑖
origin node
𝑒𝑖
destination node
𝑙𝑖
length
𝑟𝑖
release time
𝑑𝑖
deadline
𝑀1 1 4 2 0 14
𝑀2 1 4 3 0 14
𝑀3 1 4 2 0 14
𝑀4 1 4 1 0 14
Table 3. Message Constraints of 𝑀 for the Unidirectional Ring 𝑀𝑅𝑁𝑆
If the lemma is not true, then there is a feasible preemptive transmission 𝑆𝑝 such that 𝑀𝑆(𝑆𝑝) <
∑ 𝑙𝑖 + (𝑒 − 2)(max1≤𝑖≤𝑛 𝑙𝑖)
𝑛
𝑖=1 . Without loss of generality, we may assume that the common
release time of the messages is time 0. Let 𝑀𝑃 be the longest message; i.e., 𝑙𝑝 ≥ 𝑙𝑖 for each 1 ≤

12
𝑖 ≤ 𝑛. For each 2 ≤ 𝑗 ≤ 𝑒, let 𝑡𝑗 be the time instant at which 𝑀𝑃 is completely transmitted to
node 𝑗 on the edge (𝑗 − 1, 𝑗) in 𝑆𝑝.𝑀𝑆(𝑆𝑝) < ∑ 𝑙𝑖 + (𝑒 − 2)(max1≤𝑖≤𝑛 𝑙𝑖)
𝑛
𝑖=1 [6].
In our example, we have 14 < (2 + 3 + 2 + 1) + (2)(3). In the following example, in Figure
14, the red lines represent the time instant that the longest message
𝑀𝑃 = 𝑀2 is completely transferred to node 2 ≤ 𝑗 ≤ 𝑒.
Figure 14. Time Instants 𝑡𝑗 of 𝑆𝑁𝑃 (Top) and 𝑆𝑃 (Bottom)
The top instance in Figure 14is the non-preemptive 𝑆𝑁𝑃with 𝑡𝑗 = 𝑡2 = 5 (the time 𝑀2 completely
transmitted to node 2 on the edge (1, 2)), 𝑡𝑗 = 𝑡3 = 8 (to node 3 on the edge (2, 3)), and 𝑡𝑒 =
𝑡4 = 11 (to node 4 on the edge (3, 4)), the bottom instance is the preemptive 𝑆𝑃 with 𝑡𝑗 = 𝑡2 =
5, 𝑡𝑗 = 𝑡3 = 12, and 𝑡𝑒 = 𝑡4 = 16.
It is obvious that 𝑡𝑒 = 𝑓(𝑀𝑝, 𝑆𝑝). In our non-preemptive instance, the common destination node
𝑒 = 4. So, the time to finish transfer the longest message 𝑀𝑝 at 𝑡4 is also the completion time of
𝑀𝑝on the last communication link (edge) [6].Next, we will categorize the messages into two
categories based on the following definition.
For each 1 ≤ 𝑖 ≤ 𝑛 and 2 ≤ 𝑗 ≤ 𝑒, let 𝑋(𝑖, 𝑗) and 𝑌(𝑖, 𝑗) be the total amounts of 𝑀𝑖 transmitted
on the edge (𝑗 − 1, 𝑗) in 𝑆𝑝 in the time intervals [0, 𝑡𝑗] and [𝑡𝑗, 𝑀𝑆(𝑆𝑝)], respectively.
This yields the two message categories 𝑋(𝑖, 𝑗) and 𝑌(𝑖, 𝑗) for the non-preemptive transmission𝑆𝑁𝑃
shown on the left of Figure 15 and for the preemptive transmission 𝑆𝑃 shown on the right. We
have 𝑋(𝑖, 𝑗) + 𝑌(𝑖, 𝑗) = 𝑙𝑖 for each 1 ≤ 𝑖 ≤ 𝑛 and 2 ≤ 𝑗 ≤ 𝑒 since each message is either
transmitted before or after 𝑡𝑗.
Figure 15. 𝑋(𝑖, 𝑗) and 𝑌(𝑖, 𝑗) of 𝑆𝑁𝑃 (Left) and 𝑆𝑃 (Right)
Next, we will categorize the messages into 3 sets. Each set has a distinct characteristic which will
play in a role in determining the makespan 𝑀𝑆(𝑆𝑝) of our transmission. We will define set
𝐴, 𝐵, 𝐶 as follows [6]:
𝑨 = {𝑀𝑖 | 𝑀𝑖is completely transmitted to node 2 before𝑡2in𝑆𝑝}

13
It is clear that the messages in set 𝐴 accounts for the time (length) exclusively between node 1
and node 2 on link (1, 2) that finishes arriving at node 2 before 𝑀𝑝 (the longest message) does.
𝑩 = {
𝑀𝑖 | 𝑀𝑖is completely transmitted to node 2 after𝑡2in𝑆𝑝, and
𝑀𝑖is completely transmitted to node 𝑘 before𝑡𝑘for some node 𝑘, 3 ≤ 𝑘 ≤ 𝑒
}
The messages in set 𝐵 is the ones that arrives after 𝑀𝑝 to node 2 but later actually beats (finishes
arriving earlier than) 𝑀𝑝 on a later node (potentially any node from 3 to the destination node 𝑒).
𝑪 = {𝑀𝑖 | 𝑀𝑖is completely transmitted to node 𝑗 after𝑡𝑗in𝑆𝑝for each 2 ≤ 𝑗 ≤ 𝑒}
The messages in set 𝐶 are the ones that arrives after 𝑀𝑝 does to any node ranging from node 2 to
the destination node e. The total message length in all these 3 cases in fact, sum up to the entire
span (makespan) of our preemptive transmission 𝑆𝑝.
Figure 16. Sets 𝐴, 𝐵, 𝐶 of 𝑆𝑁𝑃 (Top) and 𝑆𝑃 (Bottom)
By our definitions, the sets 𝐴, 𝐵, 𝐶, and {𝑀𝑝} are pairwise disjoint, and the set 𝐴 ∪ 𝐵 ∪ 𝐶 ∪ {𝑀𝑝}
contains all the messages. Note that 𝑀𝑝 itself is not included in any of the sets 𝐴, 𝐵, 𝐶.With our
𝑆𝑁𝑃, the sets for our instance is the follows with 𝐵 being the empty set containing no messages.
𝐴 = {𝑀1}, 𝐵 = {∅}, 𝐶 = {𝑀3, 𝑀4}. On the other hand, with our 𝑆𝑃, the sets for our instance
is the follows with 𝐵 being the empty set containing no messages. 𝐴 = {𝑀1}, 𝐵 = {𝑀3},
𝐶 = {𝑀4}.
We now compute a lower bound for 𝑓(𝑀𝑝, 𝑆𝑝) (or equivalently, 𝑡𝑒). This is the shortest/least
possible time for the longest message 𝑀𝑝 to finish transfer to the destination node. We will
examine how these 3 sets of messages 𝐴, 𝐵, 𝐶 form the entire makespan. So far, we have three
types of messages: [5, 6]
 Each 𝑀𝑖 ∈ 𝐴 is completely transmitted to node 2 before 𝑀𝑝 does, and hence it delays the
finishing time of 𝑀𝑝 by at least 𝑙𝑖 amount.
 For each 𝑀𝑖 ∈ 𝐵, there is a node 𝑘, 2 ≤ 𝑘 < 𝑒, such that 𝑀𝑖 is completely transmitted to
node 𝑘 after 𝑀𝑝 does, but it is completely transmitted to node 𝑘 + 1 before 𝑀𝑝 does.
Thus, it again delays the finishing time of 𝑀𝑝 by at least 𝑙𝑖 amount.
 Each 𝑀𝑖 ∈ 𝐶 is completely transmitted to node 𝑗, 2 ≤ 𝑗 < 𝑒, after 𝑀𝑝 does. However, the
total amount transmitted on the edge (𝑒 − 1, 𝑒) by time 𝑡𝑒 is 𝑋(𝑖, 𝑒), and hence it delays
the finishing time of 𝑀𝑝 by at least 𝑋(𝑖, 𝑒) amount of time.
Recall that every message 𝑀1≤𝑖≤𝑛 is also divided into 2 groups, 𝑋(𝑖, 𝑗) and 𝑌(𝑖, 𝑗)representing
their total amounts being transmitted on the edge (𝑗 − 1, 𝑗) for all edges 2 ≤ 𝑗 ≤ 𝑒 in 𝑆𝑝. 𝑋(𝑖, 𝑗)

14
is the portion between time intervals [0, 𝑡𝑗] and 𝑌(𝑖, 𝑗) between [𝑡𝑗, 𝑀𝑆(𝑆𝑝)]. In other words,
𝑋(𝑖, 𝑗) and 𝑌(𝑖, 𝑗) arrives before and after 𝑀𝑝 does respectively.
The properties above for 𝑀𝑖 ∈ 𝐴 and 𝑀𝑖 ∈ 𝐵 are self-explanatory. For 𝑀𝑖 ∈ 𝐶, we this property
so the entire set 𝐶 consists of 2 parts of ∑ 𝑋(𝑖, 𝑗)
𝑀𝑖∈𝐶 ∪ ∑ 𝑌(𝑖, 𝑗)
𝑀𝑖∈𝐶 . Moreover, it is specifically
used to account for the case of 𝑋(𝑖, 𝑒) when 𝑀𝑝 is preemptively chopped in the last edge (𝑒 −
1, 𝑒). Hence, there exists some message 𝑀𝑖 ∈ 𝐶 that has a portion being transferred before 𝑀𝑝
finishes at 𝑡𝑒. This is the last portion to be included in 𝑓(𝑀𝑝, 𝑆𝑝) as shown in By referring to
Figure 30, we see that our 𝐵 = {∅}. In our example 1, 𝑋(1, 4) = 2, 𝑋(2, 4) = 3, 𝑋(3, 4) =
0, 𝑋(3, 4) = 0 and only 𝑀3 and 𝑀4 are in 𝐶 (since by definition, they finished arriving after time
𝑡𝑒). Therefore, ∑ 𝑋(𝑖, 4)
𝑀𝑖∈𝐶 = 0.
3.2.1.1.1. Direct Proof
Putting it altogether, we are able to directly proof the Lemma 37.1. We will also show it works
for both our 𝑆𝑁𝑃 and 𝑆𝑁 instances. From the above we see the finishing time of 𝑀𝑝 is essentially
𝑡𝑒, which is greater than or equals the sum of the following three parts:
1. Total span (delay) of the messages in set 𝐴 (denoted by ∑ 𝑙𝑖
𝑀𝑖∈𝐴 )
2. Total time to send 𝑀𝑝 through number of edges (from node 1 to 𝑒) denoted by (𝑒 − 1)𝑙𝑝
3. Total span (delay) of the messages in set 𝐵(denoted by ∑ 𝑙𝑖
𝑀𝑖∈𝐵 )
4. Total time of the messages in set 𝐶 that has part of it transferred before 𝑀𝑝 finishes at 𝑡𝑒
on edge (𝑒 − 1, 𝑒).
𝑓(𝑀𝑝, 𝑆𝑝) = 𝑡𝑒 ≥ ∑ 𝑙𝑖
𝑀𝑖∈𝐴
+ (𝑒 − 1)𝑙𝑝 + ∑ 𝑙𝑖
𝑀𝑖∈𝐵
+ ∑ 𝑋(𝑖, 𝑒)
𝑀𝑖∈𝐶
Now, we know the makespan 𝑀𝑆(𝑆𝑝) has to be the finishing time of 𝑀𝑝 (𝑓(𝑀𝑝, 𝑆𝑝)) plus the
time to finish the rest of the messages that complete after 𝑡𝑒 (∑ 𝑌(𝑖, 𝑒)
𝑀𝑖∈𝐶 ). So, we have:
𝑀𝑆(𝑆𝑝) = 𝑓(𝑀𝑝, 𝑆𝑝) + ∑ 𝑌(𝑖, 𝑒)
𝑀𝑖∈𝐶
Substituting 𝑓(𝑀𝑝, 𝑆𝑝) with ∑ 𝑙𝑖
𝑀𝑖∈𝐴 + (𝑒 − 1)𝑙𝑝 + ∑ 𝑙𝑖
𝑀𝑖∈𝐵 + ∑ 𝑋(𝑖, 𝑒)
𝑀𝑖∈𝐶 ,
≥ ( ∑ 𝑙𝑖
𝑀𝑖∈𝐴
+ (𝑒 − 1)𝑙𝑝 + ∑ 𝑙𝑖
𝑀𝑖∈𝐵
+ ∑ 𝑋(𝑖, 𝑒)
𝑀𝑖∈𝐶
) + ∑ 𝑌(𝑖, 𝑒)
𝑀𝑖∈𝐶
Combining the two parts of set 𝐶 (parts partially transferred before 𝑡𝑒 during preemption and the
parts after 𝑡𝑒) since ∑ 𝑋(𝑖, 𝑒)
𝑀𝑖∈𝐶 + ∑ 𝑌(𝑖, 𝑒)
𝑀𝑖∈𝐶 = ∑ 𝑙𝑖
𝑀𝑖∈𝐶 . (Recall 𝑋(𝑖, 𝑗) + 𝑌(𝑖, 𝑗) = 𝑙𝑖)
= ∑ 𝑙𝑖
𝑀𝑖∈𝐴
+ (𝑒 − 1)𝑙𝑝 + ∑ 𝑙𝑖
𝑀𝑖∈𝐵
+ ∑ 𝑙𝑖
𝑀𝑖∈𝐶
Factor out a 𝑙𝑝 from (𝑒 − 1)𝑙𝑝:
= ∑ 𝑙𝑖
𝑀𝑖∈𝐴
+ (𝑒 − 2)𝑙𝑝 + 𝑙𝑝 + ∑ 𝑙𝑖
𝑀𝑖∈𝐵
+ ∑ 𝑙𝑖
𝑀𝑖∈𝐶
(1)
(2)
(3)
(4)
(5)

15
By definition, the pairwise disjoint sets of 𝐴, 𝐵, 𝐶 and 𝑀𝑝 form the set of all messages (
𝐴 ∪ 𝐵 ∪ 𝐶 ∪ {𝑀𝑝} = ∑ 𝑙𝑖
𝑛
𝑖=1 ). Combining𝐴, 𝐵, 𝐶,
= ∑ 𝑙𝑖
𝑛
𝑖=1
+ (𝑒 − 2)𝑙𝑝
We know that the message of the longest length can be expressed as 𝑀𝑃 = 𝑙𝑝which is the
maximum of length 𝑙𝑖 ofall message1 ≤ 𝑖 ≤ 𝑛
= ∑ 𝑙𝑖
𝑛
𝑖=1
+ (𝑒 − 2) (
𝑛
max
𝑖 = 1
𝑙𝑖)
Finally, we obtain what we proposed in the Lemma. ∎ [6]
Figure17 illustrates the non-preemptive instance𝑆𝑁𝑃 on the left and the preemptive instance 𝑆𝑃on
the right. Both visually demonstrate each section of 𝐴, 𝐵, 𝐶 in the above proof.
Figure 17. Visual Representation of Proof for 𝑆𝑁𝑃 (Left) and 𝑆𝑃 (Right)
3.2.2. Non-Preemptive Transmission
We have previously proved Lemma 37.1 for a particular case of bothnon-preemptive and
preemptive transmission. In fact, all non-preemptive transmissions are literally preemptive
transmissions. It is just a special case by choosing not to utilize the preemptive functionality.
Non-preemptive transmissions are a subset of preemptive transmissions. The following
complexity hierarchy in Figure 18 shows the relationship between preemptive/non-preemptive.
Figure 18. Non-Preemptive/Preemptive Complexity Hierarchy
This is just the rule of the general computational complexity. We have preemptive transmissions
as the more general case with fewer constraints (such that each transmission must not finish
transferring completely before it can be transferred elsewhere on another communication link
(6)
(7)

16
(link)). Intuitively, we claim that: If the more general case is easy, then the special case is easy;
if the more special case is hard, then the general case is hard. This brings our next lemma:
3.2.2.1. Lemma 37.2
Let 𝑆𝑁𝑃 be a feasible non-preemptive transmission for a set of messages with the same origin
node, destination node, release time, and deadline. If the common origin node and destination
node of all the messages are 1 and 𝑒, respectively, then we have 𝑀𝑆(𝑆𝑁𝑃) ≥ ∑ 𝑙𝑖 +
𝑛
𝑖=1
(𝑒 − 2)(𝑚𝑎𝑥1≤𝑖≤𝑛 𝑙𝑖). [6]
This is obvious and the formal proof is as follows:
3.2.2.1.1. Direct Proof
Let 𝑆1 be a feasible non-preemptive transmission for the set of messages such that it has the
minimum makespan among all feasible non-preemptive transmissions. Let 𝑆2 be a feasible
preemptive transmission for the set of messages such that it has the minimum makespan among
all feasible preemptive transmissions [6].
Since 𝑀𝑆(𝑆𝑁𝑃) ≥ 𝑀𝑆(𝑆1) ≥ 𝑀𝑆(𝑆2), the lemma follows immediately from Lemma 37.1.
Figure 19. Visualization of Proof for Lemma 37.2
We can clearly see in Figure 19that by constructing 𝑆1 and 𝑆2, we intermediately show that the
non-preemptive transmission (our claim 𝑆𝑁𝑃) is at least as efficient (at least as long in total
transmission length makespan 𝑀𝑆(𝑆𝑁𝑃)) as the non-preemptive transmission 𝑆1 (its makespan
𝑀𝑆(𝑆1)) which is at least as efficient as those of the preemptive transmission 𝑆2 (its makespan
𝑀𝑆(𝑆2)) [6]. Therefore, we have showed:
𝑀𝑆(𝑆𝑁𝑃) ≥ 𝑀𝑆(𝑆1) ≥ 𝑀𝑆(𝑆2) = 𝑀𝑆(𝑆𝑃) ≥ ∑ 𝑙𝑖 + (𝑒 − 2)(max1≤𝑖≤𝑛 𝑙𝑖)
𝑛
𝑖=1
Which holds true for both preemptive and non-preemptive cases 𝑀𝑆(𝑆𝑃) ≥ ∑ 𝑙𝑖 +
𝑛
𝑖=1
(𝑒 − 2)(max1≤𝑖≤𝑛 𝑙𝑖) and 𝑀𝑆(𝑆𝑁𝑃) ≥ ∑ 𝑙𝑖 + (𝑒 − 2)(max1≤𝑖≤𝑛 𝑙𝑖)
𝑛
𝑖=1 respectively for our
Lemma 37.1 and Lemma 37.2 [6].
4 . KINESTHETIC LEARNING ACTIVITY MODEL
Pereducation theory, one of the most challenging aspect lays in the theory of learning. Learning
theories are the conceptual frameworks in which knowledge is absorbed. The primary learning
technique in most academic disciplines is the traditional learning that has been followed for ages.
This typical learning technique places students in a passive rather than an active role. Traditional
learning encourages one way communication and therefore, the lecture must make a conscious
effort to become aware of the students’ understanding without verbal feedback. Furthermore,
traditional learning requires a considerable amount of time outside of the classroom to enable a
long-term retention of the content [8].

17
To overcome the shortcomings of traditional learning and to improve the performance of all types
of students, we present a technique called Kinesthetic Learning Activity (KLA). This interactive
approach was first presented at California State Polytechnic University, Pomona(Cal Poly
Pomona) [8]. Throughout the years of KLA implementation, research has shown substantial
positive feedback from students as well as its effectiveness proven by the quizzes conducted
directly after. Studies also show that students prefer a “hands on” or “learning by doing”
approach to build understandings [8]. Below is the our proposed KLA specifically designed for
our Real-Time Network Message Routing research.
4.1. Overview of the Proposed KLA Approach
In our KLA, we want the students to actively engage in the material directly following our
presentation to provide best learning outcome. While the students may or may not comprehend
the concepts of message routing over a 𝑀𝑅𝑁𝑆, by using a “quiz” approach with hands-on, our
KLA will stimulate the student’s creative thinking abilities. We have developed a hands-on
crafting activity using KitKat chocolate bars to simulate what the message routing configuration
would be like under various preemptive/non-preemptive models.
The following Figure 20 shows the presentation slide of the KLA introduction on left. On right is
the slide detailing the assigned main activity.
Figure 20. Presentation Slides
4.1.1. Learning Goals
This study will examine the consequences of subjecting kinesthetic learning techniques on
students enrolled in an upper division Computer Science course, CS331 Design and Analysis of
Algorithms, at Cal Poly Pomona for the Summer 2017 academic quarter. Those results will be
compared with the effectiveness of traditional instructional methods. The objective of this study
is to provide evidence that kinesthetic learning activities supplement the learning and
comprehension of course materials. Our hypothesis is that KLAs, when used to complement
traditional classroom instructional methods will increase the amount of material absorbed by the
students. We suggest at the end of this exercise, students will be able to experience the practical
aspect of constructing a feasible transmission (whether it be preemptive or non-preemptive) given
a set of messages with certain constraints. This research study has been reviewed and approved
by the Cal Poly Pomona Institutional Review Board (IRB): protocol#17-116. It is in compliance
with federal and state regulations as well as the Cal Poly Pomona policies regarding the
protection of the human subjects used in the research.

18
4.2. KLA Instructions and Topic Information
4.2.1. KLA Execution Instructions
Students were given one pre-test to gauge the prior knowledge. The class will then be divided in
halves. Each half will alternately be taught using the traditional method and the kinesthetic
method over two class sessions by a single instructor. Following each lecture, the class was given
a post-test to evaluate the level of material comprehend. Students were also asked to complete a
survey afterin effort to gather opinions on the effectiveness of the KLA.
Figure 21.
During the KLA, each student was supplied a pack of 1.5-Ounce Kit Kat Bars (Milk Chocolate,
but flavors may vary). Each pack consists of 4 finger bars. In addition, each participant would
receive their own plastic knife, and edge/time table worksheet. The activity provides students a
𝑀𝑅𝑁𝑆 with a diagram of an arbitrary directed graph 𝐺 and a set of 4 messages 𝑀 =
{𝑀1, 𝑀2, 𝑀3, 𝑀4}. Students were instructed to construct a feasible non-preemptive transmission
satisfying the network and message constraints table. They were to simulate each message-span
with the provided KitKat bars by crafting them into the appropriate length. Moreover, they were
to place the crafted message/KitKat over the correct edge/time slot on the worksheet. Students
have 15 minutes to complete the exercise. To add competitiveness, we added an additional
constraint that disqualifies the students if they over-chop the message length into segments. In
other words, we do not allow “gluing” pieces together. Figure 21 shows the instructions slide on
the left. On the right is the solution slide.
4.3. Evaluation of KLA Results and Analysis
The pre/post-test consists of 2 questions where Question 1 has part a (total 12 points) and b (total
3 points). The test provides a feasible 𝑀𝑅𝑁𝑆 transmission and asks the students to fill in the
blanks with 4 given available options. Question 2 is True or False (total 5 points). The testhas a
total possible score of 20 points shown as follows:
4.3.1. Pre/Post Evaluation Test
Consider the above feasible 𝑀𝑅𝑁𝑆 Routing Configuration.
1a. Fill in each blue blank with one of the following choices (𝑀1, 𝑀2, 𝑀3, 𝑀4, or
N/A) (2 points per blue blank)

19
1b. Fill in each green blank with one of the following choices (𝑀1, 𝑀2, 𝑀3, 𝑀4, or
N/A) (1 point each green blank)
2. True/False (1 Point Each)
a) By changing 𝑀4’s deadline to 5 will cause this 𝑀𝑅𝑁𝑆 to be non-feasible
b) By changing 𝑀2’s release time to 3 will cause this 𝑀𝑅𝑁𝑆 to be non-feasible
c) By changing 𝑀1’s origin node to 2, this 𝑀𝑅𝑁𝑆 will remain feasible
d) By changing 𝑀1’s length to 3 and deadline to 4, this 𝑀𝑅𝑁𝑆 will remain
feasible
e) This 𝑀𝑅𝑁𝑆 is still feasible if the transmission becomes non-preemptive
The solutions of Question 1a & 1b is shown in Figure 22 and the solution of Question 2 is a) False b) False
c) False d) False e) False.
Figure 22. Solutions to Pre/Post-Test Question 1a & 1b.
4.4. KLA Experimental Results
The results of our KLA turned out well. Students all had a positive response to the activity. The
scores from the tests are anonymously recorded in Figure 23showingthe pre-test and post-test
scores, the mean, median, max, min points and percentages, along with the total number points
possible. Overall, although the KLA half of the class had a lower score throughout, they
improved more significantly than the traditional teaching half.
Figure 23. KLA Pre/Post-test Results
?
(3, 4) M4 ?
(4, 5) ?
? ?
?
M1
Edges Time 1 2 3 4 5 6
(1, 2) ? ?
(1, 3) M1
(2, 4) ? M3
M1
M3
M1
M3
M2
M2
M4 M4
(1, 3)
(2, 4)
(3, 4)
(4, 5)
Edges Time 1 2 3 4 5 6
(1, 2) M3

20
Figure 24. % of Average Test Scores & Learning Effectiveness
Figure 24shows the difference in the pre-test and post-test results and separates the traditional vs
KLA groups. The improvement in the average score for the group that learned via the traditional
method was 15.88%, while the KLA group saw their average score jump by 21.67%.
4.4.1. KLA Survey Feedback
At the end of our KLA trial, the total of 29 students completed the survey consisting the
following questions:
 Question 1: My overall evaluation of the kinesthetic learning activity (KLA) is (circle
one): Excellent, Good, Fair, and Poor.
 Question 2: Rate the following questions on a scale of Strongly Agree to Strongly
Disagree:
a) I enjoyed the KLA experience.
b) The kinesthetic activities were easy to understand.
c) The kinesthetic activities were easy to follow.
d) KLA is effective in helping me learn concepts.
e) KLA helped me be more engaged in learning process.
f) I would recommend KLA as an alternative way of teaching
The responses from Question 2 is shown in the bottom of Figure 25 indicating most the students
had a favorable experience with the KLAs. There were very few responses of Disagree or
Strongly Disagree. This result is consistent with Question 1. At last, all students participated and
enjoyed eating the KitKats. It was a good attempt considering it was the first KLA involving
food. We have also discovered many improvements that we can make to better the learning
outcome.
Selection # of respondents
Excellent 7
Good 13
Fair 9
Poor 0

21
Figure 25. Results of Survey Question 1 (Top) Question 2 (Bottom)
Figure 26. Students Executing KLA in Action
5. CONCLUSIONS AND FUTURE WORK
In our survey of routing real-time messages, we have considered the problem of how to construct
a feasible transmission on a given 𝑀𝑅𝑁𝑆 if one exists. In the case where the 𝑀𝑅𝑁𝑆 does not
have a feasible transmission, we identified the specific constraints that caused it to miss the
deadline. In addition, we analyzed and proved the computational complexity of an Arbitrary
Directed Graph network to be NP-complete even when all four parameters are fixed. We then
considered a simpler network which is a special case of the Arbitrary Directed Graph called a
Unidirectional Ring. The same complexity results hold for preemptive transmission.
Lastly, to better demonstrate this 𝑀𝑅𝑁𝑆, we proposed a Kinesthetic Learning Approach way to
model the scheduling with KitKat chocolate bars. This KLA method promotes student
interactivity and improves student learning by engaging a fundamental and easily conceived
learning style. As previous research suggest, courses on analysis of algorithms are ideally suited
for KLAs due to the complexity of the subject matter and the ability to adapt algorithm steps into
physical actions. Our KLA results a positive feedback from the students. In addition, our case
study shows an improvement in pre-test/post-test scores to support its effectiveness. For future
work, we plan to expand our research to more offline routing variations as well as online routing
of variable-length messages with more variations in network complexity such as out-tree, in-tree,
bidirectional tree, and bidirectional ring. We also plan to improve our KLA model by providing
better instructions through video examples for better participating results. We will compare the
experimental results to quantitatively analysis on its effectiveness.

22
REFERENCES
[1] R. Koch, B. Peis, M. Skutella and A. Wiese, "Real-Time Message Routing and Scheduling", TU
Berlin, Institut f¨ur Mathematik, Straße des 17. Juni 136, 10623 Berlin, Germany,, 2017.
[2] U. Shaw and B. Sharma, "A Survey Paper on Voice over Internet Protocol (VOIP)", International
Journal of Computer Applications, vol. 139, no. 2, pp. 16-22, 2016.
[3] D. Hemmendinger, "The ACM and IEEE-CS guidelines for undergraduate CS education",
Communications of the ACM, vol. 50, no. 5, p. 46, 2007.
[4] P. Mueller, "New Algorithm Significantly Boosts Routing Efficiency of Networks", Phys.org, 2008.
[Online]. Available: https://phys.org/news/2008-08-algorithm-significantly-boosts-routing-
efficiency.html. [Accessed: 30- May- 2017].
[5] J. Leung, T. Tam and G. Young, "On-Line Routing of Real-Time Messages", Journal of Parallel and
Distributed Computing, vol. 34, no. 2, pp. 211-217, 1996.
[6] J. Leung, Handbook of Scheduling, 1st ed. London: Chapman & Hall/CRC, 2004, pp. 795-826.
[7] S. Joosten, "Relaxations of the 3-partition problem", Department of Applied Mathematics, University
of Twente, Enschede, The Netherlands, 2011.
[8] Alraddady, Sara, Danny Luong, and Gilbert Young. 2014. “A Study of Kinesthetic Learning
Activities Effectiveness in Teaching Computer Algorithms.” The 2014 International Conference on
Frontiers in Education: Computer Science & Computer Engineering. 400-406
AUTHORS
Paul is a Graduate Student of Computer Science at Cal Poly Pomona. His research
interests include assistive technology for the disabled, software accessibility testing, web
and mobile applications, and CS Education, etc. He will be pursuing his Ph.D. in CS and
wishes to one day become a CS Professor.
Dr. Sun Assistant Professor of Computer Science at Cal Poly Pomona. His research
interests include software engineering, cloud and mobile computing, etc. Sun leads the
research and development for numerous start-up companies. He has a solid background in
both the industry working at Amazon as well as academia working as a Post-Doc research
associate at Vanderbilt University.
Dr. Young Professor of Computer Science at Cal Poly Pomona. His research interests
include parallel and distributed computing, computer networks, algorithm design,
scheduling, combinatorial optimization, etc. Young is the creator of the KLA and has
received the College of Science Distinguished Teaching Award for his innovative and
continuous research on this alternative CS teaching approach.

A SURVEY TO REAL-TIME MESSAGE-ROUTING NETWORK SYSTEM WITH KLA MODELLING

More Related Content

Similar to A SURVEY TO REAL-TIME MESSAGE-ROUTING NETWORK SYSTEM WITH KLA MODELLING (20)

More from ijfcstjournal (20)

Recently uploaded (20)

A SURVEY TO REAL-TIME MESSAGE-ROUTING NETWORK SYSTEM WITH KLA MODELLING