Is the act of dividing a network into smaller logical subnetworks.

3 Pipeline network topology identification

Pipeline network topology is described by adjacent matrix A. A is an n-dimensional matrix, where n is pipe segment number. When the element of row i, column j is equal to 1 in the adjacent matrix, i.e. Aij=1, there is a path from pipe segment i to j; when Aij=0, no path from pipe segment i to j. There are many similar sub-networks in a pipeline network. Typical sub-networks can be roughly classified as series and parallel structures.

3.1 Series sub-network and its identification

Series sub-network is defined as two pipe segments connected one after the other. Series sub-network topology and adjacent matrix are shown in Fig. 2a.

Figure 2. Series and parallel sub-network topology and corresponding adjacent matrixes

It can be seen in Fig. 2a, A23=∑j=14A2j= ∑i=14Ai3=1. It means pipe segment 2 connects to pipe segment 3. Thus, if pipe segment k connects pipe segment l in the adjacent matrix, it can be identified by Eq.(6).

(6)Akl=∑j=1nAkj=∑i=1nAil=1

5.1.3 Packet recovery

In previous approximate designs, some annotation frameworks have been proposed that label sections of the approximable data [5, 45]. We manually annotate benchmarks in a fashion similar to these methods. In the AMNoC, a multiflit data packet is annotated as approximable only when the packet stores words of the same type (integer or floating point) and only if those words are all approximable. At the destination node, packet recovery is implemented to approximate the missing data in an approximable data packet. Two stages are required to recover a data packet: (1) determine the missing flits of a data packet and (2) approximate the values in the missing flits.

In the approx-subnet, flits of an approximable data packet are injected (or dropped) within a single-cycle interval. Due to the ASW design, these flits are transmitted in succession. Therefore, at the destination node, flits from the same approximable packet can be easily gathered together. The location of a flit in the packet is also stored in its header bits and transmitted along with the flit. After an approximable packet is transmitted,f the missing flits can be determined based on the locations of the received flits.

Previous studies have proposed many value approximation designs, such as last value, stride, FCM [37], and VTAGE [38]. Reducing the complexity and enhancing the accuracy are the two main challenges of value approximation. In this context, the above-mentioned methods are either insufficiently accurate [37] or excessively complex, thereby incurring a high power consumption and a high overhead [38]. We choose linear interpolation due to its low complexity and reasonable accuracy. Linear interpolation requires only one addition to perform value approximation. The data in an approximable packet are generally fetched from successive memory blocks; hence, the data in a packet (e.g., the adjacent pixels in an image) are considerably similar. The data of dropped flits are most relevant to the preceding and following flits. Therefore, the linear interpolation error is small. If the dropped flit is the first flit or last flit, its approximation is to copy the received flit that is the closest to it.

3.6 Packet routing

Once flits are buffered in AS buffer (Path 3) or BS buffer (Path 1), they should go through pipeline stages of the current router. The proper output port for a packet is determined based on a minimal routing scheme, so a packet is only allowed to take the ports along one of the shortest paths toward the destination (at most two directions and four n/2-bit ports). The algorithm checks the related free ports and allocates one of them to the packet. Also, in particular, our algorithm considers whether there exists a free VAL that begins from the current router and ends at some router along the path toward the packet's destination. If there is such a VAL, it will be prioritized over the BS ports and used to shorten packet's journey. If more than one VAL is found, the longer VAL is considered first. This involves maintaining the end point address of each VAL in the VAL's source router to use for making decision. Thus, as mentioned before, packets in our network may switch between the AS and BS sub-networks several times to reach their destinations.

The network guarantees deadlock freedom by employing the well-known escape channel concept [45]. One of the virtual channels of the BS ports is used as an escape channel and adopts the dimension-order XY deadlock-free routing algorithm. To guarantee deadlock freedom, once a packet enters an escape channel, it cannot switch again to adaptive VCs and the AS sub-network. If there is a single VC per port, the entire BS sub-network uses the XY algorithm and acts as an escape channel for the AS. For flow control over VALs, we use the methods presented in Ref. [12].

4.2 Method implementation

The core of the proposed model-driven method consists of two model transformations. The first, namely UML-to-EQN, is the model-to-model transformation that generates the EQN-based performance model. Specifically, it maps an input UML Activity Diagram (AD) to an output EQN model. The second, namely EQN-to-jEQN, is the model-to-text transformation that takes as input the EQN model and yields as output the jEQN code implementing the distributed simulation. The next two sections give a detailed view of such model transformations.

4.2.1 Generation of the EQN performance model

The UML-to-EQN model transformation has been specified by use of QVT [24] and has been implemented and executed by use of the Eclipse MMT plugin [32].

QVT prescribes that both the source and the target models used in a transformation must be instances of a MOF compliant metamodel [23]. To this respect, the target EQN model is an instance of the EQN metamodel in Figure 14.3.

The service-oriented system has been designed as an orchestration of services implemented by use of the Web Services technology. It is assumed that the related AD includes a detailed view of the party that acts as the coordinator of the service composition (the orchestrator, in SOA terminology), while other parties, which act as black-box entities providing a single service, are represented by swimlanes including an activity for each provided operation.

In this perspective, a service invocation is represented in the UML model by an activity edge that connects a pair of activity nodes, the first one belonging to the swimlane associated to the orchestrator, and the second one belonging to the swimlane associated to the service provider.

In order to give the rationale of the model transformation process, Figure 14.4 outlines the mapping of the UML fragment representing a service invocation into the corresponding portion of the EQN model.

Figure 14.4. EQN mapping of the UML-based service invocation pattern.

The EQN model includes a subnetwork associated with each participant (e.g., the orchestrator and the set of service providers). Each subnetwork can be simulated by different simulation components of the distributed simulation infrastructure, as detailed later on. Two classes of jobs are introduced: the first one, named “toServe”, is used to represent jobs that have to be served by a participant; the other one, named “Served”, is used to model a job just served by a participant. The pattern shown in the figure is related to jobs processed by the orchestrator service center that have to be served by the next service center, according to the orchestration paradigm. The job class is initially set to “toServe”, briefly denoted as “C0”. A request to the next service center is structured as follows:

•

the job passes through the WAN Service Center, to model the request message that the orchestrator sends to the service provider;

•

the job passes through the Provider Service Center, to model the service execution performed by the participant. It should be noted that the router R forwards jobs of class “C0” to the Provider Service Center. Moreover, the access to the Provider Service Center is controlled by Allocate/Release nodes, in order to model the capacity of the server that provides the requested service. Finally, as the job leaves the Release node, its class is changed to “Served”, briefly denoted as “C1”;

•

the job passes through the WAN Service Center, to model the response message that the service provider sends to the orchestrator;

•

the job returns to the Orchestrator Service Center. It should be noted that the router R forwards all jobs belonging to class “C1” to the “Set C0” node and then to the Orchestrator Service Center.

The parameterization of the EQN model has been carried out by implementing an algorithm based on the one specified in [33], which takes into account the MARTE annotations included in the input UML model.

A complete outline of the mapping rules between UML elements and EQN elements is provided by Table 14.1.

Table 14.1. Mapping of UML Elements to EQN Elements

UML Element	EQN Element
Swimlane	Subnetwork
MARTE annotation	Users and think time parameters
(associated to swimlane)	(for closed EQN)
MARTE annotation	Distribution of interarrival time
(associated to swimlane)	(for open EQN)
Start/Final Node	Terminal node (for closed EQN) Source/Sink node (for open EQN)
Opaque Action Node	request to Orchestrator Service Center
Call Operation Node	see Figure 14.4
Fork Node	Fork Node
Join Node	Join Node
Decision Node Decision Node	Router Node
Control Flow	used for defining the routing within the EQN

The next section describes some implementation issues regarding the EQN-to-jEQN model- to-text transformation.

2.1 Model building

By dividing the main network into sub-networks, one can determine the parameters step by step. The isomerization and hydrogenation sub-networks are shown in figure 2a and 2b.

The mass transfer of hydrogen through the gas–liquid interface is modeled as:

N˙H2=keff,H2(CH2L,I-CH2L)

Using the Henry's Law with the Van't Hoff temperature dependency, the interfacial concentration on the liquid side can be described as follows

CH2L,I=pH2HH 2(T),HH2(T)=H0,H2exp(HadsRT)

The ractionrate for the isomerization and hydrogenation reaction are as follows

r61/62=Ccatkr,61 /62CDoce/IsoDoceL, r5/8=CCatkr,5/8CH2 LCDoce/IsoDoceL

The temperature dependency of the reaction rate coefficients are described as proposed by Buzzi-Ferraris et al. [2009]:

kr,i=kref,iexp(- Ea,iR(1T-1Tref))

Bootstrapping

It is standard practice that new subnetworks must obtain their subnetwork address mask from a central, trusted authority assuring the uniqueness of the address when first connecting to a network. In the Internet, this is the Internet Assigned Numbers Authority (IANA). We propose the following procedure for this:

The new subnetwork or node NX submits its signature verification public key in a secure way (e.g., out-of-band via a trusted messenger) to the IANA.

The IANA builds CertX by signing the certificate content with SKIANA. The IANA ensures that there is only one valid certificate per address NX. All nodes have the PKIANA available (e.g., hard coded or manually installed).

The new node sends CertX inside a special type of LSU having a lifetime roughly equal to the lifetime of the distributed certificate. This is a part of the normal “flooding” procedure as implemented today in link-state routing protocols to distribute information among nodes.9

Each recipient node verifies the new node’s signature verification public key by validating the certificate with PKIANA. It then stores the key and the certificate.10 The LSUs carrying certificates may have a higher priority in the flooding process and need not be authenticated by the previously described process because of the certificate verification’s inherent authentication.

IV.C. Router Networks

Informally, a router network is a collective of subnetworks that may be owned by different entities (thus, management delineation may be problematic) and connected over some medium that may be owned by yet another entity.

When several subnetworks need to be joined, the connection may be made using a switch or a router. Switches may only be used to connect subnets that use similar protocols (they operate at layer two of the OSI model), whereas routers may connect subnetworks that use different LAN technologies (they operate at layer three of the OSI model).

Switching technology was considered to be both complex and wasteful of resources. Lately, with the introduction of modern switching hubs in the LAN, switching has been shown to be faster, simpler, and cheaper than routing. Switching does not involve the disassembly and reassembly associated with internet protocol (IP) routing, instead it switches packets at very high speed. Unfortunately, switches are unable to process frames and control their distribution, leading to problems with scalability and management. They also cannot recognize layer three protocols such as Appletalk, IPX, and DECnet.

The versatility of a router makes it an attractive solution; however, with careful design a network may be constructed using a combination of switches (for subnet segments) and reserving routers for points where they are strictly necessary.

IP Subnet addressing

IP networks can be divided into smaller networks called subnetworks (or subnets). Sub-netting provides the network administrator with several benefits, including extra flexibility, more efficient use of network addresses and the capability to contain broadcast traffic. For the purpose of network management, an IP address is divided into two logical parts, the network prefix and the host identifier. All hosts on a subnetwork have the same network prefix. This routing prefix occupies the most-significant bits of the address. The number of bits allocated within a network to the internal routing prefix may vary between subnets, depending on the network architecture. The host part is a unique local identification and is either a host number on the local network or an interface identifier.

This logical addressing structure permits the selective routing of IP packets across multiple networks via special gateway computers, called routers, to a destination host if the network prefixes of origination and destination hosts differ, or sent directly to a target host on the local network if they are the same. Routers constitute logical or physical borders between the subnets, and manage traffic between them. Each subnet is served by a designated default router, but may consist internally of multiple physical Ethernet segments interconnected by network switches.

The routing prefix of an address is written in a form identical to that of the address itself. This is called the network mask, or netmask, of the address. For example, a specification of the most-significant 18 bits of an IPv4 address, 11111111.11111111.11000000.00000000, is written as 255.255.192.0. If this mask designates a subnet within a larger network, it is also called the subnet mask. This form of denoting the network mask, however, is only used for IPv4 networks.

The governing bodies that administer Internet Protocol have reserved certain networks for internal uses. In general, intranets utilizing these networks gain more control over managing their IP configuration and Internet access. A subnet allows the flow of network traffic between hosts to be segregated based on a network configuration. By organizing hosts into logical groups, subnetting can improve network security and performance. Sub- netting works by applying the concept of extended network addresses to individual computer (and other network device) addresses.

An extended network address includes both a network address and additional bits that represent the subnet number. Together, these two data elements support a two-level addressing scheme recognized by standard implementations of IP. The network address and subnet number, when combined with the host address, therefore support a three-level scheme.

A subnet address is created by borrowing bits from the host field and designating them as the Subnet field. The number of borrowed bits varies and is specified by the subnet mask. A given network address can be broken up into many subnetworks. For example, 192.168.1.0, 192.168.2.0, 192.168.3.0, and 192.168.4.0 are all subnets within a network 192.168.0.0. All 0s in the host portion of an address specifies the entire network.

Subnet masks use the same format and representation technique as IP addresses. The subnet mask, however, has binary 1s in all bits specifying the Network and Subnetwork fields, and binary 0s in all bits specifying the Host field.

For all 0s in the binary representation above it is possible to have that many hosts, where each 0 position goes to the power of two, increased by one as we go from right to left (left is most insignificant bit).

For example, in CCTV a typical default number of bits is 8, as shown in the chart for Class B Subnet- ting on the left. This gives 254 possible hosts addresses, a total of 256 combinations, less one for the network address and less one for the broadcast address = 254.

This is sufficient in most of the small IP CCTV LANs and hence rarely changed from the default setting.

Sub-networks Identification

A designed HEN can be divided into a number of independent sub-networks based on the structure of the HEN. A sub-network can be defined as a set of process streams and heat exchangers that are integrated to each other to form an independent network. Each sub-network in the considered HEN should be treated separately in the next controllability assessment steps. Fig. 8 shows an example of a HEN consists of three hot streams and four cold streams where the coolers and heaters use the same cooling and heating utility respectively, such as cooling water for coolers and HP steam for heaters. The figure shows three independent sub-networks as follow:

Process streams: H1, H2, C1 and C2 and contains exchangers: 2, 3, 4, 5 and H1.

Process streams: H3 and C3 and contains exchangers: 1, H3 and C1.

Process stream C4 and heated up by exchanger H2.

Network layer security

Network layer security is used to provide secure tunnels between subnetworks. The general outline of a Virtual Private Network (VPN) that connects two networks through a secure tunnel is shown in Figure 10-11. All traffic between networks 1 and 2 is protected through network layer encryption. It is also possible to connect a single end system to a subnetwork through a VPN (as is commonly done by business travelers who need to connect to their corporate networks).

Network layer security is achieved through the IPSec protocol. Figure 10-10 shows the header fields used for data transfers in IPSec. The packet headers include a new IP header that is used by the tunnel. The IP source and destination addresses correspond to the external interfaces of the VPN gateways that are the end points of the tunnel. The ESP header contains information about the key material used for encryption and the sequence number of the packet. The trailer contains padding (to ensure alignment to a multiple of the block size of the cipher) and a next protocol field to ensure correct handling of the embedded packet. There is also an authentication trailer that verifies the source of the tunneled packet.

In IPSec, the encapsulated packet remains encrypted along the path between the tunnel end points. Within networks connected by the tunnel, the packet is sent in cleartext.