The COSMIC ISO 19761 method is used in software industry as a contributor to estimation improvements and for comparability across projects. The COSMIC method is based on the identification of data movements but it does not consider the data manipulations. To take into account data manipulations associated with data movements at a detailed level of granularity, the measurement of control structures, also referred to as structural size, has been suggested previously. While the previous work focused on the use of constructs (alt, opt, and loop) for one structural level, the multi-level had not been considered. This work proposes refinements to our previous structural size measurement method through the assessment of the nested (multi-level) control structures in the sequence diagram. This refined method proposes detailed measures in different situations where the three constructs can be nested. These measures can be very useful for the software project planning that requires better effort estimation. A web site case study “Digital-Training Center” is used to illustrate and apply the proposed measurement algorithms.
The software development industry as a whole has a disappointing track record when it comes to completing a project on time and within budget. The Standish Group wellknown Chaos Report confirms that only 32% of software development projects are completed successfully within the estimated schedule and budget [21].
Software developers are constantly under pressure to deliver on time and on budget.
As a result, many projects focus on delivering functionality at the expense of meeting
the details of functionality, defined as the different scenarios describing one
functionality [20]. When software details become visible and clients’ demands on
software quality increase, functionality details can no longer be considered of
secondary importance. Many systems fail or fall into disuse precisely because of
inadequacies in these details. In an object oriented technology such details are modeled
in an UML sequence diagram and some others diagrams related to modeling
functionality. In practice, UML sequence diagrams notation is considered as the most
popular diagram [
Early in the software project lifecycle, details of the specific requirements of the software to be built as well as the staffing needs and other project variables are unclear. The variability in these factors contributes to the uncertainty of project effort estimates. As the sources of variability are further investigated and pinned down, the variability in the project decreases, and thus the variability in the project effort estimates can also decrease.
A number of measurement methods (e.g., Mk II [
The remainder of the paper is organized as follows: Section 2 presents an overview of the COSMIC method, the Structural Size Measurement method and the UML sequence diagram and its Combined Fragments. It also discusses some related works. Section 3 describes the refined Structural Size (SS) Measurement method. Section 4 illustrates its application combined with the COSMIC through the “Digital-Training Center” case study. Section 5 discusses and presents some limitations of our work. Finally, section 6 concludes the presented work and outlines some of its possible extensions. 2 2.1
The COSMIC Functional Size Measurement (FSM) method is becoming popular
because of its ability to size different types of software (e.g., business application,
realtime software, embedded software, mobile apps, neural networks, etc.) [
A Functional process always has at least one entry data movement with either a Write (W) to persistent storage or an exit (X) data movement. This will account for at least two data movements and will be sized as two CFP. Thereafter, these data movements are used to get better insights into sizing the software product.
In our previous work [20], we proposed the Structural Size Measurement (SSM)
method. It was designed by following the measurement process recommended in [
The main reason for creating such method was the need of detailed measures to
quantify data manipulations. As in the COSMIC method where the software functional
size is derived by quantifying the FUR [
The proposed SSM is applied on the combined fragments of a sequence diagram to measure its structural size (SS). This SS, also named control structural size, refers to the structural size of both conditional control structures CCS and iterative control structures ICS, described respectively through the alt, opt, and loop constructs. The SS of a sequence diagram is defined at a fine level of granularity (i.e., the size of the flow graph of their control structures).
The use of SS requires the identification of two types of data manipulation depending on the structure type CCS (alt and opt combined fragments in the flow graph) and-or ICS (loop combined fragment in the flow graph). Each data manipulation is equivalent to 1 CSM (Control Structure Manipulation) unit. The sequence structural size is computed by adding all data manipulations identified for every flow graph.
The UML Sequence diagrams are a popular dynamic modeling solution in UML
because they specifically focus on lifelines, or the processes and objects that live
simultaneously, and the messages exchanged between them to perform a function
before the lifeline ends [
Sequence diagrams can be useful references for software developers who must perform detailed design for each software component, and at different levels of details. Basically, a sequence diagram is drawn to: represent the details of a UML use case; model the logic of a sophisticated procedure, function, or operation; see how objects and components interact with each other to complete a process; plan and understand the detailed functionality of an existing or future scenario.
In sequence diagrams, combined fragments are logical groupings, represented by a rectangle, which contain the conditional structures that affect the flow of messages. A combined fragment contains interaction operands and is defined by the interaction operator. The type of combined fragment is determined by the interaction operator in which the type of logic or conditional statement are identified. The interaction operator defines the behavior of the combined fragment that can be used to describe several control and logic structures in a compact and concise manner.
A combined fragment can also contain nested combined fragments or interaction containing additional conditional structures that represent more complex structures that affect the flow of messages. The combined fragments opt, alt, and loop are summarized in Table 1. The combined fragment “opt”: encloses a sequence that might happen or not. The condition under which it occurs can be specified in the guard.
The combined fragment “alt”: Contains a list of fragments that contain alternative sequences of messages. Only one sequence occurs on any occasion. A guard in each fragment indicates the condition in which it can run. A guard of ELSE indicates a fragment that should run if no other guard is true. If all guards are false and there is no ELSE, then none of the fragments executes.
The “loop” combined fragment repeats the condition as it is indicated in the guard a number of times. The “loop” combined fragments have the properties Min and Max that indicate a minimum and maximum number of times to be repeated by the fragment. The default is not a restriction.
Over the past two decades, researchers have proposed object-oriented software metrics
to measure the quality of software design and improve the productivity [
In this paper, we focus on how to measure the dynamics aspect of software design (e.g., sequence diagrams) at a fine level of granularity. In other words, measuring the sequence diagram structural size not only at one level (as it is described in our previous work [20]), but also at the multi-level where combined fragments (opt, alt, loop) are nested.
This section presents how the structural size measurement method can be improved by taking into account the in-depth nested (multi-level) control structures in the Sequence Diagram (SD). For this purpose, we propose different algorithms for assessing the nested (multi-level) combined fragments (alt, opt, and loop).
In our earlier work we presented a structural size method for measuring data manipulation expressed by the combined fragments opt, alt and loop based on Sequence diagram descriptions. In this section, we extend our earlier work as follows: First, we classify the nested Combined Fragments into three levels and each level is further divided into several cases. Second, we propose different algorithms to support the nested multi-level combined fragment in the sequence diagrams and its corresponding flow graph. 3.1
Before sizing the sequence diagram containing the multi-level nested control structures in terms of CSM units, it is important for measurers to distinguish the combination of nested control structures’ categories in each level that require more attention. This is to avoid misinterpretation of measurement results.
There are many more cases to be identified for classifying the combined fragments into categories at a multi-level hierarchy (including the following three cases). For simplicity sake, we will focus only on the two first categories.
1. Case when one type of the SD control structures is used at all levels. Note that the number of levels can be one or more. 2. Case when two types of the SD control structures are used at all levels. 3. Case when all the above SD control structures are used at all levels.
This category includes the alternative ALT Combined Fragment with several alternatives, the choice OPT Combined Fragment, and the iterative LOOP Combined Fragment.
If the alternative ALT Combined Fragment contains or not one or more nested blocks having the same control structure type (ALT Combined Fragment nested in an ALT Combined Fragment), the following situations should be distinguished: − All levels have n sequence flows and each alternative has nested ALT
Combined Fragment (where n is a constant number) (See Algorithm 1.1)
All levels have n sequence flows and some alternatives do not have nested ALT Combined Fragment (See Algorithm 1.1) 1stlevel has n sequence flows and all other levels have p sequence flows (with p! = n) (See Algorithm 1.2) 1stlevel has n sequence flows and only some other levels have p sequence flows (with p! = n) (See Algorithm 1.2)
Each level has different number of sequence flows If the choice OPT Combined Fragment (two sequence flows where one branch is connected to an end event) and all the nested blocks have the same control structure type (OPT Combined Fragment). − Each level contains OPT Combined Fragment restricted to two sequence flows where one sequence flow is linked to end event (because it contains always a single path) (See Algorithm 1.3).
For an iterative LOOP control structure, we denote the case when each level includes a number of iterative control structures (the number of iterations is arbitrary). The following situations can be observed: − Loop with n iterations at all levels (See Algorithm 1.4) − Loop with n iterations in the first level and p iterations in the next levels (See
Algorithm 1.5) − Loop with different iterations in each level (See Algorithm 1.6)
This category may include the following cases: If the Alt combined fragment in the first level is followed by a nested OPT Combined Fragment, the following situations are considered: − All levels having OPT Combined Fragment (See Algorithm 2.1) − Some levels having OPT Combined Fragment in which the following cases are established: − Each alternative in the first level contains the Opt combined fragment and the other levels may have or not the OPT Combined Fragment (See Algorithm 2.2) − Each alternative may or not have the OPT Combined Fragment in all levels (See Algorithm 2.3)
The OPT Combined Fragment in the first level is followed by a nested ALT Combined Fragment.
The OPT Combined Fragment in the first level is followed by a nested LOOP Combined Fragment (See Algorithm 2.4).
The LOOP Combined Fragment in the first level is followed by a nested OPT Combined Fragment.
The LOOP Combined Fragment in the first level is followed by a nested ALT Combined Fragment.
In this category, all possible combinations of the different control structures can be used.
With the information from the sequence diagram and its corresponding flow graph, measurers can quickly identify the combined fragments at different levels (including the nested level). Thereafter, by applying the proposed algorithms (See Tables 2 and 3), the detailed structural size of a sequence diagram can be generated.
Let:
SDm be a Sequence Diagram containing a nested (multi-level) combined fragments, GSS be its corresponding flow graph, GSS(SDm) be the graph-based structural size function expressing the
Structural Size of the whole Sequence diagram, and GSS(F) be the graph-based structural size function expressing the Structural
Size of the combined fragment within the Sequence diagram.
The sequence diagram structural size is derived as the sum of the Structural size of all the combined fragments (multi-level) as shown by equation (1): (
) = ∑ =1 ( ) (1) where: n is the number of combined fragments Fi in the SDm, SS(Fi): the Structural size of a fragment Fi
Since each fragment represented in a SD can be itself decomposed into a new fragment that refines it, each Fi can contain (or not) one or more nested control structures. The structural sizes of these fragments are derived by using algorithms as illustrated in section 3.2.2. Note that: − GSSic(F) is the (sub)graph representing the structural size of the iterative control structure SS(ICS), − GSScc(F) is the (sub) graph representing the structural size of the conditional control structure SS(CCS) that may include alt combined fragment and-or opt combined fragment.
To measure the structural size of a fragment, we propose algorithms based on the defined strategy.
When a single control structure type is used at all levels, we propose to use Algorithm 1.1 and Algorithm 1.2. Algorithm 1.1 represents how sizing an ALT Combined Fragment nested within another ALT Combined Fragment in the same SDm, where all levels have n alternatives (with n a fixed number). Algorithm 1.2 illustrates the ALT Combined Fragment nested within another ALT Combined Fragment where the 1st level has n alternatives and all other levels have p alternatives (with p! = n). =1 Note that d= ni if there is a nested Alt combined fragment in each alternative ( ) = ∑ ( ( ) ∗ ( ( )/ ) ) Algorithm 1.2 Alt combined fragment nested in an Alt combined fragment: 1st level has n alternatives and all other levels have p alternatives (with p! = n) =1 =1 int m = [1..d]; where d is the number of nodes in Flow graphs modeling Algorithm1.2 Flow graphs modeling Algorithm 1.3 the nested blocks have the same control structure type (Opt combined fragment) int i=0; // i = [0, l] i is the root or the1stlevel alt=n=1; // n is the number of nodes in each level For each level i For each node xi End for End for i++; For each level i For each node yi out-degree (xi) = n=1
( ) = GSS( ) out-degree (yi) = 1
( ) = GSS( ) End for End for End ( ) = ( ( ) ∗ ( ( )/ ) )
= 1 CSM Algorithm 1.4 Each level contains iterative control structures (LOOP Combined Fragment) where the number of iterations is fixed (r) inti=0; // i = [0..l] where i is the root or the1stlevel int itr=r; // r is the number of iterations i=1; For each level i having r number of iterations For each node xi out-degree (xi) = r
( ) = GSS( ) End for i++; For each level i having r number of iterations (r is a fixed number equal to the fixed number in the previous level)
For each node yi out-degree (yi) = r
( ) = GSS( ) End for End For
( ) = End for int m=1….d; where d is the number of nodes in 2nd level out-degree (yim) = 1 = = ∑(GSS( )+GSS( )) = ∑ ( ∑(GSS( )+GSS( )) ) Algorithm 2.2: Each sequence flow in the Alt combined fragment (i=0) contains Opt combined fragment, and other alternatives in the followed levels may have or not an Opt combined fragment fragment Else combined fragment may or not have Opt combined fragment in all levels For each level i contains or not opt For each node y If Opt combined fragment=True End if End For End For End = = ∑( ) =1 Algorithm 2.4: LOOP Combined Fragment nested in the Alt combined fragment: Each alternatives contains LOOP Combined Fragment (with u a fixed number of iterations) Begin int i=0; // i is the number of levels outdegree(x)=0; Alt=n; // n is the number of nodes
For each level i For each node x out-degree (x) = n*i
= GSS( ) end for i++ For each level i having u number of iterations For each node y out-degree (y) = u
= GSS( ) end for i++ For each level I (i=2) having w number of iterations (wis a random number and it is NOT equal to the previous levels)
For each node z //z represents all nodes in any levels except level 0 and 1.
out-degree (z) = w
= GSS( ) = ( ) ∗
( ) =
= ( ) + ( ) End for End For End for end for End Flow graphs modeling Algorithm 2.4 4
This section illustrates the application of our proposed method through the case study “Digital-Training Center” web site. It considers the interactions among trainings in this web site, as an example of UML2.x SD modeling the behaviors of a distributed system (see Fig.1 and Fig.2). The web site has independent components: the training officer, the home page, the custom page, the training page and the authentication page.
We model the update of the training catalog by the training officer (Fig. 2): the training officer must authenticate at first (Fig. 1). He enters the web site URL and he can be redirected directly to the custom page if he has chosen, in the last connection, the option to remain connected for a limited duration. Otherwise, he is redirected to the authentication page. There are three possible cases: 1. successful authentication; 2. missing information - in this case the site asks the user to complete them; 3. wrong login and-or wrong password - in this case the site asks the training officer to correct them.
In the second case, domain name system (DNS) attacks may accidentally occur. Fig. 2 depicts the update of the training catalog. The training officer requests to update the training catalog: he is redirected to the training page where he has three possible alternatives: i) he can add new training and create a calendar session; ii) he can choose a training and update a session, iii) he can remove a training - the list of participants to this training is displayed.
User requirements are the basis core of software projects and sizing requirements is a crucial task for software project planning. Hence, if these requirements are poorly described (in the analysis phase of the SLC) and modeled (in the design phase), the software development/maintenance planning will be vulnerable. In practice, well detailed User requirement are not always available since they require much time and comprehension. However, the more time spent at the beginning of the process, the less time will be required later. Consequently, detailed descriptions of user requirements are needed: these detailed requirements can be easily represented in the form of sequence diagrams that can be measured.
In our work, the proposed algorithms dealing with the nested (muti-level) Combined Fragments in the Sequence Diagram have been illustrated through the case study “Digital-Training Center”. To explore the efficiency of our refined measurementbased algorithms, these algorithms must next be applied and tested in an industrial setting, particularly in projects having a complex functionality that requires to be modeled with these nested multi-levels. 6
This paper extends our previous work [20] on a structural size measurement method by proposing additional algorithms for measuring the combination of all nested (multilevel) combined fragments. This extension is based on the most popular combined fragments (ALT, OPT and LOOP) identified in the sequence diagrams (SD) as a whole (without parsing the SD). More specifically, we proposed several algorithms to measure the different categories of partial order between the events. We also proposed some additional contributions for refining our proposed structural size method by focusing on the different combinations of nested combined fragments of the SD in order to support nested multi–level Combined Fragment. In our future work, we will focus on automating the combination of functional and the extended Structural Size Measurement methods. Furthermore, we plan to investigate the combination of these measures to improve projects and tests effort estimation models by taking into account both Functional size measurement and this refined Structural Size Measurement. Moreover, this combination of measures should be explored by scrum masters to evaluate and prioritize user stories and make appropriate decisions within an agile context.