Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

MULTI-PROJECT SCHEDULING AND CONTROL: A PROCESS-BASED COMPARATIVE STUDY OF TH...Izack Cohen; Avishai Mandelbaum; Avraham ShtubProject Management Journal; Jun 2004; 35, 2; ABI/INFORM Globalpg. 39

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

June 2004 Project Management Journal • 41

able delays of the critical chain.Consequently, feeding buffers areadded at the end of each non-criticalactivity chain (“pushing” the latterback in time, in response to a “latestart”). The feeding buffers thus pro-tect the critical chain from variationsof non-critical chains and allow criti-cal-chain activities to start early, whenpossible. According to Leach (1999),a feeding-buffer capacity is set to 50%of the duration of its non-criticalactivity chain.

Step S5: Control.Buffer monitoring provides a

quick grasp of project status, which, inturn, enables adaptive control.Specifically, buffer consumption thatreaches a predefined threshold (e.g.,two-thirds of the buffer size or, equiva-lently, one-third of the slack timeremains unused; Leach, 1999) triggersan early warning toward taking somepreventive managerial action. Moredetails are provided later in this paper.

Multiple projects are accommo-dated by combining single-projectscheduling with TOC (Goldratt, 1984)and CC principles, notably theemphasis on reducing multi-tasking(Herroelen & Leus, 2001; Leach,1999). To this end, project start-timesare staggered, which turns the multi-project system into a “pull” systemwith newly determined release/starttimes. Following are the relevantdetails.

Scheduling and control of a multi-project system:

Step M1: Treat each project as asingle project.

Individually schedule each of themulti-projects, using the four steps forscheduling a single project, as describedin Steps S1–S4.

Step M2: Stagger projectsaccording to the bottleneck resource.

First identify the bottleneck,namely the most constrainingresource (often by simply usingmanagerial experience). Thenrelease projects sequentially, by stag-gering them, so that the bottleneckworks continuously and there is noidle time.

Step M3: Create a capacity buffer.A time buffer, called a capacity

buffer, is associated with the bottle-neck, and its role is to ensure bottle-neck availability. The capacity bufferdecouples between bottleneck activi-ties that belong to successive projects,thus determining projects’ start times.Since, based on a literature survey,there is no standard way to set the sizeof this capacity buffer, we set its base-case size at 50% of the duration of thebottleneck activity. We then analyzethe effect of alternative sizes by varyingthe values through 8.3%, 16.7%,83.3% and 116.7%.

Step M4: Control.As with single projects, scheduling

control of multi-projects is buffer-based: when allocating an idleresource, top priority is given to criti-cal-chain activities over non-critical-chain activities; secondary priority isgiven to activities of projects with thehighest level of project buffer utiliza-tion or, equivalently, the least slacktime. Least priority, in turn, is given toactivities of projects with the highestfeeding buffer consumption.

From Project to Process ManagementFollowing Adler et al. (1995), wemodel a multi-project organization asa stochastic processing network. Adleret al. (1995) validated the modelbased on an actual research and devel-opment organization, showing thatthe model simulated quite accuratelyits performance.

In the model of a stochastic pro-cessing network, each network noderepresents a group of (one or more)statistically identical resources, whoperform the same type of activities andwho are able to do so in parallel.When several activities of a project can

start being processed at the same time,we refer to the phenomenon as a“fork;” when an activity cannot beginuntil its predecessor activities havebeen completed, we call it a “join.”(Consequently, such models are oftenreferred to as fork-join queues. Forexample, see Nelson & Tantawi, 1988.)The time required to complete anactivity is called its processing time(duration) and the intervals betweensuccessive project releases are “inter-arrival times.” The reciprocal of the

Table 1 / Cohen

Inter-arrival Exp(1/3.25)

Activity A 1 3 Exp(1/6)

Activity B 2 2 Exp(1/5)

Activity C 3 3 Exp(1/4)

Activity D 4 1 Exp(1/3)

Resource Number of Time � Type Resources Distribution

Table 1. Characteristics of our multi-project system: number of resource-units per type, processing time distribution and inter-arrival time distribution. The notation Exp(l) represents anexponential distribution with probability density function f(t)= le–lt (and expectation 1/l)

Fig 1 / Cohen

A, 1

B, 2

C, 3

D, 4

Resource Queue Synchronization Queue Activity Type (I), Resource Type (#)I, #

FS

Figure 1. The Stochastic Processing Network approach for representing a multi-project system

PMI-010 June_PMJ 9/22/04 12:55 PM Page 41

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.