Free Goods

Free goods are defined as goods that do not require any resource constraint involvement in their production-they require solely non-constraints. Free goods represent an opportunity for immediate increase of Throughput with little to no increase in Operating Expense (recall that Throughput accounts for raw material expense, items that are truly variable costs). However, Chakravorty and Atwater (2005) found that DBR is very sensitive to levels of free goods. Therefore, schedules using DBR need to be aware of how orders of free goods are accepted. Specifically, they found that the number of tardy orders increased as the level of free goods released to the shop increased. They attribute this phenomenon to the loss of protective capacity at certain non-constraint resources. Atwater, Stephens, and Chakravorty (2004) discuss the impact of free goods on system Throughput. They found three basic insights for the system they modeled. First, operating the resource constraint at a level above 98 percent resulted in erratic Throughput performance. Second, increasing protective capacity above 7 percent did not significantly improve on-time performance. That is, once a nonconstraint's capacity reached 107 percent of the constraint's capacity, further increases in capacity did not improve on-time performance. Of course, this value would be very sensitive to the number and duration of statistical fluctuations included in the model. Third, when demand for constraint goods is high, managers can improve on-time performance by limiting the orders they accept for free goods (refusing such an order would reduce future utilization of non-constraint resources).

What If the Market Is the Constraint?

What if all goods are free goods? That is, what if the market is the constraint? Pass and Ronen (2003) define a market constraint as a situation in which the production capacity of every resource exceeds demand for it; they address this issue for a high-tech firm. They note that it is usually easier to control an internal constraint that is under the roof of management than to be tossed by the ups and downs of the market. The R&D department is usually a constraint because new products are not coming out fast enough. In addition, marketing or sales may be a constraint. Since lead time is a factor of competition, small batches may be run in order to shorten lead time. This involves more setups but most non-constraints can afford that (as Goldratt noted in The Goal [Goldratt and Cox, 1984; 1993]) A dummy constraint is a resource constraint that is inexpensively eliminated. Pass and Ronen (2003) note two common dummy constraints in marketing and sales: (1) Shortage of inexpensive administrative assistance and (2) lack of laptops and communications equipment such as portable fax machines. They further note three common dummy constraints in R&D: (1) Shortage of low cost components and accessories, (2) shortage of low cost administrative assistance, and (3) lack of computers and IT tools. Breaking these dummy constraints may give a significant elevation to the market constraint.

Smith et al. (1999) also mention DBR as an aid in product development at Allied Signal and Alcoa.

Re-Entrant Flows

Wu and Yeh (2006) describe the use of DBR in a situation in which a part passes through the constraint twice in flowing through the plant, known as "re-entrant flows." This situation commonly occurs in semiconductor manufacturing. According to Wu and Yeh, the method of scheduling using DBR as described in The Haystack Syndrome (Goldratt, 1990) cannot effectively schedule environments with bottleneck re-entrant flows. They cite a number of articles describing the use of DBR in re-entrant flows including Huang et al. (2002), Kayton et al. (1996; 1997), Kim et al. (2003b), Klusewitz and Rerick (1996), Levison (1998), Mosely et al. (1998), Murphy (1994), Murphy and Dedera (1996), Rose et al. (1995a, b), Tyan et al. (2002), and Villforth (1994). Wu and Yeh (2006) then propose a scheduling method for DBR that they feel is appropriate for manufacturing facilities with bottleneck re-entrant flows. Rippenhagen and Krishnaswamy (1998) simulated a wafer fabrication facility with re-entrant flows using a variety of dispatching rules and Theory of Constraints. Kim et al. (2009) report on a simulation study of a hypothetical wafer facility with re-entrant flows and protective capacity. They are interested in, among other things, the trade-off between protective capacity and protective inventory. The study is based on a six-station line with re-entrant flows and times per part ranging from 8 min to 12 min and protective capacity ranging from 1 min per part to 4 min. They found that simply knowing the percentage of downtime at non-constraints was not sufficient to understand the need for protective capacity and inventory. Specifically, they found that infrequent long outages required more protective capacity/inventory than did frequent short outages even though the proportion of time the station was out was the same. They also found that resource downtime had more impact on the constraint than did processing time variation. They found that allocation of protective capacity throughout the line was more important than protective inventory. WIP inventory involves a tradeoff between Throughput level and cycle time. Beyond some point, adding more inventory does not improve Throughput, so an appropriate level must be chosen.

Recoverable Manufacturing and Remanufacturing

Guide (1997) discusses the successful application of DBR to recoverable manufacturing, where used products are returned from the consumer to the manufacturer, who then remanufactures the product. Guide uses the term "recoverable product environment" to describe the processes to recover materials via recycling at the end of the product life. Guide (1996) showed that DBR could be a successful production planning and control system for remanufacturing.

Buffer Management Literature

While few of the above simulation or case studies above recognize Buffer Management as a necessary condition for an effective Drum-Buffer-Rope planning and control system, most TOC experts today agree on its vital importance both in expediting orders before they become late and also as the foundation of a process of ongoing improvement. The TOCICO Dictionary (Sullivan et al., 2007, 7) defines buffer management as "A feedback mechanism used during the execution phase of operations, distribution and project management that provides a means to prioritize work, to know when to expedite, to identify where protective capacity is insufficient, and to resize buffers when needed." ( TOCICO 2007, used by permission, all rights reserved.) When an item is released to the floor, it is released into a buffer-constraint, shipping, or assembly buffer depending on the shop's configuration. Buffers are sized so that each batch or order should arrive at the buffer in time to maintain the buffer approximately half-full. The buffer is actually divided into three regions9, each representing one-third of the buffer length. Region I (Red) consists of the oldest batches, which should be processed soon; Region II (Yellow) represents intermediate batches, about half of which should be in the buffer; and Region III (Green) represents the most recently released material, which is generally expected to still be en route to the buffer. If material released into the shop and under the control of the buffer has not yet reached the buffer it is called a buffer hole. Simatupang (2000) provides a good description of Buffer Management activities.

There is a person called the buffer manager whose responsibility is to steer material into the buffer on time. Holes in the Green Region of the buffer require no action. If a hole moves into the Yellow Region, the buffer manager will locate the item and remind the workstation holding the batch that it is soon due in the buffer. If a hole reaches the Red Region, the buffer manager will expedite the batch through the station holding the material and any stations between the batch's location and the physical buffer. The buffer manager will also note the location of the expedited batch and the reason for delay for prioritizing future improvement efforts. Gardiner et al. (1992) state that 90 percent of orders should require no expediting if the buffer is properly sized. The buffer size is dynamic-if too much expediting is occurring, the buffer can be made bigger; if virtually no expediting is occurring, the buffer can be made smaller. Because of Buffer Management and dynamic buffer sizing, the initial size of the buffer is not that critical-if it is initially the wrong size, Buffer Management activities will quickly reveal that fact and the buffer can be resized.

When a job is released to the shop floor, its paperwork should show the due date of the job in the buffer toward which it is moving. The supervisor of each workstation can use this information as an aid in sequencing jobs. The buffer manager of the buffer involved has a sequenced list of jobs due in the buffer, which he can use in determining the location of holes in the buffer and to decide whether to begin investigative action or expediting.

Tseng and Wu (2006) describe Buffer Management in a modified system employing five buffer regions rather than three: early arrival zone, ignored zone, mentioned zone, expediting zone, and delayed zone. The three middle zones correspond to the normal three regions of the buffer, while the first zone represents material released to the shop too early and the fifth zone represents material not processed in time. Simatupang (2000) describes how Buffer Management can be used to direct the application of preventive maintenance activities.

Schragenheim and Dettmer (2001) describe a variation on Buffer Management called the "red-line control mechanism," which collects data on jobs that are about to be late and assists managers in determining the stability of the shop floor.

Buffer Sizing

One of the questions that must be addressed in establishing a DBR system is, "What size should the buffers be?" Goldratt has suggested to Jonah courses that an initial buffer size can be developed by taking one-half the current lead time and dividing that time between the constraint time buffer and the shipping buffer. This initial buffer size can be adjusted up or down by whether too few or too many jobs require some expediting via Buffer Management. This suggestion has worked its way into the literature.

Louw and Page (2004) state that the determination of the time buffer lengths is a trial-and-error approach that consists of first determining the initial size of the time buffers through simple empirical rules (Srikanth and Umble, 1997; Tu and Li, 1998). The buffer lengths are then monitored and adjusted through a process known as Buffer Management (Goldratt, 1990; Schragenheim and Ronen, 1991). Goldratt (1990) suggests determining the initial buffer lengths by estimating the current average lead time of the tasks to the specific buffer origin and dividing it by five. Srikanth and Umble (1997) suggest the total time buffer for any product should be approximately one-half the firm's current manufacturing lead time, whereas Schragenheim and Ronen (1990) suggest a constraint buffer size of three times the minimum cumulative processing time to the constraint.

Louw and Page (2004) use a procedure for estimating the sizes of the time buffers based on a queuing model in a multi-product open queuing network. Details of this network are beyond the scope of this chapter. Ye and Han (2008) use a mathematical approach to estimate both the time buffer and the assembly buffer sizes.

Weiss (1999) presents a queuing network using separated continuous linear programs, which he says is similar to DBR in that it tends to form buffers at the busiest stations.

Taylor (2002) points out that attempting to remove all system variability is not cost effective. It is better to buffer the constraint and to some extent buffer CCRs in order to protect them from starvation. Taylor simulated MRP, JIT, and DBR systems and compared their influence on a number of operations performance measures.

Some companies have been hesitant to start DBR because they do not know how to set the buffer sizes. Because adjustments to buffer size suggested by Buffer Management will quickly correct any initial buffer size estimate, companies should simply pick a conservative buffer size and get started.

Buffer Sizing and Lead Time

In a serial line with a single resource constraint, there should be two buffers-the time buffer at the constraint and the shipping buffer. The manufacturing lead time through the system should approximate the sum of the two buffer sizes. Even in arrangements that are more complex, this statement would be true unless there is a non-constraint assembly between constraint parts and non-constraint parts and one of the non-constraint parts had a longer lead time to the assembly point than the constraint part. In this case, the lead time should approximate the sum of the assembly buffer and the time allowed to flow from the assembly point to the shipping dock. This relationship, and its importance, is explained at length in Chapters 9 and 10 of this book and by Fry et al. (1991).

TOC and Distribution

Little has been written about the TOC solution to a distribution environment such as a supply chain. However, in the early 1990s, Goldratt utilized a distribution simulator to teach the TOC approach to distribution in various classes. Recently, Schragenheim et al. (2009) published a chapter on the distribution environment with a very thorough treatment. Imagine an environment in which a manufacturer produces a variety of products that are distributed via a set of warehouses to a larger set of retailers. Under traditional management, it is common for the retailer to order an entire season's supply of an item to arrive before the season begins based on a forecast of what sales may be. However, forecasts are always wrong, so the retailer usually runs out of stock before the season ends or has excess stock left at the end of the season that must be sold at greatly reduced prices. In addition, there is the problem of storing that inventory during the season. The TOC solution, as described by Schragenheim et al. (2009), begins with a plan to deliver rather frequently during the season an amount equal to actual sales during the previous delivery period. This requires the retailer to begin the season (and each replenishment cycle) with a stock equal only to the maximum likely sales during the replenishment period. The regional warehouses will hold some stock but most stock will be held in a central warehouse at the manufacturer. This approach takes advantage of the fact that relative variation is much smaller at the manufacturer than it is at the typical retailer. There is less stock in the system, but availability of the item at the retailer is increased because of the frequent deliveries. This is essentially a DBR process applied over the supply chain. Experience has shown that the increase in Throughput far surpasses any increase in the transportation costs for more frequent deliveries.

Supply Chain Management

Simatupang et al. (2004) discuss the application of TOC to supply chain management. A supply chain consists of different firms that deliver products and services from raw materials to end customers. All the different players such as manufacturers, distributors, and retailers play significant roles in creating value for the ultimate customer. They also note that reliable global performance measures help the chain members to measure progress. They introduce the performance measures Throughput Dollar Days (TDD), a measure of things done too late and thus endangering Throughput, and Inventory Dollar Days (IDD), a measure of things done too early (or that should not have been done) and thus incurring extra inventory carrying costs.

An important aspect of supply chain management is the decision of whether to out-source particular components-the make-or-buy decision. Of course, component quality is one of the most important aspects of this decision, if not the single most important aspect. Traditionally, cost has been the second most important factor-the cost to make versus the cost to buy. However, in TOC, the decision's impact on Throughput is important and Throughput is impacted in different ways depending on whether making the component requires time at a resource constraint (and perhaps also whether it requires time at a CCR). If making the part requires only non-constraint time and no worker will be laid off because of outsourcing, then traditional cost accounting overestimates the marginal cost of making the part. If the part does require constraint time, then purchasing the part allows additional units of the least profitable part to be added to the drum schedule, thus increasing Throughput. Traditional cost accounting underestimates the opportunity cost of making the part. Either way, TOC arrives at different numbers for the decision than does traditional accounting. This decision is discussed at length in Gardiner and Blackstone (1991) and is updated by Balakrishnan and Cheng (2005), who point out that if the part is a strategic part, then the cost to buy may not be the most important consideration. The make-or-buy decision is also mentioned in Hilmola (2001).

Walker (2002) provides an excellent discussion of the application of DBR to a supply chain. He discusses how to choose which partner should be the drum, how to tie the rope, total system Inventory measures, and managing as demand goes up and down. Walker (2005) states that the applicability of DBR has been expanded to include the entire supply chain network.

Cox and Walker (2006) have published a board game that uses poker chips in a stochastic supply chain. The players can alter the order policies and batching policies at various points in the supply chain and observe directly the impact on Inventory and service levels.

Service Environment

One of the reasons for keeping buffers as small as practical while not starving the constraint is that if there is too much work in a facility, then workers have a tendency to move back-and-forth between jobs, thus wasting some of their time with extra setups. In Chapter 21 of this handbook, Herman and Goldratt point out that this problem is also true in sales. They include a Current Reality Tree (CRT) describing the problem. Umble and Umble (2006) describe how Buffer Management was used in two accident and emergency facilities in Oxfordshire, UK to track patient care. Motwani, Klein, and Harowitz (1996a; 1996b) have a two-part article describing the use of TOC and DBR in services, in general, with a specific example from health care.

TOC and Other Modern Philosophies10

TOC and Lean

Dettmer (undated) states that the Toyota Production System is better known than TOC primarily because it is a much older system (development started in the 1950s) versus TOC in the 1980s). He continues by saying that both systems use continuous improvement and have the goal of obtaining higher profits. Both methods recognize that the customer is the final arbiter of what value is.

Berry and Smith (2005) provide a comparison of TOC with Lean and with several other approaches-MRP, MRP II, ERP, and Supply Chain Management.

Sale and Inman (2003) surveyed over 900 firms and received 93 responses. They found that firms using TOC had significantly higher performance improvement than firms using JIT and traditional manufacturing.

Moore and Scheinkopf (1998) compare TOC and Lean. Both TOC and Lean concentrate on continuous improvement and control the flow of material on the shop floor. Both have had dramatic improvements of profitability and lead times and have resulted in operations being drastically simplified.

TOC and TQM

Lepore and Cohen (1999) suggest there are many synergies between TOC and Total Quality Management (TQM). Cohen is one of Goldratt's early partners. Lepore is an academic specializing in Total Quality Management. They suggest a 10-step strategy for implementing the two philosophies together. Step 4 is to implement the 5FS. Step 5 is to implement Buffer Management. However, the book contains little specifics on DBR, per se.

TOC, Lean, and Six Sigma

Pirasteh and Farah (2006, 3233) state that the top elements of TOC, Lean and Six Sigma work well together. They report on a company that combined the "best components" of these three approaches into what they called TLS. They applied Six Sigma alone to 11 plants, Lean alone to 4 plants, and TLS to 6 plants. They measured plant performance regarding "on-time delivery, warranty costs, customer returns, Inventory reduction, cycle time reduction, and scrap expense." The company concluded that "the TLS process improvement methodology delivered considerably higher cost savings to the company."

Problems with DBR