How Is the Planning and Execution Viewpoint Addressing the Issue of Scheduling and Buffering the CCR?

This viewpoint requires validating that whatever is included in the planning is a must. Thus, the question is, "Do we need to schedule the CCR?" Is detailed scheduling the only way to ensure a good enough exploitation of the system constraints?

Once we recognize that even the CCR has to subordinate to the commitments made to the market, then we have to conclude that scheduling the CCR in detail is not required in most cases (later we deal with the one exception). This means also that the CCR buffer, used to protect the schedule of the CCR, is not required and the only buffer that is truly necessary is the one aimed to protect the commitment to the market. The CCR should prioritize its sequencing decisions according to BM at the shipping buffer. However, the load on the CCR still requires monitoring. The insight here is that there is a difference between monitoring the load on a CCR resource and dictating a sequence on that resource.

How Does Refraining From a Detailed Schedule of the CCR Affect the Execution?

The traditional DBR requires three different buffers: the constraint, the shipping, and the assembly buffers (detailed in Goldratt, 1990), but if one concentrates just on the due dates of the firm orders at hand, then only one buffer, the shipping buffer (now called the production buffer in S-DBR), covers the whole production time from material release until order completion, is required. Following the green-yellow-red buffer priorities8 as they emerge from the use of having one-time buffer per order is much simpler than deciding between a red assembly buffer versus a red CCR buffer or a red shipping buffer.

What Does the Emphasis on Flow Add to the Challenge to Traditional DBR?

This view is fully concentrated on the trigger of the flow-the customer order. The point of the flow is to be able to commit to the client as fast as possible. With this as the main objective, then the challenge to DBR is obvious: Do we really need the constraint buffer or is it a disruption of the flow? After all, the constraint buffer initiates early release of the parts, so on average they reach the constraint and then wait for their schedule to be processed. Having that planned waiting time at the CCR is a disruption to the flow.

It is obvious that there is a need to choke the release to only what is truly required now. This would prevent the tendency in traditional DBR to release certain orders very early in order to exploit the CCR's capacity.9 Looking on the CCR as "disruptive to the flow" also highlights the need to be able to quickly elevate the CCR capacity whenever necessary, because even the unplanned wait time at the CCR could be significant due to its relative lack of capacity and any wait time represents a certain disruption to the flow. Of course, this is not always practical, but the basic thinking is right. We now recognize in TOC that the ultimate goal is to both grow constantly and remain stable at all times. Thus, the constraint should not be the capacity of a resource that can be easily elevated because whenever such a resource cannot fully subordinate to the demand (forces too long wait time, thus blocking the flow) it should be elevated. The underlying assumption is that the financial worth of the additional demand that would not be possible to maintain with such a blockage to the flow is worth more than the capacity of a common resource that is easy to elevate.

Outlining the Direction of the Solution

Challenging the DBR procedure of finite-capacity scheduling of the CCR does not mean we are looking for something drastically new. Actually, most of the wisdom included in the original solution is still intact. The most critical insight in DBR, which we have already mentioned, is worth mentioning again: As complex as the production shop floor may be, the performance of the shop as a whole is impacted mainly by a single work center, which determines both the response time and the maximum potential output of the floor.

This insight is relevant for S-DBR as well, even though the CCR is not scheduled in detail and does not have a specific time buffer. The term "weakest link" is perhaps more appropriate than CCR because the weakest link is not always a constraint. Nevertheless, it is always something to monitor because both the potential maximum output and the possible response time are impacted by it. Thus, the weakest link could be used to signal the sales department when additional efforts would be most beneficial and when more care should be given to the quoted delivery time given to potential customers.

The Main Ingredients of the Solution

The solution described here refers clearly to make-to-order (MTO). Another chapter is dedicated to make-to-stock (MTS), or rather make-to-availability (MTA).

S-DBR is targeted at the very short-term. Capacity planning for medium- and long-term is not included within the S-DBR methodology, even though certain information could be extracted from the S-DBR and BM that could support longer-term plans.

Regular short-term planning concentrates on: 1. When to quote the due date for production completion. The underlying assumption is that the due date has to be reliable (safe).

2. When to release the materials.

Two critical tools are required for the planning: 1. The time buffer (production buffer) to be assigned to a manufacturing order for a particular product.

2. The load control on one resource. It is possible though to extend the load control to several resources, but only one of them is truly material in dictating the due date and the material release dates.10 There is yet another piece of information that is important-the standard lead time for the product in the relevant market. It is important because the plan will not always dictate the earliest date that the algorithm based on the load control and the time buffer would come up with. For example, suppose the standard lead time in the industry is four weeks, but because sales are, at the moment, rather low you can deliver an order in just one week. Would you offer one-week delivery? Well, if the customer is willing to pay a high markup you might agree. Otherwise, this seems like a marketing mistake. First, you might give the impression that you are so pressed for money that you are ready to do anything, even at the expense of quality, just to get the order. Moreover, if you get the order, the future expectation might be that you can always complete the order in a week and when you refuse the request, it is because you don't respect your client.

Thus, the standard lead time is a reference and whenever feasible this should be the date to quote for the regular price and with the promise of absolutely guaranteed on-time completion. Of course, in off-peak periods a company could use the advantage of shorter lead time. Generally speaking, Marketing and Sales should set a clear logical approach to quoting lead times.

The Time Buffer

At the time Manufacturing at Warp Speed (Schragenheim and Dettmer, 2000) was written, the shipping buffer was understood as the shortest time we could safely commit to deliver. For instance, if a certain order could be safely delivered in 12 days, then the order had a time buffer of 12 days.

Additional insight from Goldratt led us to distinguish between two different periods that comprise the shortest safe time from order receipt11 to order completion.

1. The time the order has to wait in queue until the signal from the rope to be released to the shop floor. This prerelease queue time depends on the prior work that the CCR has to do. Assuming the CCR has quite a lot to do, then releasing the order immediately does not add any benefit and could cause damage by creating confusion of priorities.

2. A liberal estimation of the time it takes from order release until completion (the production buffer 12). As we are going to consider the current queue for the CCR and possibly delay the actual release of the materials for that order, we do not expect a peak load within the shop floor itself. When a peak load happens, then new orders would need to wait longer before being released to the floor, thus decoupling the flow in the shop from the natural pace of the CCR.

The time buffer mentioned in Item 2 is now called the production buffer, as it describes the flow time through the shop under regular load. The production buffer does not include transportation or in-transit time to the client.

The issue of delivery date to the client requires a short discussion. The transportation time is an issue only when it is a significant part of the lead time. The question is whether the commitment of the producer includes transportation. In other words, is the transportation time part of the production planning or part of the customer's planning?

Suppose the producer takes responsibility until the goods reach the customer. Then, the production planning should have a due date for completing the production and then have the final delivery date where the transportation time (and possible fluctuations in it) is considered.

Figure 9-1 shows time elements in lead time, but note that from now on we'll treat the completion of the production as the due date to which we refer.

How should the production buffer be determined? In implementing S-DBR in traditional planning and control environments, the usual recommendation is to cut the current production lead time by half. The rationale is that by eliminating the large batching and huge WIP levels, the main disruption (waiting in queues at each work center) to the flow has been vastly reduced.13 Take into account that the net processing time is just a fraction of the production lead time and you realize that by cutting the wait time to half, the total production lead time is cut by half. The priority system of BM supports very high reliability within that time. Thus, cutting the standard production lead time by half is a good initial production time buffer.14 Noting the standard lead time of the industry is a good gauge for a test measurement and the time buffer to be used should not be longer than half of that number. In most cases, this reduction is not only possible but often the production time buffer can be cut even further. These further cuts should be done only after first implementing S-DBR with buffers that are equal to half the current lead times. After the shop floor has stabilized, the further reduction of the production buffers is achieved through BM with the focus of improving the flow. Recall improving flow is the main mission of operations. Marketing and Sales should then capitalize on this reduced production lead time to get higher prices and expand the market almost at will. It means Marketing and Sales have to be fully updated with the new capabilities and current status in production.

FIGURE 9-1 The elements of lead time.

Load Control

In traditional DBR, the role of the drum, the detailed schedule of the CCR, was to measure the load on the CCR and determine whether the promised due date was safe. When no detailed schedule is going to be used, then a replacement tool for measuring the load is required.

The planned load is the accumulation of the derived load on the CCR, or any other relatively loaded resource, of all the firm orders (released and awaiting release) that have to be delivered within a certain horizon of time. It is clear that more than one horizon of time might be defined. Each horizon provides a planned load for a specific use. However, a horizon is required for ensuring we do not promise delivery dates that cannot be met is of special importance. The time horizon for such a decision has to include all the orders already received (both released and not released to production) that might compete for the capacity required for the new received order. The important parameter is the realistic expectation of the market regarding the due date. If we assume that a client submitting an order expects to get it no later than the standard lead time of that industry, then the horizon must include all orders to be delivered within the standard lead time. We might need to extend that horizon just a little to cover for a peak load that would force quoting a somewhat later due date.

There is definitely no wish for extending the horizon significantly more than the clients' expectations for response time to an order. It could be that some orders in the logbook have dates further out into the future, due to some clients wishing to ensure the availability of capacity for their orders. Nevertheless, do we need to consider those orders when we check the feasibility of delivering an order just received? Only when such an order enters the planning horizon should the capacity required for that order be considered.

What benefit does the planned load give? The most important information to deduce from the planned load is the approximate prediction regarding the time a new order will be processed by the CCR (the weakest link). Such information is critical for judging a "safe date" for completion of the order. It is only a gross approximation of the time the CCR would really process the order because our data are not necessarily precise and we do not guarantee the sequence of processing at the CCR. In addition, in order to obtain the planned load we simply add the load on the CCR of every order to be delivered in the horizon. Therefore, we have actually assumed that the CCR would not have idle time. See Fig. 9-2. Thus, the timing when a new order would have the chance to be processed by the CCR is far from being accurate, but it can serve as a gross assessment. All we need to know is what due date is safe enough to promise completion of the order. For that, we need the planned load and to add to it a certain time buffer, as we'll soon see.

FIGURE 9-2 Demonstrating the calculation of the planned load.

Technically, the planned load looks like a schedule. It is generated by taking all the planned orders to be delivered and adding the required time the CCR has to invest on the timeline. The end of the planned load is a date-the approximate date when the CCR would finish processing all the currently known orders. In Fig. 9-2, this date is 06/13/10. The important aspect of this date is that this almost arbitrary sequence is not forced on the CCR. The CCR is expected to follow the general priorities of BM and when some customer orders are stuck upstream, then other orders are to be processed early and the delayed customer orders would get higher priority when they show up at the CCR.

To demonstrate further that the planned load is not a schedule, let's review the following example. Suppose that we are in a re-entrant environment where the order goes through the same resources several times. In the planned load, all the accumulated time the CCR has to invest in that order appears once. This view definitely does not provide a realistic schedule as the order is processed by the CCR several times and in-between these separate processing times other resources would be used to process this order. Suppose there are four machines, M1M2M3M4, and a typical order goes through this sequence of resources six times. No matter which one of the four machines is the weakest link, there will be six times where the CCR would work on the order. Therefore, when such an order is put on the timeline of the CCR, it is placed on one continuous location where all the required capacity for six operations is included. On one hand, such a description is not realistic. However, as a gross average, this approximation is good enough for the time frame when the CCR would process the order and the total amount of capacity required. The first operation would probably be completed soon after the order is released to the floor, while the last one would be done significantly later. However, on average, all the operations will require approximately the time allotted in the planned load. We want to estimate the possible due date promised to the next order received. Then, the average time frame when the CCR would process the order is good enough, assuming, of course, the production buffer is long enough for the six iterations through the CCR.

In what sense does the planned load serve as a load control? The planned load represents a time frame that the CCR has to process every order to be delivered within the horizon. Therefore, the planned load, the date when it will finish on the CCR, should definitely be earlier than the horizon at hand. It needs to be shorter than the horizon by at least the time that is required by an order to move through the CCR to completion. For example, if the planned load finishing date on the CCR is three weeks from now and the horizon is four weeks, then the minimal request for load control is that one week is more than enough for an order to flow from the CCR until completion of production for the order.

Surely, the planned load does not ensure that every single order, which is included somewhere in the order logbook, has enough time to complete production on time.

Determining the Safe Dates

Traditional DBR, like most production planning methods, assumes that once an order is received for production planning it already includes a mandatory completion date. Of course, there are times where a time quotation is required, but in most cases production planning gets the orders with the dates and then they must do their best to meet all of the requests on time.

In DBR methodology, there is a check, which is done immediately after the finite-capacity schedule of the CCR is completed, to validate that all the promised deliveries are secure. The check compares the time the CCR is scheduled to finish processing the order and the completion time for producing the order. The time difference must be equal to or longer than half of the shipping buffer. As a reminder, the shipping buffer in DBR is the liberal estimation of the required time between the CCR and order completion. Eventually, in the process of scheduling the CCR, that time difference cannot always be maintained. Sometimes the schedule provides more time than required for the order to pass from the CCR to completion, but many times less time than the shipping buffer is provided. Does it mean that when the time difference between the CCR schedule and the due date is somewhat less than the ideal shipping buffer the order is doomed to be late? The assumption used in that check is that as long as half of the shipping buffer time is provided, then the completion time is secure enough because BM would give the order enough priority to pull the order through the remaining work centers to completion.15 In S-DBR, we need to develop the procedure for ensuring that the completion time can be secured. However, let's now challenge the assumption that production planning is given a due date and then, in some cases, Sales needs to contact the client and renegotiate that date. In other words, we must let Sales know at all times what dates can be promised to any new orders coming in. DBR production planning could not give a due date to an order that is not yet included in the schedule. Of course, one could always look at the DBR schedule and based on it assume what a safe date might be. Only later will the DBR schedule confirm the date or advise to delay the promised delivery date. In S-DBR, we can be much more flexible.

A procedure that would make it clear what dates can be safely promised to any incoming order generates important benefits. First, once Sales is convinced that those dates are not manipulated by Production, then it settles one of the main causes for the inherent tension between Sales and Production. Second, it opens a way to draw more of the CCR at times of peak demand by constantly smoothing the load on the CCR by giving dates that are based on the current load.

The rule for determining a safe due date is the current planned load date (the first opening in the planned load) plus half of the production buffer time.

In Fig. 9-3 we see a graphic illustration of the time segments involved in computing both the safe date for committing orders to customers and for determining the release date for orders to production. Looking at the top right of the figure, we see that a full production buffer is placed at the safe date committed to the customer. Continuing at the top, we notice that we back off one full production buffer length in time to determine the release date for the order. In the lower segments of the figure, we deal with the elements that go into the computation of the safe date. (Determine the date where the planned load ends on the CCR and add half of the production buffer.) We see that the production buffer (a liberal estimation of the production time) is logically divided in half. It can be seen that the point in time when processing is to occur on the CCR (depicted using a minimized picture of the CCR planned load profile) is the dividing point (in time) for splitting the production buffer in half. One half of the production buffer time is added to the planned completion date for the last order loaded in the CCR planned load, approximating when this next order will start processing on the CCR. This effectively adds the required time for completion on the CCR and mainly the end of the production process for the new order. Adding this half of the production buffer to the planned CCR date then gives us the safe date for commitment of the new order to the shipping dock. From that same point on the load chart, the other half of the production buffer (in effect the upstream half) is subtracted from the time when processing is to occur on the CCR. It is worth noting that in these calculations, the actual processing time on the CCR (touch time) is ignored in computing the release of raw materials and estimated shipping dates because CCR touch time versus queue time in the production process is usually negligible.

FIGURE 9-3 The timelines for safe date, order release, and buffer placement in S-DBR.

Taking this in another sequence, we see that from the point of order release to processing on the CCR we provide half the production buffer to get the order to the CCR, and the other half of the production buffer to get the order from the CCR to order completion. (Both of these time components are implied in releasing the order one full production buffer time ahead of the safe date.) Taking the time of processing on the CCR plus half the production buffer, we get the safe date for customer commitment.

The rule does not specify the location of the CCR operations within the routing. So, fixing the required time between the planned load and the safe completion date to be half the production buffer for the product requested by the order seems arbitrary. This works well except where the CCR is very close (measured in time) to the material release or shipping in the product structure and in such an extreme case, some other division (of the production buffer other than half before and after the CCR) should be used as will be noted later.

Another question is whether the size of the manufacturing order has a major impact on the size of the production buffer and on the planned load. Suppose the production buffer for a "normal" order size of 50 units is 10 days. If the order is for 200 units, then it seems the processing time at the CCR would take significantly longer than usual, and it would impact the time the order requires from the downstream operations. Wouldn't it?

The rationale behind the rule of using the default buffer for orders with different quantities is to have a simple and straightforward method to monitor the load on the CCR and determine good safe dates. In a world where the data is frequently not accurate (processing times being, many times, just gross assessments) and being exposed to significant uncertainty from both external and internal sources, the only way to manage well is to look for "good enough" planning and execution rules. The time buffers introduced are usually not optimal, but as long as they are good enough, they do the job. A large order takes more processing time, but that time is usually small relative to the wait time, so as long as we do not speak of an ultra-large order, then probably the same production buffer should be used. In a case where an order, whose size is four times the regular size, is dealt with, then it seems reasonable that such an order has a higher chance of penetrating the red. If this occurrence is persistent, then we might introduce an increase to the production buffer when the specific order size is that much higher than the average order size. The same goes for considering the planned load. As it is not a real schedule and there is no guarantee that the CCR will process the order at the "scheduled" time, it is enough to assert that the given safe date is good enough and let BM take the lead by establishing the required priorities. What else do we need to ensure? We need to release the order at such time that it'll have a full production buffer time to go through the whole shop, including the CCR.

Thus, the release time of the order should be according to the planned load minus half of the production buffer for that type of order.16 When the safe dates are given to the clients, then the order gets a full production buffer time to cover operations from the material release until order completion.

This rule applies for the vast majority of the cases. Of course, when the CCR is located at the very end of the routing (a very rare case17 in reality), then we adjust the material release and shipping buffer points by shifting the production buffer upstream by adding to the planned load end point just 20 percent of the production buffer to determine the shipping point. The same adjustment is used when a CCR is at the start of the routing. We release the material at least 20 percent of the production time buffer ahead of the planned load, and add 80 percent of the production time buffer to the planned load to determine the safe date for shipment. Only these extreme cases warrant deviation from the rule.

What Happens When Sales Quotes a Different Due Date Than the Safe Date Given by the Planned Load?

The recommended action is for Sales to quote the standard lead time whenever the safe date is prior to the standard. The problem exists when Sales does not follow the safe-date directive and quotes an earlier date than the safe date. If this case is a rare occurrence and most orders are quoted at safe date or later, then our recommendation is to release the manufacturing order to production one buffer time (production buffer) before the due date. If this practice of quoting due dates earlier than the safe dates is not a rare case, then it is not possible to use the safe-date mechanism at all and the company must revert to the behavior of "we do our best to meet the dates, but sometimes we simply are not able to."

Suppose the safe date is much earlier than the actual shipping date given to the client. Should the release date still be planned load minus half the production buffer? In this case, the actual time buffer is much longer than the production buffer.

The first impulse is to release the order one production buffer time prior to the committed due date. There is, however, one important reason why we should keep the release of the materials at the date the CCR is supposed to process it minus half the production buffer. If we do not release the order at that time, because it is more than the production buffer time until the due date, there is high probability that the CCR would be idle at that time or a little after that. One might argue that having the CCR idle, when it is not an active constraint is not as bad as it might seem. The fact that the date of the current planned load plus half the production buffer is earlier than the standard lead time date means sales are somewhat lower than what we are capable of handling. However, letting the CCR be idle when we do have a firm order at hand means we might lose that capacity if we'd need it in the near future. Let's view an example.

The standard lead time is eight weeks. The production buffer is four weeks. The current planned load is only two weeks. The rule says that the safe date is two weeks (the current planned load) plus half of four weeks (one-half the production buffer). It means the safe date is four weeks from now and the release date is now. However, if Sales quotes eight weeks, the pragmatic question is, "Should we release the order now, or wait for four weeks and then release it (leaving four weeks of production buffer for the order)?"

If we delay the release of the current order and do so to every order that gets the due date of eight weeks from now, then in two weeks the CCR would be idle for a while. If, within two to four weeks, many more orders would come, orders that would push the planned load to be more than six weeks, then due to lack of capacity at that time the quoted time for some orders might be more than eight weeks. Here is the damage-we do not fully utilize the CCR at the off-peak time and later we find we need the lost capacity.

Thus, the recommendation is to release the material based on the planned load minus half the production time buffer even for orders whose due dates are later than the safe dates. This ensures that as long as we have orders to be delivered within the standard lead time, we'll release them at the appropriate time to load the CCR continuously. Of course, there are obvious negatives to starting production earlier than the time truly required for safe on-time completion of that particular order. Nevertheless, given the situation that the shop floor is not fully loaded and wasting the capacity of the CCR might cause significant future damage, we still suggest releasing the order early in these circumstances.

Capacity Reservation