Dennis,

I will let you know, I can generate xer files of my jobs and statistical analysis but not using PRA, a software I do not have. For the statistics we use other method that is not like original PERT that was based on a single critical path nor like Monte Carlo that runs thousand times a single model that recognizes changes in the critical path as activity durations change, would be difficult to compare probabilistic models but I am interested in PRA, it has a few things I like.

I always heard about Monte Carlo running different activity durations, cost variations and even modeling of some correlation between activities but this thing of modeling calendar windows of non-work time is new to me and at first sight makes a lot of sense.

Maybe the non-work time windows for hurricane events are based on some interesting distributions or combination; like zero duration if no hurricane with high probability and a long duration if a hurricane with low probability. Perhaps the probability of a hurricane happening is modeled using a separate probability distribution because at the beginning of hurricane season the probabilities are not the same as during the peak. Two probability distributions?; one for the occurrence [hurricane or no hurricane and when] and another for the intensity [window length]  Definitively you posted an interesting subject.

Have a good week end.

Regards,

Rafael

Dennis,

I will follow your suggestion, in a few days after I finish with a construction estimate I am working, a parking garage for an hospital, similar to another we are already building.

I will concentrate on the differences that will result by using resource leveling and other functionalities. 

Discussion under certain modeling assumptions such as the use of soft-links as to force resource constraints by any "brute-force method" will also have its place if you want to discuss such perspective. 

Brute force methods freeze resource dependencies, therefore as the Monte Carlo simulation run durations that can substantially change the resource dependencies or  "soft-links" then for every run this happens the model is no longer valid, the probabilities are no longer valid and that can be a good warming up for the debate.

Best regards,

Rafael

My post was in direct response to your own comment,

The linkage between P6 and PRA seems to be getting stronger with each update, which only enhances its value.

I thought you understood the implications of using different software and that this was the reason for your statement. Running different lag values, running different resource leveling and other functionalities might lead to different project durations on each and every Monte Carlo iteration, therefore the resulting probability curves would be different, this applies to weather modeling as well [it is just a component of the total].

It makes no sense modeling weather or any other risk using a simulation based on a different software you use to mange your schedule. If you use PRA formerly Pertmaster and control your job using PRA it makes sense but not if using any other software such as MSP, P6, SureTrak or any other. I liked Pertmaster better than P6. If you say P6 is better, that will yield better schedules then you would be accepting PRA and P6 are different models.

Sorry if you did not liked me commenting on your own comments, some people use resource leveling, even some of our contracts require us to use it and explicitly prohibit the use of "soft-links" to force resource constraining.

I had no way to know you do not care about software resource leveling nor about the possibility of lag having a different calendar than predecessor or successor. It is not uncommon in our schedules for activities to have lag relationships in calendar days while predecessor and successor calendars are in work days.

Regards,

Rafael

Dennis,

The linkage between P6 and PRA seems to be getting stronger with each update, which only enhances its value.

It cannot be better said.

I believe the linkage should be 100%. For example P6 has a limited option for lag calendars, is based on an across the board rule for all the activities. I believe there are about 4 options, none allows for an individual assignment for lag calendar. Then my question is, do PRA and P6 handle lag calendars in the same way? How the lag calendar handling is transferred to PRA or maybe there is still no link? If there is no link then you will be running a different model.

Calendar on lags can make a huge difference if calendar difference/exceptions/non-work days differ. Same goes with resource leveling algorithm, they can yield very different results. It makes no sense to model risk on an engine that yields different results than the software you are using to manage your job. Until 100% integrated it will only make sense to use PRA when managing your job using PRA. But PRA falls short of my expectations/needs.

I wish I can have Monte Carlo running with the same CPM engine and functionalities of my software, it uses other algorithm that is not equal, each has it advantages over the other, having both would be great. Would be like the resource leveling, we have several options each with its own prioritization you select but the one that looks for Optimum (meaning shorter duration) looks for best prioritization itself. The developer recognizes best option can be other than shorter ["Optimum"] and provides for your selection of algorithm.

It is funny, we are quoting a job that requires Risk Modeling but only few vendors provide Risk Modeling truly integrated, and Primavera is not one of these.

Regards,

Rafael

Dennis,

Although I do not have P6 or Primavera Risk I believe you explained it very well. It is about a pre-defined way in Primavera Risk to model weather. It is a Monte Carlo Simulation, a perfect fit for these issues. When you go back to P6 it becomes a deterministic schedule.

If P6 makes the creation of the resource easier but is empty then it means it has no effect on the deterministic schedule as expected, this trick must be of value to P6 users who also use Primavera Risk. Guess this means in Primavera Risk you assign the other values that are not transferred from one software to the other.

I have some issues with modeling Monte Carlo using a different resource leveling engine for Monte Carlo than the one used in P6, this will yield different durations for resource leveled jobs; P3, P6 and Pertmaster/Primavera-Risk have their origin on different software developers, they are different.

I believe it is better to simulate weather on the risk model and not on the deterministic schedule using "weather calendars" as they distort the last. No matter what, the impact of a hurricane is not merely about days lost to the hurricane it is also about the time to recover under different scenarios, this I do not know how you do it but seems like an interesting issue.

I use Spider functionality for shift modeling in a way an activity can change crews when it falls in summer weather and another crew when it moves to winter weather, but is based on seasons representing a shift, based on an estimation of expected normal weather conditions for the season, the productivity is deterministic. What you are doing is very interesting, is a different issue.

For the deterministic model it would be convenient to use some critical chain functionality, there is a customized application called Aurora CCPM but being customized for this and other sophisticated needs is of high cost for the average user [~200,000 US$ custom implementation with a few licenses]. I got well impressed with the literature especially with the idea on customization and how it improves P6 resource leveling. Asta PP have CCPM embedded into their application at no extra cost, perhaps not as sophisticated, do not know if can work on multi-project models.

I recall a job that was impacted by Hurricane Georges, the issue was not the work days lost to the hurricane but the time it took the utilities to recover on a "remote" area, about 2 months to get power. Also a basketball court roof was gone, a US pre-engineered building not designed for our hurricanes. The repacement cost was covered not under warranty but under the Builder's Risk insurance. Neither the structures nor windows were damaged as they were designed  for our hurricanes, perhaps some glass panels were gone. Obviously we got EOT but extended overhead was by each party to pay for their own.

Is a pity I cannot afford Primavera Risk because I would like to continue the debate on this issue on the modeling of weather how it should be, using statistical modeling. I will follow your posts in the hope of learning something from you.

On the average every 5 years some hurricane hit us, a major hurricane every 10 years. Still the water by non-hurricane storms can cause some disruption. I wonder how useful this can be on weather events that have such an impact on the schedule that it is no longer valid after the event, Monte Carlo simulation will be running a wrong schedule. Do not missunderstand me, I believe in those models, just that we must recognize their practical limitations.

Best regards,

Rafael

Adding weather to calendars distort the schedule, it is better managed through statistical modeling.

Though in the absence of statistical modeling I believe it can be modeled as a project buffer, a special type of activity. Same as Primavera, Asta PP (as an optional package) and Spider Project (embedded in all versions) provide for statistical modeling.