on 03-05-2015 7:09 PM
Hello everyone!
We use a SNP profile with strict prioritization settings (tab “Solution Methods”, group “Priority Decomposition”) for prioritization demand by types. It works perfectly from functional perspective, but as for performance it takes 9 hours in comparison to cost-based prioritization which takes just an hour.
Could you please give advice how to improve the performance of the optimizer run? And how to evaluate whether it is a performance issue or the system works just fine?
Here some statistics:
No. of Demands 1.648
No. of Periods 17
No. of Location Products 15.345
No. of Products 6.221
No. of Locations 24
No. of Resources 65
No. of Prod.-Spec. Transportation Lanes 11.903
Number of PPMs/PDS 6.521
Details About Internal Model of Optimizer
No. of Variables 270692174
No. of Discrete Variables 0
No. of Discrete PPMs/PDS 0
No. of Discrete Product-Spec. Transportati 0
No. of Constraints 127426101
Time Decomp. (No. of Periods) 0
Product Decomposition (%) 0
Memory consumption 1.913.776 kb
Thanks in advance.
We did some SNP Optimizer prototyping a year or 2 back, before changing tack to a different way.
some early jobs.
8549 MB 4 hours
Then from memory, we ramped up the memory to about 24 GB to be able to handle our model.
That was using DISCRETE OPTIMIZATION though. TO be able to download blocks. LINEAR was a lot less memory. ie in the selected method above all the tabs in the profile.
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
Hi Vladimir,
Can you please shed some light how you are doing this : "prioritization demand by types"?
How many demand types do you have?
Are all demands within each demand type have the same priority?
How many LP solve run do you have?
Kind Regards,
Omar
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
Hello Omar.
Optimizer can consider 3 types of demand, they are assigned to ATP categories: BM, FC and FA respectively. The only settings in the optimizer profile is "strict prioritization".
Within each type the demand has the same priority and the solution is found with a single run.
Best regards,
Vladimir.
Dear Vladimir,
I understood that you don't want to use Cost base optimization but use "Strict prioritization" option. I would suggest to you to use Time Decomposition (3 Buckets) option. This is suitable only if you are looking for an accurate plan in the short term/medium term and ready to compromise the quality of result for the few buckets in the end.
Example since you have 17 buckets the plan will be same for first 12-14 buckets. Last buckets may not have the same accuracy. I have used Time Decomposition in the past and the results were satisfactory.
Thanks and Regards,
Gajanan Patki
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
Hi Vladimir
Please do specify reasonable Maximum Runtime in Solution Method Tab in Optimizer Profile. I have experience with similar issue when Optimizer user to take close to 7-8 hours to complete without doing much(feasible solution not optimum) for Product deco iteration which used to run for longest amount of time.
Once we reduced the Max runtime from 720 mins to 180 mins, we were able to achieve quicker completion of optimizer and without much impact on quality of solution.
We tracked Service Levels, Number of fulfilled and Unfulfilled demands, Number of Product Deco iterations, Number Optimal solutions and Number of feasible solution by using Optimizer log.(/SAPAPO/SNPOPLOG). If you decide to go with this option, please do track these key indicators for 8-10 runs to know the impact of this change.
Hope this will help
Thank you
Satish Waghmare
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
User | Count |
---|---|
6 | |
4 | |
3 | |
2 | |
1 | |
1 | |
1 | |
1 | |
1 | |
1 |
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.