The idea of this solution is not new. It’s actually pretty old. In literature, the technical design of this method can be found in publications from the year 2006. Again and again, however, research advises that it is currently not feasible for reasons of performance. This opinion seemed like a good reason to try implementing it with HANA.
So for us, it was primarily a question of determining whether a process that to date, for performance reasons, was either infeasible or only had a very limited level of feasibility, could be brought into the realm of feasibility with HANA.
As far as I can see, the evidence shows that we have succeeded.
What’s it all about?
Imagine the following scenario. A Partner A gets a rush order and in his turn passed out rush jobs to two of its 1st-tier suppliers with the request for appropriate delivery promises. These 1st-tier suppliers in their turn contact their suppliers to order various purchased parts or raw materials and for their own planning, they in turn request delivery promises.
That was the starting point and we were not the first to plough through this scenario. Of course we have made some idealizing assumptions for our lab test and the PP scenario is not defined in depth.
Why is this so difficult? Let’s say all parties are SAP users, which we of course like to do and would have to change their production planning to fulfil this order. This means in the technical sense, they would need to initiate a planning run, the duration of which is often in the region of one hour.
In our model, the 1st-tier suppliers make quasi proposals for a delivery date, which the 2nds could either refuse or confirm. One of the simplifications that we have adopted for our laboratory experiment; was a stock of 0 for all materials. All materials had to be first produced by the supplier. We have also tried to cover a wide range of material types, which went from mass-produced items such as screws to piece work. The 2nd-tiers also had very different workloads, production times, etc.
The 1st-tier suppliers only had information on which part they could get from which supplier and the only information they received in response was the confirmation or rejection of the request date. This is of course not the most elegant way to solve this problem, since 2nd tier could also have advised its earliest possible date, thus making further inquiries superfluous.
That is exactly what we did not want. The agents were to contact each other under maximum stress, this was the goal. Many planning runs and many read operations on large databases.
The agents were able to selectively operate from a ”normal” database, or from a HANA-column-store using the same data.
The measure by which the experiment was judged was the time that was required for the coordination of all suppliers. The working hypothesis for the experiment was a correlation between the degree of capacity utilization and the planning effort for the 2nd-tiers and the drifting apart of maturities between HANA and non-HANA systems for coordination between these APO-SCM agents.
This hypothesis was confirmed and the observable effect was in the order of factor 200 for the HANA system.
A few words about our team.
We met by chance in developer forums and quickly found that we all deal with areas of current AI research …… and with HANA.
We all have access to relevant armoured systems and have a high standard of networking both technically and intellectually. We sit in Boston (MIT), Bombay and Bielefeld. The youngest of us is 21, the oldest 52 (that’s me) and I hope for many more joint experiments in HANA.
– by Mario Luetkebohle, Consultant, itelligence AG –