Which is okay but I just need it to run in 1-2 hours regardless =) what do I need. Can Epicor give me an external MRP so we can line it up on 100 NVIDIA GPU’s ?
We have 10-15 Companies in our Database each one has MRP and each one runs long and when everyone wants to regen at 10pm it must be done before AM shift otherwise record locks.
Also what if they all were 3 shift shops… No Downtime possible but once a month. We have been asking this question Epicor several times without a proper response… One rep replied “MRP running 9hrs is normal” thats one crappy MRP product, need to leverage Multi-Threading properly. =). What if we had 1 million jobs a day that needed to be generated.
We’re at 140GB, going on 2.5 years live in Epicor. Discovered db purge, and starting to test that.
It seems when Epicor talks about scalability, they always bring up # of concurrent users. Our concern is performance slowdowns from millions of parts, customers, orders, shipments, rather than # of users.
There’s quite a few Insights courses this year about the PDT, tuning, and BPM tuning which I hope to learn from.
MRP is and always been a medium horizon tool, not a daily execution tool. Many companies, including every single one I’ve worked at, ran MRP daily.
Now, the 100 GPUs idea, I believe that’s the next logical step for ERP companies - sort of. What I hope to see in the market is ability to leverage on-demand computing power in the cloud. I don’t want to buy 100 GPUs (or VMs) that will sit idle for most of the day when I can rent as much capacity I need. However, the software needs to be designed to do so. Starting with V8, Epicor split the UI from the Business Logic. Having Business Objects gave them the ability to change the technology stack. On the new stack, they added the REST interface which is more loosely coupled. It should be possible to architect portions of these items to be less coupled with the system, like MRP. The benefit would be a single MRP engine that runs across all of their product line and one that can be run in the cloud with all the power on tap.
We have a relatively large database, (350Gig). Typically, we have 5000-6000 open Jobs, counting unfirmed, and about 4500 PO suggestions or open PO’s. We are a 3 shift company and only run regen MRP once a week, at midnight on Saturday. It takes about 4.5 hours to complete. We run 3 schedulers and two processors and that was based solely on experimentation. Any more schedulers and we found there were locks that crashed the run. We run a net change every day of the week, other than on Saturday, and that completes in about 35 minutes.
Like others have suggested, we limit the horizon and we do not use finite scheduling.