Vehicle Routing Problem - Too many orders?

2854
6
Jump to solution
01-27-2013 04:56 AM
Highlighted
New Contributor II
Hi,

I'm trying to use the Vehicle Routing Problem Solver to calculate 4 routes for trash pickup. There are nearly 2,800 orders. Is this too many for Network Analyst to process? I ran the solver around 12 hours ago and the progress bar says it's at 66 percent - where it's been for at least 2 hours. I am new to this extension and I have no idea whether it can handle this many orders but takes a really long time, or if my computer's processor can't handle it, or if something else is going on. Thanks for any insight!
Tags (2)
Reply
0 Kudos
1 Solution

Accepted Solutions
Highlighted
Regular Contributor III
Trace, this should be absolutely possible to solve with Network Analyst. Working on an older machine I had, I was able to get several hundreds routes with roughly 5,000 orders solved. It is very important to have 64bit OS though. I am using ArcGIS 10.1.

Of course having so many orders would have a significant impact on the scalability of the solver. So if you want to solve a problem with such a large number of orders, please consider do this. Instead of feeding all the 4 routes into the VRP solver at once, you might try use just 2 of them in the solver at once. After that, you can solve the orders that were left unassigned by using the other 2 routes - in the second solve. Naturally, you would get a more viable solution than if you'd divide the number of orders in half and then run the solve by using two separate problems.

View solution in original post

Reply
0 Kudos
6 Replies
Highlighted
New Contributor II
I have only done a couple hundred before not thousands. I am sure it can handle it, but like you said; very time consuming.
Reply
0 Kudos
Highlighted
Regular Contributor III
Trace, this should be absolutely possible to solve with Network Analyst. Working on an older machine I had, I was able to get several hundreds routes with roughly 5,000 orders solved. It is very important to have 64bit OS though. I am using ArcGIS 10.1.

Of course having so many orders would have a significant impact on the scalability of the solver. So if you want to solve a problem with such a large number of orders, please consider do this. Instead of feeding all the 4 routes into the VRP solver at once, you might try use just 2 of them in the solver at once. After that, you can solve the orders that were left unassigned by using the other 2 routes - in the second solve. Naturally, you would get a more viable solution than if you'd divide the number of orders in half and then run the solve by using two separate problems.

View solution in original post

Reply
0 Kudos
Highlighted
New Contributor II
Thanks--I gave up after about 36 hours of processing, but it helps to know that solving this is possible! I forgot to mention that I am using ArcGIS 10.0, SP4 (this version/SP is compatible with another extension we use). The machine is fairly old and definitely not 64 bit. I will try again on a newer, faster machine.
Reply
0 Kudos
Highlighted
by
Occasional Contributor
Hi,

It looks like you are trying to assign about 700 orders to each route. When the number of orders assigned to a route is significant large, the solver will run slow due to the large optimization search space. If you can limit the size of orders per route to 100, you should see big performance improvement.

The best way to solve your problem is to create multiple copies of each route so each route represents the workload of a day or a week, i.e. Route1-Day1, Route2-Day1, Route3-Day1, Route4-Day1, Route1-Day2, Route2-Day2,... Route4-DayN. The solution will be much easier for you to interpret and use.

Thanks.

Anna
Reply
0 Kudos
Highlighted
New Contributor
So any thoughts on whether the VRP of the 9.3 version of ArcGIS/Network Analyst would be able to divide ~10-25k service stops amongs multiple different vehicles either with or without seed points?  Haven't had any luck so far on my 32bit machine but am able to test on a 64bit tomorrow.  Whenever I try to solve I get this NA error message:

Error: Failure while working with internal OD cost matrix problem instance.
Error: Out of memory.


Any thoughts or suggestions would be appreciated.  Thanks!
-Jason
Reply
0 Kudos
Highlighted
by
Occasional Contributor
Hi,

You might want to split orders into multiple study areas using location-allocation solver. Then solve VRP on each study area. Seed points won't solve the memory problem. It usually helps to build solutions with better clusters.

Thanks.

Anna
Reply
0 Kudos