GeoProcessing Pack Routing performance concerns

[JamesMcElroy]

We're seeing some concerning performance issues related to routing using local data via the GeoProcessing tools.  Thanks to Ralf we were able to get a functional gp model to create a gpk for routing.  I tried creating two packs, one using just the data for Arvada, CO, and one using the data for 10 states (Western US + Alaska).

Inside ArcMap creating a route using Arvada's data took 3 seconds.  Using the gpk via Runtime took 4 seconds (this is for a very short route, maybe 2 miles total).  This is a long time for a short route.

I saw the same results using the 10-state pack as the Arvada pack.  That is good because it means performance is the same irregardless of data size after startup.

After that it seems to be almost a linear progression of distance to time.  A route from Longmont, CO to Arvada, CO took 11 seconds to build, and a route from Laramie, WY to Arvada, CO took 17 seconds to build.

This is much slower than the routing engine in MapObjects, and even Bing web services takes less than a second for the same route.

It seems like there are serious performance concerns for doing routing via the GeoProcessing tasks.

Thanks,

James

[MiriRevivo]

We're also using a gpk built from a model in order to perform routing operations through WPF code. We've seen that only the first route per session takes longer (about 7-8 seconds), and the rest of the routes, long routes as well as short ones, take about 3 seconds.

Naturally, we only initialize the geoprocessing service and the geoprocessor object only once per session.

When you run the model you've built in ArcMap on the Network Dataset and try to perform the same route operation - is the performance any better?