MapForums

Also seeing lockups on dual-core, even after setting processor affinity on the application and mappoint to pin each to the same CPU. The lockup usually occurs on a GetLocation() call. When this occurs, my application message loop is hung inside the mappoint code never to return, hence mappoint must be killed via task manager.

I have setup my test to repeat the same drawing calls over and over again-- and it may run ok for several minutes, or it may lock up within just a few seconds. I have also throttled back the application requests to the point where there is at least 20 milliseconds between any two requests to the map.

Of course, when I disable dual core in the BIOS the problem goes away. MP02/04/06 all exhibit the same behavior. This is a bummer.

I've just been doing some proof of concept stuff with .NET 2 (C#), and I'm not seeing this slow down, although I'm using MapPoint as an external application rather than as a control. This may explain the difference?

Initially I was running with MapPoint 2002 due to these objects being registered correctly.
Each unit of work involved FindPlaces() and route calculation.

On a Core 2 Duo (2400 I think), this ran 4898 units of work in 1hr 13min 41 sec (real time). Cpu usage was about 90% - ie. it was using more than one cpu.

With MapPoint 2006, 4355 units of work ran in 1hr 3min 14sec (real time), but cpu usage dropped to about 70%. Ie. still more than one cpu, but it appears to be running more efficiently in terms of pure cpu cycles.

Richard

__________________
Winwaed Software Technology LLC
http://www.winwaed.com
See http://www.mapping-tools.com for MapPoint Tools
Pre-Order MapPoint 2009 today: http://www.mapping-tools.com/mappoint2009

I think the performance issue (at least the one I've seen) is related to drawing shapes on the map. I'm not using FindPlaces or routes.

The lockup usually occurs inside GetLocation().

Just wondering if anyone ever found a fix for this? Am using VB5 with the MP2006 activex. It appears to run about 6x slower on a Dell D430 duo-core than on a Dell D800 non-duo-core.

In my case shutting down the 2nd core did NOT fix anything.
Initial placing of shapes:
D800 = 2.2 seconds
D430 = 13.06 seconds (dual core ON and "SpeedStep Enabled")
D430 = 18.328 seconds (dual core ON but "SpeedStep" Disabled)
D430 = 25.72 seconds(dual core OFF but "SpeedStep" Enabled)

Moving Shape times: Dual core enabled and SpeedStep enabled on D430
D800 Add Shape = 20 ms
D800 Delete Shape = 5 ms
D430 Add Shape = 110 ms
D430 Delete Shape = 12 ms

Also, I was at the open source developers conference last week and
spoke with the parallel processing guru at Intel about this and his
response was that Microsoft(writer of mappoint) messed up in their
thread management causing the processor(s) to have to flush the cache
when doing certain operations. In our case its the "getlocation"
mappoint method used when creating shapes.

Any comments?
Thanks,
Bartj

I have a message into someone at Microsoft, even provided some source code about a month ago, but still havent heard anything back. Apparently they are very busy working on the upcoming MP release. I had half-hoped the new version would fix the problem, but my guess is the problem is too difficult to fix in the short run, or they wont even look at it until after the new version is ready.

I will post if/when I get some kind of response.