Newbie having trouble with stability & autosaves

[S.o.G.]

Hi, my name is Michael. I found darwinbots because I was interested in checking out genetic programming.

I have an amd dual core machine, and I am running Windows XP64.

I have a lot of crashes with 2.37.6 (it crashes about once every 1.5-2 hours). Usually overflow errors, once a complete machine crash.

Also I am unable to open most autosave files. The program either hangs or crashes. The few that do open have no graphic details (but this is fixed by turning all the graphics options on and back on again).

Worse is that a simulation that was opened from autosave runs slower and reports more total objects than it should.

Is it possible all these problems are related to running on XP 64? Is anyone else using darwinbots on XP 64?

Also, how fast should it be (for example on 16000 x 12000 I get speeds from 8 cps to 40cps depending on population, with fast mode & flicker mode turned on. Darwinbots only ever uses 50% cpu -- is there a way to make it use more? The load seems to distribute over both cores, so I don't think it is because it is bound to a single core.

[EricL]

Hi Michael.  I'll never remember that, so I'll also say hi SoG.

Welcome aboard.  I have machine envy.

I can't help you too much with 2.37.6 though others can.  I've been working on the 2.4X code fork.  You might want to try to see whether your issues repro in 2.42.5.  If so, I can help track them down.

That said, I don't think the issues are related to XP64, at least the overflows.  Crashing the OS though... now that has to be a bug in the OS or the VB runtime.  DB is all 32 bit code written in VB6 and I pretty sure Microsoft's 32 bit emulation is solid though there could be bugs using the 32 bit VB6 runtime on 64 bit platforms.  But my bet for most of your issues in bugs in the DB program itself.  Give 2.4X a shot and tell me if you see the same.

Unfortuantly, VB6 programs are single threaded so even though your OS will schedule the thread across both your CPUs, the program is entirely serial and thus will never (until the C++ port comes along) be able to take advantage of more than one cpu at a time.  When you look at it in perfmon, it will look like both cpus are being used at 50% but what is really happening is that the program is using one at 100% for 10ms then the other for 10ms, etc. as the OS schedules it.  You can run two instances side by side however and use internet sharing to have them share bots between them.  This would also give you some topological isolation.  I want to make this super easy in coming versions....

cycles/sec will be very dependent on number of bots and to a lesser extent on the length of their genomes, the physics of your sim and the bot's morphological characteristics.  Sims with 1000 bots may run at X, but the same sim with bots that create a lot of ties or shots will run slower or if you have fulid resistance turned on, etc..  The program has to do math for evyerthing, shot collision detection, tie drag calculations, etc. so perf is gernerally proportional to the number of bots, shots and ties in a non-linear way.  The numbers you site are typical for smaller sims.
 
Again, welcome.

[Numsgil]

Wow, I think that's the most powered computer ever attempting to run Darwinbots.
 
 It could have to do with the VB run time libraries.  While crashes are relatively rare for me on different windows OSes, the program does behave differently.  Deffinately not a what-you-see-is-what-you-get environment.
 
 Eric, that sounds really cool running two instances and internet sharing between them

[S.o.G.]

I started a new sim, and now darwinbots has run for about 4 million cycles without a crash!

The difference was a LARGER field, so that the bots are drawn as colored blobs & not rings.

I also discovered that such a sim runs much much faster -- even with display turned off. I'm not sure why it would be faster with display turned off. But it is -- about twice as fast.

I also discovered that walls really slow things down. Why is this? I know that partly it's when there is an organism that likes to indefinetly attempt to tie-feed to them, thus creating a lot of ties.

Anyway, now I'm getting speeds as high as 60-70 cps with display turned off, when there is a wall & about 100 bots & 100 veggies. If I run without any walls, it can get as high as 100cps.

Cool.

I'm having a lot of fun playing with this. I haven't tried writing any bots yet, just trying to set up environments that result in evolution.

Huge thanks to everyone who has worked on this program! It's a lot of fun.

[S.o.G.]

You can run two instances side by side however and use internet sharing to have them share bots between them.  This would also give you some topological isolation.  I want to make this super easy in coming versions....

That sounds like a good idea...so, I have to run an ftp server on localhost, yes? I have cygwin installed so I'll probably use that as I'm more familiar with unix than windows (I'm a mac guy originally).

Thanks for creating such a fun program!

[Numsgil]

Darwinbots has really been optimized, when it has been optimized, for larger sims (like in excess of 1000 bots).
 
 Walls are slow because they're treated as a special kind of bot, which increases the complexity way too much.  Walls should probably be defined as lines or something like that.

[EricL]

so, I have to run an ftp server on localhost, yes?

Correct.  I worked on this several versions ago and it was working to a certain extent, but I did not fully test it.  It may be buggy.  Let me know if you have any issues.  I'll probably focus on this post 2.42.6.