Do you have those ports (5020, 5040, and 5060) open on your firewall?

Regards,
Eric Winemiller
Digital Carvers Guild
Carrara plug-ins

Port 1 needs to be changed to something else besides 5020. (that little note at the bottom of the Nework Rendering pane on your screenshot) 5025, 5030 etc. The others can stay as shown Also, since I've got some other devices on the network that dynamically grab the 192.... IP addresses I've changed them all to manual "locking" the last digit preventing another device from grabbing the address and messing with the Carrara network. You may not have this issue though. I'm running several nodes with no problems.

I have no problems ... working with 18 render nodes ....

Wow I'd like to see that! Next time you've got them all running, please take a screen shot of all the Ns on the screen and post it here. That would just be cool.

Thanks,
Eric Winemiller
Digital Carvers Guild
Carrara plug-ins

Indeed! :)

I guess the issue would be creating a scene that wouldn't be completely rendered so fast you could actually see all 18... :-D

Which node is faster? :)

LOL, I love it. Regards, Eric

All nodes are pentium IV from 2.4 to 3.2 Ghz, plus three Pentium D 3.2 Ghz ... but faster is the dual xeon :)

I have 12 machines but don't have the grid yet. I'm wondering if it is worth it? I think that two changes would make network rendering better. One, the source computer should go back and try to render any tile that the network has sent to finish. If a slow machine get a tile to render that is very complex, like dynamic hair, then it's almost faster to let the fastest computer to render everything. Two, have two frames rendering at once. The other nodes can move on to the next frame instead of waiting for the last few tiles to finish. Dale

dueyftw, If you've got systems with significantly different performance, reducing the rendering tile size makes a huge, positive difference in accelerating the total render. Mark

You can optimize this issues if you set a small tile size, 64 or less ... you get more network traffic and less scene optimizations, but avoid this render latency if a big tile spent a lot more time to render, and overall network render works best.

Juan J. Gonzalez

Message edited on: 01/20/2006 13:06

I changed the port 1 to 5030 and I got the net rendering working on very simple scenes. Tile size is 16. I get the same error message as above if I render a more complexe scene. And poser scenes render on each different computers but not with the network rendernodes. Thanks for the info Mark

Hi Hawk, Make sure that you've got the latest Carrara patch. There were some Poser/network rendering issues corrrected with it. I just completed a job with mulitple illustrations that had Poser/network rendering. I know it can work for you. ;) Mark

Hi Mark, when you do manual locking, do you put the IP adress of the computer where resides the main application on all the rendernodes or do you put the respective IP # on each node? as in; 192.168.1.100 192.168.1.101 192.168.1.102 Jean-Luc

I've locked each computer/node to it's own existing IP.

Hi Mark, one last question. If the port 1 on the main application is, let's say 5031, do the nodes have to match it or do they have to differ Thanks for your help Jean-Luc

I don't think Carrara cares but mine are all set to the same number.

I have reduce the tile size, and there is a big difference in the computers. Two are p4 2Ghz and the rest are P3 866Mhz. I would still like to see the changes. Why have 12 computers rendering if 11 of them are always waiting Dale

Do this need to be reported to Eovia for a new patch. Hopefully they can provide a patch that would let other PCs work on new tiles while one or more PCs still working on existing tiles? My concern is that if the PCs finished with all tiles in a frame and one or more PCs are still working on existing tile(s). As a engineer programmer, I would want unfinished tile(s) to split workload with PCs to finish it.

GWeb,

I believe CharlesB clarified this at one time. If you have a slow machine still working on a chunk for frame 1, the other machines will start on frame 2, you just won't see the UI update until all the blocks are done for frame 1. The only time they would go idle is if you are rendering a still or you're on the last frame.

Regards,
Eric Winemiller
Digital Carvers Guild
Carrara plug-ins

Yes Eric, i have seen this in some scenes ... i think that nodes are rendering while main app is waiting for last one ... but i am not sure, perhaps waiting nodes render only one tile and then wait ... , so if there is a very slow machine in the farm, it could stop all render process.

Message edited on: 01/20/2006 16:34

I wonder if one of PC fail on a tile or get disconnected or slow, the whole render animation may go direct to the recycle bin. I still think that unfinished tile should be given to all other PCs to finish it.

If you monitor the task and ethernet activity of a node, it seems that they are waiting. Their is no monitoring of the node performance while rendering, so I can't really tell. A nice thing would be to see how much work each node has done. Dale

Yeah it would be cool to have a small odometer built in Carrara

how do you change the tile size? I was just looking and can't find it... ( though it's probably one of those "if it gets any closer it'll bite you" things)

Message edited on: 01/22/2006 05:45

In that Miscellaneous section where you enable network rendering, if you scroll down just a little bit more you should see a slider called Tile Size.

Regards,
Eric Winemiller
Digital Carvers Guild
Carrara plug-ins

yaaa.. i found it :) Thanks

Ok, I finally got it to work. I had to copy the extension folder from the main app to the render nodes and it looks like it's working now. thanks to everyone. Jean-Luc

For animation: is there a way to send complete frames to different CPUs, instead of having the whole farm work on different tiles on one frame, and some CPUs wait until the last tile is done? ::::: Opera :::::

Hello Opera, No there is not. Reducing tile size of the render tiles per frame is actually a faster render scheme than sending a whole frame to a slower machine. A couple of other software packages do it that way I know. However, I'm aware of some dislike for that because it prevents using a render farm on large single images for print applications where a single image can be 50megs. Apparently, the render farm scheme is an either/or solution and Eovia chose this direction.

Thank you for the information. However, I don't see how having all cpu's render the same frame, even with small tiles, could be faster than each CPU having its own frame. At best it would be an even race. Isn't there overhead loading resources to attack a tile? In Poser they are called buckets, and you always go for the biggest possible bucket for this very reason. I accept that Eovia chose this direction. ::::: Opera :::::

I think the key here is he said "sending a frame to a slower machine"

'sending a frame to a slower machine" is not an important problem IMO. So what if a slow machine gets a full frame? The others just keep going, the slow one goes at its own pace. There very worst thing that could possibly happen is: at the end of a long animation run on many machines, say 24 hours, the slow machine coincidently gets a frame RIGHT at the very end, and it has to finish, while the others are done because there are no more frames; you might loose 1/2 hour in the extremem. In the end, the various nodes would have each rendered, say 100-120 frames each, while the slow one might have contributed 45. Meanwhile, all computers rendered at their own pace, with no downtime, and with larger tiles on each "swallow." I am presuming that on a one-cpu render, you'd want the largest possible tile, right? That's the way it is in Poser, you attempt big tiles (called buckets) but turn down the setting if it can't be digested. There is overhead involved in attacking a tile of a given size, and you want to incurr that overhead as few times as possible per frame. The Carrara renderer might not have that same reality. Let me ask this about the Evoia paradigm. What happens if you are rendering an animation for HD or digital film, and it's a two-hour budget per frame for an average node by itself. During network render, 10 machines have at it, all on the same frame. However, during the run, two of the computers crash or freeze. Is there intelligence built into the network render system to detect the partially rendered tile and send it to another machine? If anyone can shed light on that, please respond. Thank you ::::: Opera ::::

I would be worried if a slow machine still have to work on the first tile while other PCs are finishing several other tiles. I would like this tile to be shared with other PCs to finish it before they can move into the next frame. I also would like to see fast PCs to bull the slow PC out of tile task. Many of us would like to see first few frames to determine if it need some fine-tune in the scene or renderer settings.

I was wrong, the nodes do not wait for the next frame. What is happening is the sending computer updates so much faster then all the nodes (2.1 GHz compared to 866 Mhz nodes) that it does almost all of the work. If I add any more load, more Poser objects or more complex sky settings then the nodes might not get to render anything before the source computer finished the frame. This is not a very efficient use of resources. Vue send one frame to each node and the last frame get's sent to all the nodes to finish. Dale

I think a Que of 3 frames would help. And turning off the source computer so I can work and do test renders. Dale

The concept of being able to actually work on the soucre computer is interesting. If it does not slow down the rendering, and running the farm does not bog down Carrara so you can animate/practice render, that's a nifty way to go. I still like sending whole frames to each node. ::::: Opera :::::