Distributed Rendering X64 -> XP32

[Micha_cg]

Hi,

today I try to setup a network rendering, but I don't get it. Some problems:
... the DR settings show no status of the servers and
... find server button is disabled and
... if I start my rendering at the XP64 machine, than the XP32 spawner crash. Also,
... I was looking for the latest spawner, but can not find any spawner at the download section. So I used the version at my HD.


-Micha

[andreas]

Re: Distributed Rendering X64 -> XP32

... I was looking for the latest spawner, but can not find any spawner at the download

i think in the last build, after installing you can find the spawner under programme/asgvis/renderslave

hope this helps

Andy

[Micha_cg]

Re: Distributed Rendering X64 -> XP32

... it works now, Danke Andy.

An other question: if I enable my firewall, it dosn't work anymore. Is one of this services necessary?

[Micha_cg]

Re: Distributed Rendering X64 -> XP32

An other problem: the spawner at the slave tell me, that an texture is loaded, but I get balck buckets? Shouldn't I get an error message like "... problem X with the texture file ..."?

Also I see, that the IM pass use the host engine only, is it right?

[Micha_cg]

Re: Distributed Rendering X64 -> XP32

... I tested a Q6600 with a DualCore Notebook and the rendertimes was single 12min and DR 10min. I ask me, will it help to get more speed if I use a Gigabit LAN?

[joe_b]

Re: Distributed Rendering X64 -> XP32

... the DR settings show no status of the servers and

The status portion is not implemented by V-Ray at the moment

... find server button is disabled and

This is also not implemented in the V-Ray sdk at the time

... if I start my rendering at the XP64 machine, than the XP32 spawner crash.

That isn't a problem for us at the office. So I don't know if thats a specific hardware issue or something else

... I was looking for the latest spawner, but can not find any spawner at the download section. So I used the version at my HD.

As vcube said - Program File\ASGvis\DRSpawner - we need to make a shortcut...

An other question: if I enable my firewall, it dosn't work anymore. Is one of this services necessary?

I don't believe any service is the issue, but there is a specific port that V-Ray uses to communicate. I can't remember it off-hand. I'll have to ask Vlado. Or maybe Travis Serio remembers the answer I gave him about the port number? I know he asked me about that a while ago...maybe I have it in my chat history somewhere...

An other problem: the spawner at the slave tell me, that an texture is loaded, but I get balck buckets? Shouldn't I get an error message like "... problem X with the texture file ..."?

I would have to see the file in question to see what is going on there. I don't know if the black bucket is caused by the texture or if there is something else going on - colormapping, lights incorrect, etc...

Also I see, that the IM pass use the host engine only, is it right?

I was under the impression that all the maps are computed on each machine. I could be mistaken though.

[Micha_cg]

Re: Distributed Rendering X64 -> XP32

Hi Joe, the spawner crash was caused by an old spawner version, the version at the ASGvis dir works fine.

Yes, please let me know, wich port number must be free. I would like to render with active firewall.

Black buckets - I will send you a file and my texture. I wonder me, that the standard rendering works fine and the DR dosn't work and no error message is visible. EDIT: today it works fine. Great!

Missing IM pass - the framebuffer don't show buckets of the slave. Can you see IM slave bucktes at your machine? But I can see, that the slave is busy.

What is the workflow for DR and usage of GI caches? Is the full cache saved at the master and must be manualy copied to the slaves befor rendering with caches?

[dalomar_cg]

Re: Distributed Rendering X64 -> XP32

Check the mode for IM. The single frame does work alright sometimes, but I've found that the Bucket mode works a bit more consistently. Bucket mode a separate map is created for each bucket, and each bucket goes directly on to rendering the image (almost looks like a QMC render process). There isn't one single map for the whole render, which can be an issue if you need it, but I've found that it leads to more consistent results and that its easier to track down issues from one node to another.

EDIT: This can also increase render times because there is less back and forth with information from the nodes to the host. In a typical process for IM+LC there is communication during the LC pass, then for the IM pass every single bucket has to get sent back to the host, and then for the render every bucket has to get passed back. There's also the information that needs to be sent out to all the nodes. With Bucket mode all the information only gets sent out once and back once, which decreases the time/resources needed to calculate all of that. Not to mention decreases the bandwidth used. Each bucket will render "slower" due to extra geometry that needs to be loaded at the bucket edges, but the overall process should be faster.

[Micha_cg]

Re: Distributed Rendering X64 -> XP32

Thanks Damian, I will test it.

The slave bucktes problem seems to be solved - at an other rendering I have seen IM slave buckets, but very late. It seems to be, the master start therendering and the slave need some time to start.

[Micha_cg]

Re: Distributed Rendering X64 -> XP32

... I tested a Q6600 with a DualCore Notebook and the rendertimes was single 12min and DR 10min. I ask me, will it help to get more speed if I use a Gigabit LAN?

I test the notebook and get 600 cinebench points, the master engine is at ~1600 points. So, the notebook power is ~38% of the master. But the speed increase per additional notebook slave is only ~17%. So, it seems to be, the active render power of the notebook is only ~50% or 300 cinebench points. DR seems to waste much power for communication.
So, the fastest engine should be the master and it bring not much speed increase to use additional slow engines.

I suppose so, each GI method has an other yield of the DR CPU power. So, could it be, that a PPT rendering profit more by DR than LC+IM?

PS: I'm from Monday to Fryday not here - short vacation.

[dalomar_cg]

Re: Distributed Rendering X64 -> XP32

Have a good vaction Micha...I guess you'll see this when you get back.

So here's the explanation of why your notebook "seams" to be running at half power. With DR there is a certain amount of resources (in this case resources=time) that are used to take care of all of the leg work now required to send off the scene to any DR slaves. That means that the host computer needs to preprocess more buckets at a given time to find out what geometry needs to be loaded/unloaded. All that information must then be sent to the slaves, and when all that information comes back that needs to be processed again in order to be added to the other information in the render. Not to mention that all of this information has to be sent over a network and there are considerations of how much that slows things down. In the end its not that your laptop is not running optimally, its that there is allot of extra processes that need to happen for DR, so its a combination of your host machine having to process all of that and the laptop having momentary periods of rest in between buckets.

[Micha_cg]

Re: Distributed Rendering X64 -> XP32

I'm back, the short vacation was fibe and now I have got the parts for my new slave machine. ... some minutes befor I got my first render test working:

single master (Q6600@3GHz) -> 70 seconds
master + slave (Q6600@2.4GHz) -> 40seconds

So, the DR need 57% percent of the non DR time. The master give me 1600 cinebench points, the slave 1260 (78% power of the master machine). I'm quite happy: I suppose so, if I would get the same calculation power at the slave, I would get approx. the half render time per DR. Great.
The test based on a scene without textures and physical sky. The LAN conection is a 1G bit LAN, befor it was a standard LAN.


... but I got a problem: the buckets of master and slave show a different brightness. What's the reason for it and how can I fix it? I have seen this problem at the forum some weeks befor, but can not remember the solution. Any suggestion?

[merlik585]

Re: Distributed Rendering X64 -> XP32

... but I got a problem: the buckets of master and slave show a different brightness. What's the reason for it and how can I fix it? I have seen this problem at the forum some weeks befor, but can not remember the solution. Any suggestion?

I have the same problem,and have not found a solution yet.

Renee

[Micha_cg]

Re: Distributed Rendering X64 -> XP32

Today I open my DR test scene and ... it works fine.

[Micha_cg]

Re: Distributed Rendering X64 -> XP32

... it works now, Danke Andy.

An other question: if I enable my firewall, it dosn't work anymore. Is one of this services necessary?
...

Sorry, I have overseen that my Rhino3.exe was blocked for communication. :-[
Works fine now, but the bightness problem is back. I see know, that the physical sky is the problem: it seems to be, that the slave use a 180° turned sky environment - the master buckets show the sun from an other dir than the slave buckets. Also a reopen of the scene dosn't help.