Server configuration for Ben using

[Bablooos]

Hi,
I'm trying to use Ben for online playing but unfortunately he plays very slow and unstable. What is primary points to speed him up? I use virtual environment, so maybe I don't give enough CPU to him.
Maybe you can give some recomendations about tuning Ben server? What is important? What configuration is necessary to work for example with 10 requests per second? 100?

[ThorvaldAagaard]

I need to know more about your setup to give you some recommendations.

On my test-server it runs fine, but mostly as single user

On JBridge BEN is playing tournaments and I think up to 10 robots

I would probably not run more than 4 robots (a taBle) on the same server.

Tensorflow has some problems when multiple processes access the same model, so a lot of requests are serialized.

For the fastest version try speed.conf

[Bablooos]

I'm not using something special, running gameapi.py. More or less standard CPU only VM with 16cpu on Ubuntu. But sometimes requests take 5/7s or even up to ±15s (bidding). Maybe it's ok for the CPU, I'm not sure. Just asked about this. If it is ok then it looks like I should use some GPU hosting? Do you have the same timing for bidding on CPU? чт, 9 окт. 2025 г. в 20:07, Thorvald Aagaard ***@***.***>:

…

*ThorvaldAagaard* left a comment (lorserker/ben#175) <#175 (comment)> I need to know more about your setup to give you some recommendations. On my test-server it runs fine, but mostly as single user On JBridge BEN is playing tournaments and I think up to 10 robots I would probably not run more than 4 robots (a taBle) on the same server. Tensorflow has some problems when multiple processes access the same model, so a lot of requests are serialized. For the fastest version try speed.conf — Reply to this email directly, view it on GitHub <#175 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABLSC3IXTY4TJ6K5MMUJ47D3W2I63AVCNFSM6AAAAACIX4AROSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTGOBWG43DANRYG4> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[ThorvaldAagaard]

The current configuration is playing a board in about 1 min for me.
I am running the Api on an old server and not using GPU and with 16 GB of RAM

You are welcome to try the same request at http://ben.aalborgdata.dk:8080/api to see if you get same response times.

I am running on Windows and haven't looked at Ubuntu as server.

[Bablooos]

As I understood reading docs problem is not from Tensorflow etc but from trying to use something like DDS. Am I correct? For fastest play I have to cancel this at all. Since I am still not clear with speed.configuration, what do you think is possible to add to gameapi.py requests parameters like "use only AI for answer" and/or "maximum answer time" ? Or should I make different instances of the service with different configurations? вс, 12 окт. 2025 г. в 12:11, Thorvald Aagaard ***@***.***>:

…

*ThorvaldAagaard* left a comment (lorserker/ben#175) <#175 (comment)> The current configuration is playing a board in about 1 min for me. I am running the Api on an old server and not using GPU and with 16 GB of RAM You are welcome to try the same request at http://ben.aalborgdata.dk:8080/api to see if you get same response times. I am running on Windows and haven't looked at Ubuntu as server. — Reply to this email directly, view it on GitHub <#175 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABLSC3MIHNW5ANODURCSJCD3XILMLAVCNFSM6AAAAACIX4AROSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTGOJUGA4TCMJXGU> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[ThorvaldAagaard]

Using speed.conf should disable DDS

So run python gameapi.py --conf config/Robots/speed_TF2.conf