For those using bamt on 50+ gpu farm, One of the hurdle is using a quality router or using a proxy for distributing work. Because some pool require as much as 12 active connections per gpu, and most cheap router start to choke past 500 connections, add to this a bittorrent client and web browsing and this figure can easily double.
Aaron had a Proxy in the making but did not release it.
I am using a well supported Hopping proxy and # of connections is ~80% less.
Aaron had a Proxy in the making but did not release it.
I am using a well supported Hopping proxy and # of connections is ~80% less.
What proxy is this? Also, do you just run the proxy on one of the rigs and point the others to it?
yes, the idea was that you run the proxy on one machine and the other nodes just pull work from it. ideally the nodes would just find the proxy via broadcast and so could be zero configuration.
i used the proxy myself for about a month, but it has some issues that i never sorted out. it is written in python and is very high performance, 100s of Ghash should be no problem for it. if someone who knows python wants to take it over, i will give them what i have, but it is not suitable for public consumption at this time and i do not have the time to make it so.
you could also explore using one of the other proxies out there.. when i started the project all the proxies i could find pretty much sucked massively, but that may have changed by now.
Sometimes hacks work. Sockets or named pipes are far superior if you can easily modified the sender. The issue with modifying cgminer is if changes aren't incorporated into mainclient you are stuck w/ patching and rebuilding it after every version and BAMT will be unable to function without the modifed version. A ram drive while a hack provides seemless integration between the two projects.
