Published by KermMartian on April 7, 2008 at 12:58:49 PM CST | Discuss this article (9)

Since my last article about my LAMP neural networking project, Simms, I've made some substantial progress on both the algorithmic basis for the project and the underlying stability of the machines and thread handling that maintains its functionality. First and foremost, I've added two more subprocessing cores, Simms4 and Simms5. Simms4 is a dual-processor server with 2GB of RAM and dual SCSI hard drives; it has previously served several incarnations as 'Renderer' and 'Cemetech1'. I will be putting Simms4 to work performing backup duties for the Simms' cluster's brain (ie MySQL database) and producing daily renders of the full brain architecture. The second machine, Simms5, has identical specs to the existing Simms0, Simms1, Simms2, and Simms3 machines, namely a 1.4GHz P4, 1GB of RAM, and a 40GB hard drive. I have no specific job for it yet, but will put it to work sharing the load from the existing subcores.

On the physical side, I added scripting to automatically synchronize files from the supervisor machine, Simms, to each of the six subcores, and improved the thread handling to decrease the occurrence of race conditions causing duplicate threads to be spawned. On the software side, I added some minor optimizations to the core brain logic module and fixed a few bugs in the aim and secretary modules.

In order to repair some design flaws in the original Simms logic processing database (currently known as the LTM) I am forking what is now known Simms1 into a separate Simms2, which will be tested and brought live over the next few days. I'll post updates when it's live and functional. Among other things, this should vastly improve processing speed, decrease MySQL CPU load on the supervisor, and improve Simms' response time and conversational speed.

Permalink