Although it’s been several months, maybe even years, in the making, we’re happy to announce that the Pixel Plow render farm service is now running on our own farm management software platform. We’ve dubbed this software: “Harpoon”
This change brings with it a slew of internal admin features we’re pretty happy about. As well, it removes several limitations with the previous third-party tool we used to use. Future agents will likely take advantage of these limitations being gone. Things like the silly 5,000 frame per-job limit and the 5,000 frame per-user queue limit…those aren’t long for this world. We’ve tested single jobs with over a million frames in the queue without problems.
While Harpoon has been running our GPU farm since its release, letting it control our CPU farm as well has brought noticeable improvements. Now it takes milliseconds between job slice executions, instead of many seconds or even minutes. That means our farm wastes less time when nodes switch from job to job. Harpoon does better at handling error codes as well, so we hope to leverage that into more automation in the future. Our proprietary scheduling algorithm is now at the heart of Harpoon, so its scheduling decisions are based directly on that algorithm. This works much better than trying to coerce a third-party tool into doing what you want.
All told, we’re psyched about Harpoon and what it will enable us to do. We’re looking forward to everyone realizing the benefits of our software developers efforts. Great work guys!