Friday 28th October 2022
So today we are learning about IT systems. CPU - central processing unit. RAM - random access memory. ROM - read only memory. GPU - graphics processing unit. Core is the brain of a CPU. Network. Distributed rendering - render farm - headless nodes ($millions). Floating licenses on one dedicated license server. Assets - (scenes, files, plates, renders, E.c.t.) stored on one server - strict version control but requires large and fast server ($millions). NAS (network attached storage) vs SAN (storage area network). Slow continuous speeds; SAN has very fast continuous read/write speeds (e.g video) but is complex to set up and manage. Limited access to this server. Raid - redundance so if anything happens there's a copy of it somewhere else. Backup systems for copy's of assets. Data management - unstructured date - no predefined data model. Structured data - highly organised and searchable data in a relative database - strict naming conventions, version control and metadata. Database (e.g. shotgrid) a necessity. However, this requires scripts galore to provide necessary tools for populating databases and using/managing it. Hooks in every app to register saved, published, rendered files with the database. Render management - similar to data management, render submissions need to be tracked in the database but the render farm needs to be used efficiently 24/7. The efficiency is targeted at 97% - 98%. It is necessary to separate 2D and 3D tasks and associate suitable hardware as well as allocate resources. Priorities need to be set regarding the various jobs submitted. It also uses scene description (e.g. .rib, .ass, .ifd) instead of loading the whole scene (too slow and greedy). Frame chucking allows several frames to be assigned to a core. This is much easier to do with CPU rendering since GPU rendering is volatile.
Comments