2 Requirements and Goals

Multi-rendering is the means, not the end. The concept grew from a set of requirements for the current generation Silicon Graphics X server and the operating system and hardware capabilities at our disposal. The requirements are:

• Support both indirect and direct OpenGL rendering. Indirect rendering is rendering done by the X server on behalf of the client. Direct rendering is done by the client directly manipulating the hardware. Direct rendering is optionally supported by OpenGL implementations.

• Support the X Consortium's PEX 5.1 extension.

• The same OpenGL rendering library used for direct rendering OpenGL programs ``outside'' the X server should be used inside the X server.

• The solution must not diminish X server interactivity.

• Do not compromise X server reliability.

• Limit changes to the X Consortium's sample server organization to keep manageable the task of integrating future X Consortium releases and bug fixes.

• Do not require changes to extensions not using multi-rendering.

• Support dynamic loading of the OpenGL and PEX extensions to reduce overhead when not using these extensions.

• Minimize synchronization and locking overhead when using multi-rendering.

• Add no measurable overhead when not using multi-rendering.

• Render PEX using OpenGL to require only a single 3D device-dependent component.

• Provide a single mechanism to work across SGI's full line of graphics hardware.

• Minimize use of expensive operating systems resources. In particular, processes and rendering nodes should be allocated judiciously.

• Meet the schedule for release.

Other sets of requirements, goals, and system capabilities are likely to lead to other approaches. For example, IBM's OpenGL implementation [12] has a similar set of requirements but does not adopt multi-rendering.