Cluster and Node Configuration (Tokens)

The average frameloop rate (in Hz) of the COMM and DRAW thread frameloops.

NOTE: This is NOT NECESSARILY the SwapBuffers rate! - Normally, BeFrameRateFreq matches the GPU scan-out rate (VBlank rate, Swap rate, etc.) for the monitor or display device.

However for Serial MultiViewport, BeFrameRateFreq == 2X the GPU scan-out rate.

This because each frameloop produces a 1 of 2 subframes for display (1/2 of the window contents), and SwapBuffers is only called every 'other' frameloop to present both.

With Serial MV, even though the COMM and DRAW rates are very irregular from frameloop to frameloop, the average rate of subframe rendering (aka BeFrameRateFreq) is 2X the GPU scan-out rate.

CPU processor affinity mask for COMM, NET PAGER, SENSORHOST, TRACKER, and GUI threads (if LockCPUs ON).

Enable driver yield mode.

Enable time critical priority settings.

Sets the time control used for the IG.

Replaces the old ICD exe_host_time_control.

Previously enabled a hack to force the number of threads supported by the system CPU(s) to <= 2 (when Hyperthreading and DualCore = OFF and NumIsectThread <= 1).

For a deskside Nova run, don't sleep.

Sleeping plays nice with the host machine, but also makes getting benchmark numbers more difficult.

CPU processor affinity mask for ISECT threads(s) (if LockCPUs ON).

Laptop mode enabled.

CPU processor affinity mask for all process threads (used in "LockCPUs CPUMASK" and "LockCPUs FEPAGER" modes).

Enable frontend low latency mode.

ON: Use LowLatencyFrontEndFreq.

OFF: Use FeFrameRateFreq.

CPU processor affinity mask for MAIN thread (if LockCPUs ON).

Clear out missed frame counter on resets.

Spawn a separate thread for the net pager (always false).

Spawn a separate thread for the disk pager (TRUE on deskside and radar).

Force integral sensorhost to sync with main thread.

If set, spawns a separate thread on FRONTEND / STANDALONE nodes to load and call the head tracker DLL in.

See also MpTrackerSync.

Force FRONTEND head tracker thread to sync with main thread.

Total number of pager clients (not deleted for parser backward compatibility with existing CFG files).

CPU processor affinity mask for PAGER and RESOURCE PAGER thread (if LockCPUs ON).

Indicates timestamps from host are pre-recorded and should be offset.

CPU processor affinity mask for SEA thread(s) (if LockCPUs ON).

Controls the number of threads used to update waves in the uber.

2 indicates wave updates are run in a separate thread, and that two additional threads may be used for updating waves when WaveGenerationMode is not LEGACY.

1 indicates wave updates are run in a separate thread, and that one additional thread may be used for updating waves when WaveGenerationMode is not LEGACY.

0 indicates that waves update in the comms thread.

Defaults to 2.

The time during init before the deadlock prevention starts emptying this process' view of a shared stream FIFO.

The timeout allowed before the deadlock prevention starts emptying this process' view of a shared stream FIFO.

The depth of the substream queues in the shared memory tile manager.

The name used for the shared streams named shared memory.

SPINMODE_OFF, SPINMODE_SPIN, SPINMODE_SLEEP, SPINMODE_YIELD, SPINMODE_SLEEP_ONE

Enables time-killing before SWAP.

OFF: No wait.

SPIN: Busy - wait until (frame period) ms have passed since SOF.

SLEEP: Sleep until (frame period) ms have passed since SOF.

YIELD: Yield the CPU, allowing another thread of equal priority to run.

SLEEP_ONE: Sleep for 1.0ms.

OFF, ON, 1, 0, TRUE, FALSE

OFF, ON, 1, 0, TRUE, FALSE

Must be "true" to avoid resource pager thread eating 100% CPU in an idle loop.

LOCAL, LOCALFIFO, NET

N/A

0 to UINT_MAX, inclusive

Total number of uber clients.

0 to 16, inclusive

COMM thread frame time offset (in msec) since last COMM thread start-of-frame after a Swapbuffers (normally every one).

Disabled if == 0 or node type is STANDALONE, FRONTEND,
ISECT, or PLANTER (only used on cluster subnodes).

If > 0.0, causes COMM thread to stall in updateCommBuffers() after the Uber Prebuild before Uber NetRcv+Extrap and Uber Build until the specified number of msec have elapsed.

NOTE: For Serial Multiviewport, UberSampleTime specifies the wait time for the 1st subframe after a Swap. The wait time for the 2nd is implicitly UberSampleTime + 1000/BeFrameRateFreq, where BeFrameRateFreq is the subframe rate.

0 to 16, inclusive

Time in milliseconds that the front end will wait within its frame to multi-cast the uber data to the back end nodes.

OFF, ON, 1, 0, TRUE, FALSE

Enables use of efficient in-process thread sync (in place of 100% busy-waits).

This can greatly reduce CPU core utilization on cluster-on-deskside systems like nucleus.