Causes and Management of Interrupt Latencies

Interrupt latency is of particular concern to real-time systems developers. This topic examines its causes and RTX techniques for managing them.

Initialization Latencies

During the Windows boot up and shutdown, some systems exhibit latencies due to device driver or hardware initialization. Use the srtm.rtss sample program during boot time to measure these latencies for your particular system.

Software Causes

Software causes of interrupt latencies include:

PIC (programmable interrupt controller) or APIC (advanced programmable interrupt controller)-level masking of interrupts. The Windows kernel and drivers change interrupt masking via HAL IRQL manipulation calls. Drivers routinely mask interrupts for several µs.
Processor-level masking of all interrupts. The Windows kernel, HAL, and special system drivers may mask interrupts for up to 100ms.

Interrupt-processing overhead.

Hardware Causes

Hardware causes of interrupt latencies include:

Bus "hijacking" by peripheral devices. For example, a video card may stall the CPU's attempt to read the card's I/O space register.
Burst DMA by SCSI controllers.
Cache dirtying by Windows and applications.
Most systems, particularly portables, can go to a low-power state for peripherals after a configurable time-out is set, and "wake up" with a delay that is intolerable to a real-time application.
Battery management features use of SMI (System Management Interrupt) causes delays of up to hundreds of microseconds.
Intel's SpeedStep technology can change processor speed, depending on load factors.

RTX Management of Interrupt Latencies

RTX manages interrupt latencies in the following ways:

RTX eliminates PIC or APIC-level interrupt disabling by isolating interrupts (through the RTX HAL extension).
Windows and their drivers cannot disable RTSS interrupts.
In addition, all Windows interrupts are masked when RTSS is running.

Hibernation

RTX does not support hibernation or hybrid sleep. Before installing RTX on a system, hibernation and hybrid sleep must be disabled to guarantee real-time performance. To facilitate this process, the RTX installer creates and enables a new power state on your machine, called “RTX-Recommended,” with the proper settings, described below.

Before creating and enabling the RTX-Recommended power state, the installer clones your machine’s existing power state. During uninstallation, the RTX installer will reactivate the old power state and delete the RTX-Recommended state.

RTX-Recommended Basic Settings

Turn off display = Never

Put the computer to sleep = Never

RTX-Recommended Advanced Settings

Category	Settings
Hard disk	Turn off hard disk after = Never
Sleep	Sleep After = Never Allow hybrid sleep = Off Hibernate after = Never
PCI Express	Link State Power Management = Off
Processor power management	Minimum processor state = 100% Maximum processor state = 100%

On laptops, both AC and DC behaviors are set for all the above.

To view a power state’s settings:

From the Start menu, choose Control Panel.
Click System and Maintenance, then click Power Options.
Click Change plan settings for the power state you want to view. You see the “Change settings for the plan: PlanName” window. This is the location of the basic settings.
To view advanced settings for the power state, click Change advanced power settings.

Additional Latency Considerations

Other variables that can influence RTX performance include Windows device driver implementations, OpenGL usage, and legacy USB support. To minimize interrupt latencies, the OEM should disable the following:

OpenGL applications
Legacy USB support from the BIOS settings.