Tag: irq

irqbalance or set_irq_affinity – interesting cause for a network performance issue.

When it comes to high-performance computing, squeezing every bit of performance out of the system is crucial. One of the critical factors in achieving high performance is reducing system latency. Interrupt requests (IRQs) are a type of signal generated by hardware devices that require attention from the CPU. By default, IRQs can be delivered to any CPU core in a multi-core system. This can lead to cache misses and contention, ultimately leading to increased latency. Fortunately, there are tools available to help manage IRQ affinity and reduce latency, such as irqbalance and set_irq_affinity. https://github.com/majek/ixgbe/blob/master/scripts/set_irq_affinity

irqbalance is a Linux daemon that helps to balance IRQs across multiple CPU cores to reduce latency. By default, irqbalance distributes IRQs across all CPU cores, which is a good starting point. However, depending on the system configuration, it may be necessary to adjust IRQ affinity further to optimize performance.

set_irq_affinity is a script that allows users to set IRQ affinity for specific hardware devices. The script can be used to specify which CPU cores should receive IRQs for a specific hardware device, reducing the chance of cache misses and contention. Set_irq_affinity requires root access to run and must be executed for each device on the system.

To use set_irq_affinity, first, identify the device’s IRQ number using the “cat /proc/interrupts” command. Once the IRQ number has been identified, run the set_irq_affinity script, specifying the IRQ number and the desired CPU cores. For example, to set the IRQ affinity for IRQ 16 to CPU cores 0 and 1, run the following command:

sudo set_irq_affinity.sh 16 0-1
This command tells the kernel to route IRQ 16 to CPU cores 0 and 1.

Keep in mind that setting IRQ affinity is a delicate balance. Setting IRQ affinity for too few CPU cores can result in increased latency due to increased contention for those cores. On the other hand, setting IRQ affinity for too many CPU cores can result in inefficient cache usage and increased latency due to cache misses.

In summary, managing IRQ affinity is an important aspect of optimizing system performance, particularly in high-performance computing environments. The irqbalance daemon can help to balance IRQs across multiple CPU cores, while set_irq_affinity allows users to specify the IRQ affinity for specific hardware devices. By carefully managing IRQ affinity, users can reduce latency and achieve better system performance.