Most of the actions taken during clock ticks are not relevant in the HPC context, where only one task is running on each CPU. For example, synchronizing the hardware time with the kernel time is irrelevant to the application running on that core; this can easily be, and is already, handled by other CPUs. Also, process accounting in the context of high performance computing is irrelevant to the application. However, there is more to each tick than the idle case in the presence of I/O. Linux processes most of the I/O operations in its bottom half handlers, which are used for separating the acknowledgment and processing of hardware interrupts. This reduces latencies for applications that have real-time requirements as the kernel spends less time in uninterruptable mode. Bottom half handlers, a.k.a. softirqs in Linux, run before returning from hardware interrupts or are scheduled just like other user processes in some cases. This means if there is no hardware interrupt received on a core at all (such as the timer interrupt), kernel functionality that handles I/O services are not performed on that core. Furthermore, if there is no I/O device interrupt received at a core, there can still be bottom half work items queued by remote CPUs for the sake of load distribution. For example, Generic Receive Offload and more recent Receive Packet Steering mechanisms in the networking stack map and forward each incoming packet to a remote CPU for the packet’s processing rather than handling them all on the CPU where the interrupt from the network device is received. Naively turning off the ticks would not only reverse this load distribution, but also would prevent the bottom half handlers from running on tickless cores and performing critical tasks.