> Bytedance engineers thus took to reworking the kernel SMP code to allow preemption during IPI completion waiting in order to avoid scheduling latency spikes. Bytedance engineers found the existing Linux SMP behavior to be particularly bad for latency-sensitive workloads like DPDK. In their end their optimizations paid off big time with a 90% reduction in P99 latency in testing with DPDK:
I'll need to keep this in mind for the next time I encounter someone complaining that TikTok didn't do anything novel!
> I'll need to keep this in mind for the next time I encounter someone complaining that TikTok didn't do anything novel!
All big hyperscalers do a ton of kernel development to scratch their itches. It's nice when their changes can be upstreamed to solve everyone else's problems, even if it's the very few problems we have in common with people who manage a million servers.
> Bytedance engineers thus took to reworking the kernel SMP code to allow preemption during IPI completion waiting in order to avoid scheduling latency spikes. Bytedance engineers found the existing Linux SMP behavior to be particularly bad for latency-sensitive workloads like DPDK. In their end their optimizations paid off big time with a 90% reduction in P99 latency in testing with DPDK:
I'll need to keep this in mind for the next time I encounter someone complaining that TikTok didn't do anything novel!
> I'll need to keep this in mind for the next time I encounter someone complaining that TikTok didn't do anything novel!
All big hyperscalers do a ton of kernel development to scratch their itches. It's nice when their changes can be upstreamed to solve everyone else's problems, even if it's the very few problems we have in common with people who manage a million servers.