The serverless computing dispatcher receives many requests in user space and distributes them to worker nodes. These requests traverse between kernel space and user space, which increases the dispatching latency. For workflows with extremely short execution times that are invoked thousands of times per second, this latency significantly affects their completion time. Therefore, minimizing dispatching latency is crucial to ensuring the responsiveness of workflows. Drawing on the capabilities of modern programmable network devices, we introduce a serverless computing dispatcher that runs on DPU and programmable switches. At the DPU, the dispatcher uses DPDK to extract request information and add it to the packet header. At the programmable switch, the dispatcher executes the dispatching algorithm and forwards the request to the appropriate function at line speed, avoiding the request in the kernel space and user space transmission time and reducing scheduling latency. Experiments demonstrate that, in high concurrency scenarios, \(\mathbf {L^2SCD}\) significantly reduces the average and P99 latency of single function invoking, improving performance by 15 \(\times \) and 27 \(\times \) , compared to two baselines on average, respectively.