Request-response workloads are common in practice, including key value stores, RPC and other HTTP workloads. Performance of these workload is largely limited by system call and I/O overheads, because the application needs to process a large number of requests arriving at the different TCP connections at the same time (e.g., notified by kevent), issuing a pair of system calls (e.g., read() and write()) to each of them. Even worse, inside the kernel, it is difficult to perform I/O batching across multiple TCP connections. In this talk, we report how FreeBSD performance can be improved using PASTE that uses netmap(4) API atop and below the kernel TCP/IP implementation. We show that PASTE improves throughput and latency by a factor of two; In BSDCan 2018, we introduced concepts of PASTE but without meaningful performance improvements in FreeBSD. Finally, we discuss issues we are facing, for example, soupcall() that is not notified of empty-data mbufs that carry FINs. Implementation and installation documents are available at https://micchie.net/paste/.