> Without careful priority tuning Which might be as simple as marking the throughput tasks as SCHED_BATCH. It'll make their latency even worse but preserve thr

> Without careful priority tuning Which might be as simple as marking the throughput tasks as SCHED_BATCH. It'll make their latency even worse but preserve throughput, give priority to the shell and reduce context switches on an oversubscribed system even further. Or you could use taskset/cgroups to reserve 1 hyperthread for interactive things.