When using cpu with hyper-threading, whether disable hyper-threading is a alway right policy?

In the paper "haracterizing Microservice Dependency and Performance: Alibaba Trace Analysis" from SoCC'2021, Figure 14 shows that "When the host CPU utilization exceeds 40% (80%), the RT of a microservice can be degraded by more than 20% (30%) in average." Does it mean hyber-threading generate interference on others?

If someone want to use Microservice in a SMT CPU, another virtual core must stay free for QoS?

Start by reading the cited SoCC'2021 paper, especially Figure 14, and compare its CPU-utilization findings with the issue's questions about SMT and virtual cores. A useful outcome would be a documented, evidence-based explanation of whether disabling hyper-threading is required for microservice QoS and under what conditions.

Written by the indexing model from the issue text.