KPTI + Retpoline Linux Benchmarking On Older Clarksfield / Penryn ThinkPads
The percentage loss from socket activity on the mitigated kernel was roughly the same between the three laptops.
But with the context switching benchmark via stress-ng, the Clarksfield and Penryn hardware was more noticeably impacted.
Likewise, with the System V message passing micro-benchmark, the older hardware saw a more significant drop in performance.
While lastly with the Apache web server benchmark, the percentage difference was about the same across the three tested Lenovo Linux systems.
In some benchmarks the older ThinkPads with pre-PCID processors did end up seeing a larger (percentage) loss, but besides Redis that was mostly isolated to micro-benchmarks like the stress-ng kernel tester. Going into the testing, frankly, I expected a larger performance loss with these around decade old Clarksfield and Penryn laptops relative to the Broadwell reference laptop. In simple CPU workloads like GraphicsMagick and FLAC audio encoding there remained little to no performance impact from the KPTI and Retpoline code mitigating Meltdown and Spectre, but as illustrated over the past week, with workloads having high I/O and poking the kernel the overhead is more noticeable.
Those wishing to dig into the raw numbers of these benchmarks can find all of the data via OpenBenchmarking.org.
If you enjoyed this article consider joining Phoronix Premium to view this site ad-free, multi-page articles on a single page, and other benefits. PayPal or Stripe tips are also graciously accepted. Thanks for your support.