Query about "Speculative Execution" Bug and RHEL Hosted In Virtualized Environments

Latest response

If the virtualization layer (say, AWS, Azure, etc.) has been remediated and the CSP is eating the 30% hit at the hardware level, are guests riding that CSP hardware going to take a further 30% hit against their CPU allocation? Or, does the remediation set some kind of flag that gets passed up from the virtualization layer so that the fix-code sees "already safe - nothing for me to do at this layer"?

Responses

For the meltdown issue (primarily intel CPU's), the performance hit is from flushing the TLB (translation lookaside buffer) on every userspace/kernel transition. Userspace code spends a lot more time walking page tables to convert VM addresses into physical addresses as a result. You only take that hit once. Most workloads won't see 30% slowdown, but almost all workloads will see some slowdown. In the absence of a benchmark, assume something around 10%. Redhat has a more detailed analysis at https://access.redhat.com/articles/3307751.

Close

Welcome! Check out the Getting Started with Red Hat page for quick tours and guides for common tasks.