Some #Spectre/#Meltdown Intel microcode update observations:

At work we have the same hardware for each of the developers (corporate-mandated Windows laptops). We develop in GNU/Linux VMs. We received them when we were purchased back in April by a larger company.

There's a full compiler stack I wrote for the development of certain systems. It uses Saxon and is therefore really heavy on resources (and syscalls). Lots of inefficiencies. (Don't write your next compiler stack in XSLT. I wouldn't have if I knew it was going to become what it is today.)

I noticed that newer devs' systems, with identical configurations, were taking more than twice the amount of time to compile the same software using this stack. A coworker and myself spent a bit of time debugging and it was eventually found to be the microcode. Once my coworker updated his Intel microcode and BIOS to include the Spectre and Meltdown mitigations, the virtualization performance was terrible---over 100% performance degredation in this case (~5m for a normal build without mitigations, ~12m after mitigations). That's far worse than any benchmarks I've read. Disabling the microcode mitigations restored performance to previous levels.

What a mess.
Follow

@mikegerwitz Yeah, husband works for a cloud provider and the mitigations were just unworkable.

@gamehawk Did they end up using the microcode updates or have to work around it in other ways?

@mikegerwitz I think they had to roll back whatever they did, though they're getting ready to reboot the whole fleet to apply a new (and presumably better-tested for performance) fix. I *think* the first one was the microcode, but I wasn't too clear on the details. (He's a sysadmin, I'm a dev, we don't always speak the same language.)

Sign in to participate in the conversation
Mastodon

Follow friends and discover new ones. Publish anything you want: links, pictures, text, video. This server is run by the main developers of the Mastodon project. Everyone is welcome as long as you follow our code of conduct!