4 ms·
Optimizing Interrupt Handling Performance for Memory Failures
- gbrown_ 7y agoFull title is "Optimizing Interrupt Handling Performance for Memory Failures in Large Scale Data Centers", I had to truncate it to fit within HN's character limit.
- lwb 7y agoReally interesting. The paper felt like about 95% setup and 5% punchline, which I appreciated as someone with very limited knowledge of these low-level details. Their setup with all the different services and auto-remediation sounds extremely complicated and is probably massive overkill for anyone not at FB scale. On the other hand that makes a compelling case for running your infra on the cloud, as you would have to deal with these issues yourself on-prem.
- agapon 7y agoThis is not about interrupts in general but about error reporting interrupts (such as CMCI).
- londons_explore 7y agoIf you have enough memory errors that the cost of handling memory error interrupts is hurting performance, perhaps the correct solution is to drop the hardware in the bin and buy new hardware? Many people would bin the hardware with just one memory checksum failure. If they start happening a million times a day, I would bin the hardware, and I'm pretty stingy when it comes to replacing usable stuff...
- Dylan16807 7y agoBinning at one seems unnecessarily harsh, what if that was a cosmic ray? But I agree with the overall sentiment. If I'm getting more than one bit flip a week, it's time to look at hardware remedies.
- setr 7y agoIf you can deal with memory errors effeciently, then you can presumably relax the reliability requirement, which might map to lower production costs it might even allow for alternative, less stable architectures/hardware, trading off for other things (like size or speed)