When the kernel ABI has to change
Posted Jul 9, 2013 19:06 UTC (Tue) by lambda
In reply to: When the kernel ABI has to change
Parent article: When the kernel ABI has to change
I'm curious; how are you write bandwidth limiting processes? I have tried to find a way to do that recently, as I am trying to preserve available write bandwidth for high-priority processes by throttling low-priority processes. But as far as I can tell, in the latest kernels, write bandwidth limits apply only to direct I/O, not normal buffered writes (since buffered writes get accounted to the global flush process, rather than the processes that dirtied the pages in the first place).
Is there currently a way to bandwidth limit buffered writes that I'm missing? Or were you using a non-stock kernel that had one of the proposed patch series applied?
I thought that one of the primary motivations for requiring a single hierarchy was so that mem cgroups and blkio could be managed together, allowing for accounting of dirty pages to push blkio pressure up through the memory management system, rather than encountering the kinds of problems that you're describing.
Am I just confused or missing something?
to post comments)