4 ms·
Interesting! After reading it, I realized I was actually hoping for information at a lower level than VM for memory pressure. Actually finding live, actionable
by SomeHacker44 8y ago
Interesting!
After reading it, I realized I was actually hoping for information at a lower level than VM for memory pressure. Actually finding live, actionable information about DRAM bandwidth usage, delays caused by the hardware system of caches including TLBs, L1/2/3 caches and main memory contention, etc. I have not found that existing tools are insufficient in monitoring/dealing with VM swapping - OTOH I usually seek to keep that at zero and leave a little swap just to allow for some chance of alerting and recovery before OOM killer kicks in.
- grandmczeb 8y agoCan you give a little more detail about the kind of information you’d like from the hardware?
- Filligree 8y agoHow much time does each CPU core spend running/waiting for L1/L2/L3/DRAM? How often are those stalls due to cross-core contention for the same bit of memory? Which execution units are in use? What's the limiting factor on throughout? That's what I can think of in the first two minutes, anyhow. It all comes down to the last one.
- jakeogh 8y agoProb too easy to ask here, but can I get a list of pending syscalls sorted by age?