 |
Post Ax8SKSbj7DPsumWw8e by [email protected] |
 |
More posts by [email protected] |
|
Post #Ax8KI5QLXbfUneZ32m by [email protected] |
|
0 likes, 1 repeats |
|
|
|
Linux kernel quiz: Why is this program so slow and takes around 50ms to run?Wha… |
|
|
|
Post #Ax8M9WdbBvpiVIoBqC by [email protected] |
|
0 likes, 0 repeats |
|
|
|
Missing `close(sock)` in the loop? Leaking file descriptors is a common perf ki… |
|
|
|
Post #Ax8M9WjyoCwAp5nHmq by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@rgo closing the sockets would be one way to avoid the performance hit, yes; bu… |
|
|
|
Post #Ax8NG4ZI3Ff7QfTlAm by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@jann I'd guess some contention (ping pong?) on a kernel mutex when 2 threa… |
|
|
|
Post #Ax8NG4g1eD39lYd8fg by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@vbabka oh no it's much worse than just mutex contention, and doing the joi… |
|
|
|
Post #Ax8S2h7WQ03oiGvlhY by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@jann I bet there’s some evil rcu_synchronize() in there that causes threads … |
|
|
|
Post #Ax8S7YCD1heuTyRGIS by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@madcoder yes, exactly, you got it! |
|
|
|
Post #Ax8SKSbj7DPsumWw8e by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@jann I don’t like RCUs implementation. And that is one of the big reasons wh… |
|
|
|
Post #Ax8T2oNHXzrdAKvEQ4 by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@madcoder that sounds like SMR readers will still need atomics? to me, the RCU … |
|
|
|
Post #Ax8XvTMjl6GLkfzGIy by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@jann yes that’s correct, memory barriers are needed (the atomics don’t mat… |
|
|
|
Post #Ax8Y20GiL75ZXHHfe4 by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@jann (I wrote the XNU implementation fwiw so I might be very opinionated on th… |
|
|
|
Post #Ax8YfRcCbE3Ndx8r5c by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@madcoder and then Linux works around that with incredibly dodgy stuff like SLA… |
|
|
|
Post #Ax8YlzAEmIvFQnBFrs by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@jann actually there aren’t RMW atomics at all on arm.Entering the critical s… |
|
|
|
Post #Ax8ZOzKo4u6tFmiq2q by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@jann wow that’s a problem with the fd table? That’s crazy. But now I know … |
|
|
|
Post #Ax8ZTOUUUR4IQsLvXc by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@madcoder ah, makes sense. I think RCU is explicitly trying to avoid even that … |
|
|
|
Post #Ax8Znf7aMqdLnqIsr2 by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@madcoder yeah, exactly, something like close(fcntl(0, F_DUPFD, 1024)).Oh, make… |
|
|
|
Post #Ax8a6t0ix3PWNkGEeO by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@jann I’m familiar. We also have something f similar in two ways: quiescent c… |
|
|
|
Post #Ax8aFSHlgTYU6NJqRE by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@jann don’t get me started on all the things that Linux does wrong as a moder… |
|
|
|
Post #Ax8bSHeAeKt6pETjua by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@madcoder there is a patch series for teaching SLUB to allocate from virtual me… |
|
|
|
Post #Ax8c06XyQQvoFGhttA by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@jann since 2023?? Wow. |
|
|
|
Post #Ax8oW8fhGDbdYEMRpA by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@madcoder link? |
|
|
|
Post #Ax8q9Qko439wsf3Cfg by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@idkrn https://www.youtube.com/watch?v=ZXUIFj4nRjk I don't know how to make… |
|
|
|
Post #Ax8qBjoeACCQZywn0i by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@idkrn https://www.youtube.com/live/ZXUIFj4nRjk?si=v6agM1xGlg-yuSLr&t=697 |
|
|
|
Post #Ax8qOCuNp95iHcZmls by [email protected] |
|
0 likes, 1 repeats |
|
|
|
they key slides are: basically a higher working set is slower than a smaller on… |
|
|
|
Post #Ax96v022JmmpAboYLY by [email protected] |
|
0 likes, 2 repeats |
|
|
|
@jann The limitation on speed will be system memory. setsockopt() can reduce… |
|
|
|
Post #Ax97SKaAvrfwSlfzOK by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@alison nope, that's not it |
|
|
|
Post #Ax98mSAMXubZjGlH6W by [email protected] |
|
0 likes, 2 repeats |
|
|
|
@jann Must be pthread_detach() then so that all the threads need not be join()&… |
|
|
|
Post #Ax9TwygcaabWWRi5Cq by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@madcoder thanks |
|
|
|
Post #AxCoKPmSbJLzlsm2oS by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@jann Why does OpenBSD have no issues with it and is 10X faster?https://flak.te… |
