Tweet
Power regression by futex problem?
Hello everybody,
I've started playing with MySQL vs. MariaDB lately. The replacement works just fine, though I have some concerns about a power regression (and possibly more side effects). This does not only effect MySQL/MariaDB but every threaded application using fast user-space locking.
Tracing system calls of these processes with strace I can see a futex timeout every second. Searching for this behavior on Google brings up a lot of discussions about 2012's leap second, though MySQL and others are reported to generate really high load. As said for me this is just one wakeup a seconds.
Setting the time with 'date -s "$(date)"' does not help in my case. Setting the time back for X seconds makes the futex problem disappear for these X seconds. After that it just continues.
As a lot of applications are threaded these days and every running threaded process seems to wake the system each second this looks like a really bad power regression to me. I could reproduce this with Arch Linux and openSUSE, have not tested any more.
Any clue what is going on?
Regards,
Chris
I've started playing with MySQL vs. MariaDB lately. The replacement works just fine, though I have some concerns about a power regression (and possibly more side effects). This does not only effect MySQL/MariaDB but every threaded application using fast user-space locking.
Tracing system calls of these processes with strace I can see a futex timeout every second. Searching for this behavior on Google brings up a lot of discussions about 2012's leap second, though MySQL and others are reported to generate really high load. As said for me this is just one wakeup a seconds.
Setting the time with 'date -s "$(date)"' does not help in my case. Setting the time back for X seconds makes the futex problem disappear for these X seconds. After that it just continues.
As a lot of applications are threaded these days and every running threaded process seems to wake the system each second this looks like a really bad power regression to me. I could reproduce this with Arch Linux and openSUSE, have not tested any more.
Any clue what is going on?
Regards,
Chris