Java Httpurlconnection

[Florian Weimer]

* Somshekar C. Kadam:

> I am seeing  on ARM lot many seconds wasted on this as shown below logs.
> Where as in Intel I did not see a single reference, seems that is something
> to do with it, need to  look further into it.
>
> ============
> 15033 08:10:27.254816 clock_gettime(CLOCK_MONOTONIC, {4286, 606936331}) = 0
> 15033 08:10:27.255070 clock_gettime(CLOCK_MONOTONIC, {4286, 607189947}) = 0
> 15033 08:10:27.255504 clock_gettime(CLOCK_MONOTONIC, {4286, 607662716}) = 0
> 15033 08:10:27.255924 clock_gettime(CLOCK_MONOTONIC, {4286, 608074100}) = 0
> ==================================

On x86, there is an clock_gettime implementation in the vDSO, with a
fast path that stays in userspace.  It's much faster than the
implementation based on the system call, and calls do not show up in
strace (because there is no system call).

Maybe you can try to get a backtrace, to determine where the calls are
coming from?  And how long do these calls take individually?  Normally,
a clock_gettime call every 400 microseconds or so shouldn't cause a
drastic slowdown, but if CLOCK_MONOTONIC is particularly slow to access
on this platform, it could be the cause.

Thanks,
Florian