[erlang-questions] what means busy_wait in msacc erl 19.0?

Lukas Larsson garazdawi@REDACTED
Mon Jun 27 19:49:16 CEST 2016


On Mon, Jun 27, 2016 at 6:58 PM, Max Lapshin <max.lapshin@REDACTED> wrote:

> without +MBas aoffcaobf +MBacul 0  it is just leaking memory (maybe not
> leaking, but using it so much, that OOM comes earlier).
>

Running both of those seems counter intuitive, basically aoffcaobf is
needed in order to enable acul, but since you disable acul it makes little
sense to have it. If the default (bf) is not good enough, aobf should be
better than aoffcaobf.

IIRC I talked to you in early 18.x about a bug in the acul pool which has
since been fixed, have you checked if your issues are still present after
that fix? If they are, it would be great if you could send me a recon_alloc
snapshot of both a healthy system and one that is about to crash (or the
crash dump if it has already crashed).


>
> flag   +stbt s that we were using earlier is important to capture better
> multicast UDP, but make system absolutely unstable (it eats all memory and
> dies) when some other task is launched on a server, so we have refused from
> it.
>

that is indeed the problem with +sbt, which is why don't recommend you run
it unless you know that no cpu intensive tasks will compete on the same
cores.


>
> Flags  +sbwt short +swt very_low +sfwi 20 +sub true  are recommended by
> our customers.  Without them since 18.3 we get messages from sys_monitor
> like   long_schedule a and long_gc on a server with 50% usage of CPU.
>

I don't really see how any of those flags could directly influence the
amount of long_schedule or long_gc you get on your system. There could be
some subtle effect in how things are scheduled and woken that changes.

Do you run on real hw or in a virtual machine of some sort?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://erlang.org/pipermail/erlang-questions/attachments/20160627/1b837113/attachment.htm>


More information about the erlang-questions mailing list