Stanislav, Sorry about that - I hadn't seen your reply! Understood on your suggestions. I guess my ultimate question is - is virtualization still a huge problem aside from noisy neighbors? AWS of course allows you to get dedicated instances (albeit expensive), but you're still running in a VM. The complexity of including another data center in the infrastructure is a bit too much for this use case, but as mentioned - it'll definitely move this way over time. Joel, Nothing other than what's on this thread. However, I will say that I moved from Ubuntu to an Amazon Linux AMI, moved logging from direct to disk out to syslog (with async writes to disk), and increased the instance size to a c4.xlarge, and users are reporting considerable improvements. Best, Colin On Thu, Feb 16, 2017 at 5:01 PM Joel Serrano <joel@gogii.net> wrote:
Hi Colin,
Did you get any more info on FS on AWS? I'm interested on this also.
Thanks, Joel. On Mon, Feb 13, 2017 at 6:43 AM, Colin Morelli <colin.morelli@gmail.com> wrote:
Hey list,
I'll start by saying I have read the warnings and recommendations to *not* run FS on virtualized hardware. For right now, though, it's really the only option for me. I don't anticipate nor will I expect the behavior to be the same as what I would get on physical hosts, but would just like it to be as consistent (and good) as possible given the constraints.
My question is - does anyone have particular kernel and instance type recommendations for AWS? I'm getting relatively frequent jitter and what seem like timing-related issues on calls, even in simple device <-> device calls within FS (no external gateway involved). I'm currently running FS in a Debian 8 docker container (with no memory or CPU limits) on a c4.large Ubuntu 16.04 host (kernel 4.4.0-45, default 250hz timer). The particular behavior I'm seeing *feels and sounds* like a few frames of silence, followed up by the audio being sped up slightly for a second or two (almost like it's trying to catch up), happening occasionally throughout calls. Note that this is with one call, and ~5% CPU usage on the host. Thus I'm pointing a finger most likely at kernel timer resolution. Running time_test 100 on FS returns an average of 200, which seems pretty bad.
I've heard people have better success with the Amazon Linux AMIs for their packaged kernels. Is this true? Would recompiling the kernel with a 1000hz timer help a lot here?
If you're having success on AWS, would be great to know what AMI you're based off of, what kernel version (and timer), and instance size. Thanks in advance.
Best, Colin
_________________________________________________________________________ Professional FreeSWITCH Consulting Services: consulting@freeswitch.org http://www.freeswitchsolutions.com
Official FreeSWITCH Sites http://www.freeswitch.org http://confluence.freeswitch.org http://www.cluecon.com
FreeSWITCH-users mailing list FreeSWITCH-users@lists.freeswitch.org http://lists.freeswitch.org/mailman/listinfo/freeswitch-users UNSUBSCRIBE:http://lists.freeswitch.org/mailman/options/freeswitch-users http://www.freeswitch.org
_________________________________________________________________________ Professional FreeSWITCH Consulting Services: consulting@freeswitch.org http://www.freeswitchsolutions.com
Official FreeSWITCH Sites http://www.freeswitch.org http://confluence.freeswitch.org http://www.cluecon.com
FreeSWITCH-users mailing list FreeSWITCH-users@lists.freeswitch.org http://lists.freeswitch.org/mailman/listinfo/freeswitch-users UNSUBSCRIBE:http://lists.freeswitch.org/mailman/options/freeswitch-users http://www.freeswitch.org