Hello All,
it seems I might found the way to increase the stability of streams in the
room
I would ask you to test the nightly build (as soon as I commit my changes)
I don't have too much hardware of my own :(((
I'll write to this thread as soon as I will be ready
stay tuned :))
On Thu, 22 Oct 2020 at 20:54, Maxim Solodovnik <solomax...@gmail.com> wrote:
> Steams might be combined on server
> It is easier for client to display one huge video than 10 small ones
>
> there is also follow talker mode, when only one video is displayed
> you also can use presentation room and "stream on demand"
>
> Additional option might be: switching video codec
>
> I do remember the default codec for Flash (h253 if I'm not mistaken)
> consumes less CPU than h264
> (the higher is compression - the less bandwidth is required, but more CPU)
>
> On Thu, 22 Oct 2020 at 20:33, dww <dwort...@mykolab.com> wrote:
>
>>
>> It seems that currently the limitation is the CPU usage on the client
>> side just over 5 total users. At this point KMS on the server is not over
>> stressed. The KMS stream limitation could be fixed via clustering via a
>> docker swarm spanning multiple nodes.
>>
>> I am curious how proprietary applications like Zoom and Microsoft Teams
>> handle this. Is there some kind of prioritization of the speaker or round
>> robin updates?
>>
>> Up to 10 users would be a good start but we need to aim for higher.
>>
>> Dennis
>>
>> On Thu, 2020-10-22 at 19:07 +0700, Maxim Solodovnik wrote:
>>
>> Hello Denis,
>>
>> please let me know if 10 users in the room will be OK
>> we need to change the way cluster works in case current configuration is
>> not stable/powerful enough
>>
>> On Thu, 22 Oct 2020 at 13:33, Denis Noctor <denisnoc...@gmail.com> wrote:
>>
>> Actually... this is something that had crossed my mind recently.
>> Previously, I’ve had on average 5 users per group... but on occasion if
>> there more than 5 users (in and around 7-10) some users... some would
>> intentionally drop drop their cam... and sometimes mic... (I thought it
>> might have been due to bandwidth issues on their side)... but I’ve got my
>> own kids taking online classes via their schools... whereby there are a min
>> of 20 attending a class and so on (different platforms etc.)
>>
>> I have about a combination of 5 to 6 computers at home due to home office
>> / covid restrictions etc. and am happy to log in to either the OM demo
>> sites or others to try replicate this scenario. (and my own, of course).
>>
>> Yep, I’ve been reading up on WebRTC and Kurento regarding connectivity
>> limitations regarding audio and video... this has been covered a lot
>> recently in previous posts in this forum... and from memory it is somewhere
>> between 200-300 connections... (per server? / instance?)
>>
>> For example... if you have 5 users in a room (using audio and video) this
>> will result in 5 (users) x 5 audio connections x 5 video connections...
>> giving “125” connections... but if you have 7 users in a room receiving and
>> experiencing audio and vid... that is 7x7x7= 343 connections... which
>> obviously exceeds the connections as per previous posts.
>>
>> I have had a scenario whereby 3 classes were held at the same time on the
>> same AWS instance... and nobody has reported a problem (yet... or they
>> unknowingly downplayed it due to internet bandwidth problems etc.)... but
>> the max number of students per room has been 4-5....
>>
>> I am happy to test this more with others, if you are up for it and am
>> watching this carefully.
>>
>> I have a scheduled meeting with 10 participants next week and am now
>> nervous... and curious to see how it works out.
>>
>> Talk soon.
>>
>> Denis
>>
>>
>>
>> Sent from my iPhone
>>
>> > On Oct 21, 2020, at 7:57 PM, dww <dwort...@mykolab.com> wrote:
>> >
>> > Does anyone have an idea why the client browser seems to limited to 5
>> > video pods when each user connects with both video and microphone.
>> >
>> > It seems to definitely max out the CPU(s) on the client computer. How
>> > do other applications like Zoom or Microsoft Teams get around this
>> > issue. Do they throttle back the streaming of the video pods or views
>> > of many or most of the pods and do it in some kind of round-robin
>> > update?
>> >
>> > Am I just doing something wrong.
>> >
>> > Any response would be appreciated. I would like to start uses OM for
>> > meetings next month.
>> >
>> > Thanks,
>> > Dennis
>> >
>> >
>> >> On Sun, 2020-10-18 at 17:40 -0400, dww wrote:
>> >> When creating 3 guests on each of 2 other laptops that I run into the
>> >> same issues,so it seems that the main limitation is the number of
>> >> video/audio pods within a client. Is there anyway to get around that
>> >> as
>> >> it implies that one annot have more than 4 or 5 guests with both
>> >> video
>> >> and audio?
>> >>
>> >> Dennis
>> >>
>> >>
>> >>
>> >>> On Sun, 2020-10-18 at 13:00 -0400, dww wrote:
>> >>> Yes I get the same result, however, this does not appear to be a
>> >>> valid
>> >>> stress test of kms. This stress tests a client machine with
>> >>> multiple
>> >>> tabs or browser windows each with connections and determines that
>> >>> it
>> >>> is
>> >>> CPU bound.
>> >>>
>> >>> It seems the only way to stress test kms is to do this with
>> >>> multiple
>> >>> client machines. I have 3 laptops here and a couple of smart phones
>> >>> so
>> >>> I will try distributing the client windows among all of them.
>> >>>
>> >>> Dennis
>> >>>
>> >>>> On Sun, 2020-10-18 at 10:56 +0700, Maxim Solodovnik wrote:
>> >>>> I'm on Ubuntu 20.04 desktop? so i can use UI
>> >>>> if you are using server I would recommend `htop`
>> >>>>
>> >>>>> On Sun, 18 Oct 2020 at 10:54, dww <dwort...@mykolab.com> wrote:
>> >>>>> May I ask for your linux command line that got the CPU
>> >>>>> percentage
>> >>>>> for all cores?
>> >>>>> thanks,
>> >>>>> Dennis
>> >>>>>
>> >>>>>
>> >>>>>
>> >>>>>> On Sat, 2020-10-17 at 12:59 +0700, Maxim Solodovnik wrote:
>> >>>>>> Hello,
>> >>>>>>
>> >>>>>> i just have tried the following script
>> >>>>>> started as `./run10.sh 5`
>> >>>>>>
>> >>>>>> everything seems to work, but my CPU was 800% busy (all cores
>> >>>>>> were 100% busy)
>> >>>>>>
>> >>>>>> without `--use-fake-device-for-media-stream` parameter I had
>> >>>>>> lots
>> >>>>>> of permission errors due to camera was "captured" by first
>> >>>>>> browser
>> >>>>>> other have reported "Camera busy" error
>> >>>>>>
>> >>>>>>
>> >>>>>> _HASH_HERE_ - should be replaced with real hash (I have
>> >>>>>> created
>> >>>>>> endless invitation hash to the private conference room)
>> >>>>>>
>> >>>>>> the script
>> >>>>>> ===============================================
>> >>>>>> #!/bin/bash
>> >>>>>>
>> >>>>>> i=$1
>> >>>>>>
>> >>>>>> if [ -z "${i}" ]; then
>> >>>>>> i=30
>> >>>>>> fi
>> >>>>>> let "i += 0"
>> >>>>>>
>> >>>>>> rm -rf /tmp/delme*
>> >>>>>>
>> >>>>>> while ((i--)); do
>> >>>>>> #echo "${i}"
>> >>>>>> mkdir /tmp/delme${i}
>> >>>>>>
>> >>>>>> #local conference
>> >>>>>> chromium-browser --user-data-dir=/tmp/delme${i} --disable-
>> >>>>>> infobars --no-default-browser-check --allow-insecure-
>> >>>>>> localhost
>> >>>>>> --
>> >>>>>> use-fake-device-for-media-stream '
>> >>>>>>
>> https://localhost:5443/openmeetings/hash?invitation=_HASH_HERE_&language=1
>> '
>> >>>>>> &
>> >>>>>> done
>> >>>>>>
>> >>>>>> On Fri, 16 Oct 2020 at 09:10, dww <dwort...@mykolab.com>
>> >>>>>> wrote:
>> >>>>>>> I mentioned earlier that I wanted to experiment with
>> >>>>>>> clustering
>> >>>>>>> using
>> >>>>>>> docker swarm for the kms service.
>> >>>>>>>
>> >>>>>>> I wanted to establish a base line using one container for
>> >>>>>>> kms.
>> >>>>>>> All the
>> >>>>>>> components are in one Linode with 8 GB of RAM.
>> >>>>>>>
>> >>>>>>> For this experiment. I start entering the video and
>> >>>>>>> whiteboard
>> >>>>>>> room
>> >>>>>>> from my admin login from Firefox on one laptop. I allowed
>> >>>>>>> both
>> >>>>>>> video
>> >>>>>>> and microphone and the video pod was the lowest resolution.
>> >>>>>>> I
>> >>>>>>> generate
>> >>>>>>> a guest url to the room.
>> >>>>>>>
>> >>>>>>> On another laptop also on the same connection to the
>> >>>>>>> internet
>> >>>>>>> I
>> >>>>>>> enter
>> >>>>>>> the room on firefox and allow both video ( lowest
>> >>>>>>> resolution)
>> >>>>>>> and
>> >>>>>>> microphone on a new tab each time.
>> >>>>>>>
>> >>>>>>> Up to 4 guest clients load quickly and the pods are created
>> >>>>>>> almost
>> >>>>>>> immediately on all 5 tabs. On the first attempt on the 5th
>> >>>>>>> guest the
>> >>>>>>> pod for this guest on the admin laptop took a couple of
>> >>>>>>> minutes
>> >>>>>>> to load
>> >>>>>>> the video. On the 5th guest tab, the pods for guest 1, 2
>> >>>>>>> and
>> >>>>>>> 4
>> >>>>>>> would
>> >>>>>>> not refresh( pod frames present but no video.
>> >>>>>>>
>> >>>>>>> RAM and CPU usage was not significant on the server.
>> >>>>>>>
>> >>>>>>> I closed the tabs for all the guests and redid the
>> >>>>>>> experiment,
>> >>>>>>> The
>> >>>>>>> first 4 guests again loaded quickly without any issues. The
>> >>>>>>> 5th
>> >>>>>>> guest
>> >>>>>>> loaded to completion but took about 40 seconds for all 6
>> >>>>>>> tabs
>> >>>>>>> to
>> >>>>>>> complete. On the 6th guest there were multiple connection
>> >>>>>>> drops
>> >>>>>>> and
>> >>>>>>> retries and the tabs were reduced to about 3 pods working,
>> >>>>>>> the
>> >>>>>>> others
>> >>>>>>> gone, this was on all tabs.
>> >>>>>>>
>> >>>>>>> So based on this it seems that up to 5 users using both
>> >>>>>>> video
>> >>>>>>> and
>> >>>>>>> microphone seem to work fine.
>> >>>>>>>
>> >>>>>>> Does this agree with anyone else's experience? I had from
>> >>>>>>> other
>> >>>>>>> posts
>> >>>>>>> that we can expect 14-15 users per kms instance. Does both
>> >>>>>>> laptop on
>> >>>>>>> the same network have any influence on this?
>> >>>>>>>
>> >>>>>>> I will try a swarm next after I gets some feedback.
>> >>>>>>>
>> >>>>>>>
>> >
>>
>>
>>
>>
>
> --
> Best regards,
> Maxim
>
--
Best regards,
Maxim
