[jitsi-users] Jitsi Videobridge performance


#1

Hi Guys,
According to the post: https://jitsi.org/Projects/JitsiVideobridgePerformance, the jvb handled 1000+ video streams with 20% CPU usages. I tested it in my side, the results were not same as the post. It handled 20 steams with 50% CPU usages. Does it need special configuration for jvb? Please kindly help.
My environment:
Debian 8.2/x642CPUs2G MemoryVideobridge: using apt-get -y install jitsi-meet (should be least version)

Best Regards,
Leo


#2

PS. My test VM hosted on digitalocean. I also tested it on my personal VM, same things.

Best Regards,Leo

···

From: leoyey@hotmail.com
To: users@jitsi.org
Date: Thu, 5 Nov 2015 15:40:36 +0800
Subject: [jitsi-users] Jitsi Videobridge performance

Hi Guys,
According to the post: https://jitsi.org/Projects/JitsiVideobridgePerformance, the jvb handled 1000+ video streams with 20% CPU usages. I tested it in my side, the results were not same as the post. It handled 20 steams with 50% CPU usages. Does it need special configuration for jvb? Please kindly help.
My environment:
Debian 8.2/x642CPUs2G MemoryVideobridge: using apt-get -y install jitsi-meet (should be least version)

Best Regards,
Leo

_______________________________________________
users mailing list
users@jitsi.org
Unsubscribe instructions and other list options:
http://lists.jitsi.org/mailman/listinfo/users


#3

You should have in mind that the evaluation was made on a dedicated
server. On DO VMs you have lots of restrictions, you don't have a real
CPU and also I'm not sure exactly how much of the host CPU features are
exposed to the cloud stack. On top of that you also have other users
running VMs on the same host. You're not mentioning the size of the
VM (CPU, RAM). It all can depend also on the load of the datacenter
you're in, so it's kind of normal to have lower performance.

···

On Thu, 5 Nov 2015 15:46:10 +0800 Leo Ye wrote:

PS. My test VM hosted on digitalocean. I also tested it on my personal
VM, same things.

--

Yasen Pramatarov
Lindeas Ltd. https://lindeas.com
'working on GNU/Linux ideas'
Professional Jitsi Meet services


#4

Hi Yasen,
Thanks for your information. You are right that hosted in VMs will slow down the performance. I did this test because I plan to host the JVB on cloud. In this case, the servers will be hosted on VMs provided by cloud vendor. Is it normal to host jvb on cloud based on your experience? Do you have any existing cases that JVB can handle lots of streams?
PS, I did a simple 4 streams test on my personal VM. JVB took 60% CPU when there are 4 steams connected. Please find my VM information below. I want to know whether it is normal nor not. Thanks for your advice.
H/W path Device Class Description=================================================== system Computer/0 bus Motherboard/0/0 memory 2010MiB System memory/0/1 processor Intel(R) Core(TM) i5-3210M CPU @ 2.50GHz/0/100 bridge 440FX - 82441FX PMC [Natoma]/0/100/1 bridge 82371SB PIIX3 ISA [Natoma/Triton II]/0/100/1.1 storage 82371AB/EB/MB PIIX4 IDE/0/100/2 display VirtualBox Graphics Adapter/0/100/3 eth0 network 82540EM Gigabit Ethernet Controller/0/100/4 generic VirtualBox Guest Service/0/100/5 multimedia 82801AA AC'97 Audio Controller/0/100/6 bus KeyLargo/Intrepid USB/0/100/6/1 usb2 bus OHCI PCI host controller/0/100/6/1/1 input USB Tablet/0/100/7 bridge 82371AB/EB/MB PIIX4 ACPI/0/100/b bus 82801FB/FBM/FR/FW/FRW (ICH6 Family) USB2 EHCI Controller/0/100/b/1 usb1 bus EHCI Host Controller/0/100/d storage 82801HM/HEM (ICH8M/ICH8M-E) SATA Controller [AHCI mode]/0/2 scsi0 storage /0/2/0.0.0 /dev/cdrom disk DVD reader/0/3 scsi2 storage /0/3/0.0.0 /dev/sda disk 107GB VBOX HARDDISK/0/3/0.0.0/1 /dev/sda1 volume 95GiB EXT4 volume/0/3/0.0.0/2 /dev/sda2 volume 4190MiB Extended partition/0/3/0.0.0/2/5 /dev/sda5 volume 4190MiB Linux swap / Solaris partition
Best Regards,
Leo

···

Date: Thu, 5 Nov 2015 10:02:09 +0200
From: yasen@lindeas.com
To: users@jitsi.org
Subject: Re: [jitsi-users] Jitsi Videobridge performance

On Thu, 5 Nov 2015 15:46:10 +0800 Leo Ye wrote:

>PS. My test VM hosted on digitalocean. I also tested it on my personal
>VM, same things.

You should have in mind that the evaluation was made on a dedicated
server. On DO VMs you have lots of restrictions, you don't have a real
CPU and also I'm not sure exactly how much of the host CPU features are
exposed to the cloud stack. On top of that you also have other users
running VMs on the same host. You're not mentioning the size of the
VM (CPU, RAM). It all can depend also on the load of the datacenter
you're in, so it's kind of normal to have lower performance.

--
> Yasen Pramatarov
> Lindeas Ltd. https://lindeas.com
> 'working on GNU/Linux ideas'
> Professional Jitsi Meet services

_______________________________________________
users mailing list
users@jitsi.org
Unsubscribe instructions and other list options:
http://lists.jitsi.org/mailman/listinfo/users


#5

You should make sure your VMs have AES-NI. I don't believe DO have that.
Amazon and Google do though.

···

On Thursday, 5 November 2015, Leo Ye <leoyey@hotmail.com> wrote:

Hi Yasen,

Thanks for your information. You are right that hosted in VMs will slow
down the performance. I did this test because I plan to host the JVB on
cloud. In this case, the servers will be hosted on VMs provided by cloud
vendor. Is it normal to host jvb on cloud based on your experience? Do you
have any existing cases that JVB can handle lots of streams?

PS, I did a simple 4 streams test on my personal VM. JVB took 60% CPU when
there are 4 steams connected. Please find my VM information below. I want
to know whether it is normal nor not. Thanks for your advice.

H/W path Device Class Description

                            system Computer
/0 bus Motherboard
/0/0 memory 2010MiB System memory
/0/1 processor Intel(R) Core(TM) i5-3210M CPU @
2.50GHz
/0/100 bridge 440FX - 82441FX PMC [Natoma]
/0/100/1 bridge 82371SB PIIX3 ISA [Natoma/Triton
II]
/0/100/1.1 storage 82371AB/EB/MB PIIX4 IDE
/0/100/2 display VirtualBox Graphics Adapter
/0/100/3 eth0 network 82540EM Gigabit Ethernet Controller
/0/100/4 generic VirtualBox Guest Service
/0/100/5 multimedia 82801AA AC'97 Audio Controller
/0/100/6 bus KeyLargo/Intrepid USB
/0/100/6/1 usb2 bus OHCI PCI host controller
/0/100/6/1/1 input USB Tablet
/0/100/7 bridge 82371AB/EB/MB PIIX4 ACPI
/0/100/b bus 82801FB/FBM/FR/FW/FRW (ICH6
Family) USB2 EHCI Controller
/0/100/b/1 usb1 bus EHCI Host Controller
/0/100/d storage 82801HM/HEM (ICH8M/ICH8M-E) SATA
Controller [AHCI mode]
/0/2 scsi0 storage
/0/2/0.0.0 /dev/cdrom disk DVD reader
/0/3 scsi2 storage
/0/3/0.0.0 /dev/sda disk 107GB VBOX HARDDISK
/0/3/0.0.0/1 /dev/sda1 volume 95GiB EXT4 volume
/0/3/0.0.0/2 /dev/sda2 volume 4190MiB Extended partition
/0/3/0.0.0/2/5 /dev/sda5 volume 4190MiB Linux swap / Solaris
partition

Best Regards,

Leo

> Date: Thu, 5 Nov 2015 10:02:09 +0200
> From: yasen@lindeas.com
<javascript:_e(%7B%7D,'cvml','yasen@lindeas.com');>
> To: users@jitsi.org <javascript:_e(%7B%7D,'cvml','users@jitsi.org');>
> Subject: Re: [jitsi-users] Jitsi Videobridge performance
>
> On Thu, 5 Nov 2015 15:46:10 +0800 Leo Ye wrote:
>
> >PS. My test VM hosted on digitalocean. I also tested it on my personal
> >VM, same things.
>
> You should have in mind that the evaluation was made on a dedicated
> server. On DO VMs you have lots of restrictions, you don't have a real
> CPU and also I'm not sure exactly how much of the host CPU features are
> exposed to the cloud stack. On top of that you also have other users
> running VMs on the same host. You're not mentioning the size of the
> VM (CPU, RAM). It all can depend also on the load of the datacenter
> you're in, so it's kind of normal to have lower performance.
>
> --
> > Yasen Pramatarov
> > Lindeas Ltd. https://lindeas.com
> > 'working on GNU/Linux ideas'
> > Professional Jitsi Meet services
>
> _______________________________________________
> users mailing list
> users@jitsi.org <javascript:_e(%7B%7D,'cvml','users@jitsi.org');>
> Unsubscribe instructions and other list options:
> http://lists.jitsi.org/mailman/listinfo/users

--
sent from my mobile


#6

Hi Emil,
Thanks for your advice. I will definitely check whether the cloud provider will have AES-NI or not. Could you explain a lot bit more why AES-NI will affect the JVB's performance?

Best Regards,
Leo

···

Date: Fri, 6 Nov 2015 05:48:42 -0600
From: emcho@jitsi.org
To: users@jitsi.org
Subject: Re: [jitsi-users] Jitsi Videobridge performance

You should make sure your VMs have AES-NI. I don't believe DO have that. Amazon and Google do though.

On Thursday, 5 November 2015, Leo Ye <leoyey@hotmail.com> wrote:

Hi Yasen,
Thanks for your information. You are right that hosted in VMs will slow down the performance. I did this test because I plan to host the JVB on cloud. In this case, the servers will be hosted on VMs provided by cloud vendor. Is it normal to host jvb on cloud based on your experience? Do you have any existing cases that JVB can handle lots of streams?
PS, I did a simple 4 streams test on my personal VM. JVB took 60% CPU when there are 4 steams connected. Please find my VM information below. I want to know whether it is normal nor not. Thanks for your advice.
H/W path Device Class Description=================================================== system Computer/0 bus Motherboard/0/0 memory 2010MiB System memory/0/1 processor Intel(R) Core(TM) i5-3210M CPU @ 2.50GHz/0/100 bridge 440FX - 82441FX PMC [Natoma]/0/100/1 bridge 82371SB PIIX3 ISA [Natoma/Triton II]/0/100/1.1 storage 82371AB/EB/MB PIIX4 IDE/0/100/2 display VirtualBox Graphics Adapter/0/100/3 eth0 network 82540EM Gigabit Ethernet Controller/0/100/4 generic VirtualBox Guest Service/0/100/5 multimedia 82801AA AC'97 Audio Controller/0/100/6 bus KeyLargo/Intrepid USB/0/100/6/1 usb2 bus OHCI PCI host controller/0/100/6/1/1 input USB Tablet/0/100/7 bridge 82371AB/EB/MB PIIX4 ACPI/0/100/b bus 82801FB/FBM/FR/FW/FRW (ICH6 Family) USB2 EHCI Controller/0/100/b/1 usb1 bus EHCI Host Controller/0/100/d storage 82801HM/HEM (ICH8M/ICH8M-E) SATA Controller [AHCI mode]/0/2 scsi0 storage /0/2/0.0.0 /dev/cdrom disk DVD reader/0/3 scsi2 storage /0/3/0.0.0 /dev/sda disk 107GB VBOX HARDDISK/0/3/0.0.0/1 /dev/sda1 volume 95GiB EXT4 volume/0/3/0.0.0/2 /dev/sda2 volume 4190MiB Extended partition/0/3/0.0.0/2/5 /dev/sda5 volume 4190MiB Linux swap / Solaris partition
Best Regards,
Leo

Date: Thu, 5 Nov 2015 10:02:09 +0200
From: yasen@lindeas.com
To: users@jitsi.org
Subject: Re: [jitsi-users] Jitsi Videobridge performance

On Thu, 5 Nov 2015 15:46:10 +0800 Leo Ye wrote:

>PS. My test VM hosted on digitalocean. I also tested it on my personal
>VM, same things.

You should have in mind that the evaluation was made on a dedicated
server. On DO VMs you have lots of restrictions, you don't have a real
CPU and also I'm not sure exactly how much of the host CPU features are
exposed to the cloud stack. On top of that you also have other users
running VMs on the same host. You're not mentioning the size of the
VM (CPU, RAM). It all can depend also on the load of the datacenter
you're in, so it's kind of normal to have lower performance.

--
> Yasen Pramatarov
> Lindeas Ltd. https://lindeas.com
> 'working on GNU/Linux ideas'
> Professional Jitsi Meet services

_______________________________________________
users mailing list
users@jitsi.org
Unsubscribe instructions and other list options:
http://lists.jitsi.org/mailman/listinfo/users
                 
--
sent from my mobile

_______________________________________________
users mailing list
users@jitsi.org
Unsubscribe instructions and other list options:
http://lists.jitsi.org/mailman/listinfo/users


#7

The seet of instructions allow the bridge to delegate encryption to
hardware.

Emil

···

On Saturday, 7 November 2015, Leo Ye <leoyey@hotmail.com> wrote:

Hi Emil,

Thanks for your advice. I will definitely check whether the cloud provider
will have AES-NI or not. Could you explain a lot bit more why AES-NI will
affect the JVB's performance?

Best Regards,

Leo

------------------------------
Date: Fri, 6 Nov 2015 05:48:42 -0600
From: emcho@jitsi.org <javascript:_e(%7B%7D,'cvml','emcho@jitsi.org');>
To: users@jitsi.org <javascript:_e(%7B%7D,'cvml','users@jitsi.org');>
Subject: Re: [jitsi-users] Jitsi Videobridge performance

You should make sure your VMs have AES-NI. I don't believe DO have that.
Amazon and Google do though.

On Thursday, 5 November 2015, Leo Ye <leoyey@hotmail.com > <javascript:_e(%7B%7D,'cvml','leoyey@hotmail.com');>> wrote:

Hi Yasen,

Thanks for your information. You are right that hosted in VMs will slow
down the performance. I did this test because I plan to host the JVB on
cloud. In this case, the servers will be hosted on VMs provided by cloud
vendor. Is it normal to host jvb on cloud based on your experience? Do you
have any existing cases that JVB can handle lots of streams?

PS, I did a simple 4 streams test on my personal VM. JVB took 60% CPU when
there are 4 steams connected. Please find my VM information below. I want
to know whether it is normal nor not. Thanks for your advice.

H/W path Device Class Description

                            system Computer
/0 bus Motherboard
/0/0 memory 2010MiB System memory
/0/1 processor Intel(R) Core(TM) i5-3210M CPU @
2.50GHz
/0/100 bridge 440FX - 82441FX PMC [Natoma]
/0/100/1 bridge 82371SB PIIX3 ISA [Natoma/Triton
II]
/0/100/1.1 storage 82371AB/EB/MB PIIX4 IDE
/0/100/2 display VirtualBox Graphics Adapter
/0/100/3 eth0 network 82540EM Gigabit Ethernet Controller
/0/100/4 generic VirtualBox Guest Service
/0/100/5 multimedia 82801AA AC'97 Audio Controller
/0/100/6 bus KeyLargo/Intrepid USB
/0/100/6/1 usb2 bus OHCI PCI host controller
/0/100/6/1/1 input USB Tablet
/0/100/7 bridge 82371AB/EB/MB PIIX4 ACPI
/0/100/b bus 82801FB/FBM/FR/FW/FRW (ICH6
Family) USB2 EHCI Controller
/0/100/b/1 usb1 bus EHCI Host Controller
/0/100/d storage 82801HM/HEM (ICH8M/ICH8M-E) SATA
Controller [AHCI mode]
/0/2 scsi0 storage
/0/2/0.0.0 /dev/cdrom disk DVD reader
/0/3 scsi2 storage
/0/3/0.0.0 /dev/sda disk 107GB VBOX HARDDISK
/0/3/0.0.0/1 /dev/sda1 volume 95GiB EXT4 volume
/0/3/0.0.0/2 /dev/sda2 volume 4190MiB Extended partition
/0/3/0.0.0/2/5 /dev/sda5 volume 4190MiB Linux swap / Solaris
partition

Best Regards,

Leo

> Date: Thu, 5 Nov 2015 10:02:09 +0200
> From: yasen@lindeas.com
> To: users@jitsi.org
> Subject: Re: [jitsi-users] Jitsi Videobridge performance
>
> On Thu, 5 Nov 2015 15:46:10 +0800 Leo Ye wrote:
>
> >PS. My test VM hosted on digitalocean. I also tested it on my personal
> >VM, same things.
>
> You should have in mind that the evaluation was made on a dedicated
> server. On DO VMs you have lots of restrictions, you don't have a real
> CPU and also I'm not sure exactly how much of the host CPU features are
> exposed to the cloud stack. On top of that you also have other users
> running VMs on the same host. You're not mentioning the size of the
> VM (CPU, RAM). It all can depend also on the load of the datacenter
> you're in, so it's kind of normal to have lower performance.
>
> --
> > Yasen Pramatarov
> > Lindeas Ltd. https://lindeas.com
> > 'working on GNU/Linux ideas'
> > Professional Jitsi Meet services
>
> _______________________________________________
> users mailing list
> users@jitsi.org
> Unsubscribe instructions and other list options:
> http://lists.jitsi.org/mailman/listinfo/users

--
sent from my mobile

_______________________________________________ users mailing list
users@jitsi.org <javascript:_e(%7B%7D,'cvml','users@jitsi.org');>
Unsubscribe instructions and other list options:
http://lists.jitsi.org/mailman/listinfo/users

--
sent from my mobile


#8

Hi Emil,
Thanks for explaining. Is there any way that we can disable encryption for Jitis Videobridge?

Best Regards,
Leo

···

Date: Sun, 8 Nov 2015 05:53:31 -0600
From: emcho@jitsi.org
To: users@jitsi.org
Subject: Re: [jitsi-users] Jitsi Videobridge performance

The seet of instructions allow the bridge to delegate encryption to hardware.
Emil

On Saturday, 7 November 2015, Leo Ye <leoyey@hotmail.com> wrote:

Hi Emil,
Thanks for your advice. I will definitely check whether the cloud provider will have AES-NI or not. Could you explain a lot bit more why AES-NI will affect the JVB's performance?

Best Regards,
Leo

Date: Fri, 6 Nov 2015 05:48:42 -0600
From: emcho@jitsi.org
To: users@jitsi.org
Subject: Re: [jitsi-users] Jitsi Videobridge performance

You should make sure your VMs have AES-NI. I don't believe DO have that. Amazon and Google do though.

On Thursday, 5 November 2015, Leo Ye <leoyey@hotmail.com> wrote:

Hi Yasen,
Thanks for your information. You are right that hosted in VMs will slow down the performance. I did this test because I plan to host the JVB on cloud. In this case, the servers will be hosted on VMs provided by cloud vendor. Is it normal to host jvb on cloud based on your experience? Do you have any existing cases that JVB can handle lots of streams?
PS, I did a simple 4 streams test on my personal VM. JVB took 60% CPU when there are 4 steams connected. Please find my VM information below. I want to know whether it is normal nor not. Thanks for your advice.
H/W path Device Class Description=================================================== system Computer/0 bus Motherboard/0/0 memory 2010MiB System memory/0/1 processor Intel(R) Core(TM) i5-3210M CPU @ 2.50GHz/0/100 bridge 440FX - 82441FX PMC [Natoma]/0/100/1 bridge 82371SB PIIX3 ISA [Natoma/Triton II]/0/100/1.1 storage 82371AB/EB/MB PIIX4 IDE/0/100/2 display VirtualBox Graphics Adapter/0/100/3 eth0 network 82540EM Gigabit Ethernet Controller/0/100/4 generic VirtualBox Guest Service/0/100/5 multimedia 82801AA AC'97 Audio Controller/0/100/6 bus KeyLargo/Intrepid USB/0/100/6/1 usb2 bus OHCI PCI host controller/0/100/6/1/1 input USB Tablet/0/100/7 bridge 82371AB/EB/MB PIIX4 ACPI/0/100/b bus 82801FB/FBM/FR/FW/FRW (ICH6 Family) USB2 EHCI Controller/0/100/b/1 usb1 bus EHCI Host Controller/0/100/d storage 82801HM/HEM (ICH8M/ICH8M-E) SATA Controller [AHCI mode]/0/2 scsi0 storage /0/2/0.0.0 /dev/cdrom disk DVD reader/0/3 scsi2 storage /0/3/0.0.0 /dev/sda disk 107GB VBOX HARDDISK/0/3/0.0.0/1 /dev/sda1 volume 95GiB EXT4 volume/0/3/0.0.0/2 /dev/sda2 volume 4190MiB Extended partition/0/3/0.0.0/2/5 /dev/sda5 volume 4190MiB Linux swap / Solaris partition
Best Regards,
Leo

Date: Thu, 5 Nov 2015 10:02:09 +0200
From: yasen@lindeas.com
To: users@jitsi.org
Subject: Re: [jitsi-users] Jitsi Videobridge performance

On Thu, 5 Nov 2015 15:46:10 +0800 Leo Ye wrote:

>PS. My test VM hosted on digitalocean. I also tested it on my personal
>VM, same things.

You should have in mind that the evaluation was made on a dedicated
server. On DO VMs you have lots of restrictions, you don't have a real
CPU and also I'm not sure exactly how much of the host CPU features are
exposed to the cloud stack. On top of that you also have other users
running VMs on the same host. You're not mentioning the size of the
VM (CPU, RAM). It all can depend also on the load of the datacenter
you're in, so it's kind of normal to have lower performance.

--
> Yasen Pramatarov
> Lindeas Ltd. https://lindeas.com
> 'working on GNU/Linux ideas'
> Professional Jitsi Meet services

_______________________________________________
users mailing list
users@jitsi.org
Unsubscribe instructions and other list options:
http://lists.jitsi.org/mailman/listinfo/users
                 
--
sent from my mobile

_______________________________________________
users mailing list
users@jitsi.org
Unsubscribe instructions and other list options:
http://lists.jitsi.org/mailman/listinfo/users

--
sent from my mobile

_______________________________________________
users mailing list
users@jitsi.org
Unsubscribe instructions and other list options:
http://lists.jitsi.org/mailman/listinfo/users


#9

It's part of the way the system works. In WebRTC the media is
encrypted, but for the conference use case in the SFU (the
Videobridge) the streams have to be decrypted in order to get some of
the data for synchronization purposes. After that is is re-encrypted.
All of this takes some CPU.

That being said, one of the major advantages of Jitsi Meet is the
encryption, so my advice would be not to look for a way to disable it,
but more like to change the VM provider or host the bridge on a
physical machine, even if it's a small one.

···

On Mon, 9 Nov 2015 10:53:15 +0800 Leo Ye wrote:

Is there any way that we can disable encryption for Jitis Videobridge?

--

Yasen Pramatarov
Lindeas Ltd. https://lindeas.com
'working on GNU/Linux ideas'
Professional Jitsi Meet services


#10

HI Yasen,
Thanks for your advice. It is much clear now.

Best Regards,
Leo

···

Date: Mon, 9 Nov 2015 10:00:48 +0200
From: yasen@lindeas.com
To: users@jitsi.org
Subject: Re: [jitsi-users] Jitsi Videobridge performance

On Mon, 9 Nov 2015 10:53:15 +0800 Leo Ye wrote:

>Is there any way that we can disable encryption for Jitis Videobridge?

It's part of the way the system works. In WebRTC the media is
encrypted, but for the conference use case in the SFU (the
Videobridge) the streams have to be decrypted in order to get some of
the data for synchronization purposes. After that is is re-encrypted.
All of this takes some CPU.

That being said, one of the major advantages of Jitsi Meet is the
encryption, so my advice would be not to look for a way to disable it,
but more like to change the VM provider or host the bridge on a
physical machine, even if it's a small one.

--
> Yasen Pramatarov
> Lindeas Ltd. https://lindeas.com
> 'working on GNU/Linux ideas'
> Professional Jitsi Meet services

_______________________________________________
users mailing list
users@jitsi.org
Unsubscribe instructions and other list options:
http://lists.jitsi.org/mailman/listinfo/users