Media Recorder sizing for voice, video, screen - application share recording

Voice recording CPU sizing

The table below shows the supported number of simultaneous voice calls per CPU core:

The system was tested up to 16 core servers, where 2 cores must be reserved for the OS, leaving up to 14 CPU cores for the application.
When certain features are enabled on the Recording Servers, the capacity numbers change:
- Encryption: when storage file encryption is enabled, capacity numbers must be decreased by 10%
- Voice quality check: when voice quality check is enabled, capacity numbers must be decreased by 15%
Numbers in red denotes default values, should be used for server sizing
Numbers only applicable when Receive-side scaling (RSS) is enabled in the OS

Storage Codec / Network Codec
Recorded Platform		Silk	G.711	G.722
Skype for Business	GSM-FR (Wave)	140	250	160
	PCM (Wave)	154	275	176
	MS-ADPCM (Wave)	140	250	160
	Speex (Ogg)	92	165	106
		G.729	G.711	G.722
Cisco Network-Based Recording	GSM-FR (Wave)	120	175	113
	PCM (Wave)	132	193	124
	MS-ADPCM (Wave)	120	175	113
	Speex (Ogg)	79	116	75
		G.729	G.711 / PCM	G.722
IPTrade, IPC, Speakerbus, BT ITS ^{1 2}	GSM-FR (Wave)	207	300	195
	PCM (Wave)	228	330	215
	Speex (Ogg)	137	198	129
		G.729	G.711	G.722
Cisco Proxy and Dial-in recording, ACME Packet / Oracle, Cisco CUBE, Sonus, MetaSwitch SBCs, and other SIP/SIPREC recording	GSM-FR (Wave)	120	175	113
	PCM (Wave)	132	193	124
	MS-ADPCM (Wave)	120	175	113
	Speex (Ogg)	79	116	75
		G.729	G.711	G.722
Network port mirroring based SIP/SCCP recording	GSM-FR (Wave)	170	250	160
	PCM (Wave)	187	275	176
	MS-ADPCM (Wave)	170	250	160
	Speex (Ogg)	112	165	106
		G.729	G.711	G.722
Avaya ¹	GSM-FR (Wave)	207	300	195
	PCM (Wave)	228	330	215
	Speex (Ogg)	137	198	129

¹ Simplex stream recording, e.g recording channel mixing on turrets

² BT ITS supports PCM streams only

Sample calculations

Scenario

CPU sizing

Skype for Business recording for 1000 users

Skype for Business uses Silk for most call scenarios
GSM-FR is the recommended storage codec
To plan for the worst case, we assume 1000 simultaneous calls (duplex streams)
Encryption and voice quality checks are not required
CPU requirement: 1000 / 140 = 8

Trader voice recording for 500 users

Trader voice platforms use G.711 simplex streams for most call scenarios
GSM-FR is the recommended storage codec (recorders support VAD to filter out long silence)
To plan for the worst case, we assume 4 simplex recording streams per turret (depends on the recording channel mixing configuration)
Encryption and voice quality checks are both required
CPU requirement: 500 x 4 / (300 x 0.9^* x 0.85^**) = 9

^* ^{The performance multiplier of the Encryption process
** The performance multiplier of the Voice Quality Check process}

Voice recording on co-located Proxy and Media Recorder server CPU sizing

A standalone proxy server can handle 200 concurrent sessions per CPU core in the case of Skype for Business, and 150 in the case of Cisco. The following table shows the sizing in the case of co-located proxy and recording services.

Storage Codec / Network Codec
Recorded Platform		Silk	G.711	G.722
Skype for Business	GSM-FR (Wave)	82	111	89
	PCM (Wave)	87	116	94
	MS-ADPCM (Wave)	82	111	89
	Speex (Ogg)	63	90	69
		G.729	G.711	G.722
Cisco Proxy-Based Recording	GSM-FR (Wave)	56	81	54
	PCM (Wave)	60	74	57
	MS-ADPCM (Wave)	56	70	54
	Speex (Ogg)	43	55	41

Video recording sizing

Video recording CPU sizing

When recording video, besides the number of recorded endpoints, the video bandwidth also has to be taken into account.

With a single CPU core, 125 Mbps total video stream can be recorded. So with the minimum server requirements (4 cores, 2 cores considered as reserved for the OS), the system can record up to 250 Mbps video streams total.

The bandwidth usage of the video endpoints can vary based on the device types and their configuration. The following table shows the bandwidths with different video resolutions:

Video resolution and framerate	Bandwidth
Video resolution and framerate	Skype for Business	Cisco
360p @ 30 fps	300 - 800 Kbps	300 - 600 Kbps
480p @ 30 fps	400 - 1500 Kbps	600 - 800 Kbps
720p @ 30 fps	700 - 2500 Kbps	1300 - 2000 Kbps
1080p @ 30 fps	1500 - 4000 Kbps	2000 - 4000 Kbps

For more information, refer to the following articles:
https://docs.microsoft.com/en-us/skypeforbusiness/plan-your-deployment/network-requirements/network-requirements
https://www.cisco.com/c/en/us/td/docs/voice_ip_comm/cucm/srnd/collab12/collab12/cac.html

It is important to note, that the numbers above have to be multiplied by two because there is always an incoming and an outgoing stream.

Calculating with 2*2,5 Mbps 720p streams per endpoint, with a single CPU core, 25 concurrent video session can be recorded. Therefore, a 4-core Recording Server (2 cores considered as reserved for OS) can record 50 concurrent video sessions.

Numbers only applicable when Receive-side scaling (RSS) is enabled in the OS

Video recording and storage devices

In the case of a high number of recorded video endpoints, the type of the storage device also has to be considered. An HDD usually has a write speed of 50-120 MB/s, while an average SSD has 200-520 MB/s, or even more in the case of high-end devices. The disk utilization for the recording shouldn't be more than 60% and the disk must provide sustained I/O performance at this rate. Numbers only applicable when Receive-side scaling (RSS) is enabled in the OS.

The following table shows some examples for maximum total video bandwidth per Recording Server:

Storage type and maximum write speed	Maximum video bandwidth at 60% disk utilization and required recorder cores	Maximum number of recorded video endpoints (calculating with 2*2,5 Mbps streams per endpoint)
HDD - 50 MB/s	240 Mbps / 2 cores	48 endpoints
HDD - 80 MB/s	384 Mbps / 4 cores	76 endpoints
HDD - 120 MB/s	576 Mbps / 5 cores	115 endpoints
SSD - 200 MB/s	960 Mbps / 8 cores	192 endpoints
SSD - 350 MB/s	1680 Mbps / 14 cores	336 endpoints