Tele/Presence News - Technical Designer's Guide to QoS via the OSI Model for Tele/Presence (Video) & SIP Protocol (Voice) Animated Indepth Tutorial

Technical Designer's Guide to OSI-Open Systems Interconnection
For Tele/Presence (Video) & SIP-Session Initiation Protocol (Voice) Indepth Tutorial

This IS an animated tutorial - click here.

The point of this tutorial is that unless there is QoS-Quality of Service efforts "effectively and efficiently" applied at all 7 Layers of the OSI-Open System Interconnection model, the communications sessions may be disrupted or disconnected. More importantly, even small problems with voice echo, video screen jitter and many others can cause users to be confused, annoyed or worse, question the usability of the voice or video system. That is, hundreds of millions have been spent on video systems for more than thirty years and it remains a less-than-widely accepted technology.

Applications Layer 7 and Presentation Layer 6 transform, convert or digitize human interactions such as mouse, keyboard, video, voice and other UI-user interface functions into digital communications 1-0's. We see the video "jitter" and hear the voice "stutter" both may also have echo, clipping/dipping (see below), electrical surges/sags and a vast array of other events which can cause disruption of the communications session.

Session Layer 5 RTP-Realtime Transport Protocol is used to carry the video, voice and "presence" in SIP-Session Initiation Protocol. The video CODECs-compression-decompression don't measure jitter at the packet Layer 3, but compare the RTP timestamp (see above) at Layer 5 and video frame at Layer 7. The CODECS compare the first video packet of a video frame to the timestamp of the last packet of the same video frame, taking latency network "flight" time into account. While frame size may vary due to image content, all frames are expected within normalized ranges. Endpoints such as video systems, telephones and a vast array of other SIP devices such as appliances and yet-to-be developed systems should have network monitoring software installed and tested often. Measurement of RTP is provided by RTCP-Realtime Transport Control Protocol. Since video/voice/presence is not the only types of packets can impact performance, continual network monitoring is critical. RTCP-XR-eXtended Reports MRB-Metrics Report Block provides measurements (metrics) for monitoring quality of video/SIP calls and conversations. These measurements include packet loss and discard metrics, delay metrics, analog metrics, video and voice quality metrics. The Metrics Report Block reports individually on packets lost (discarded) on the IP channel as opposed to packets that have been received and then lost by the receiving jitter buffer. MRB reports on the combined effect of losses and discards which can be used to determine corrective actions on voice/video QoS.

Transport Layer 4 - TCP-Transmission Control Protocol is used to manage data connections in "connection-oriented" segments and to get ACKnowledgements before sending more data using a concept called a "sliding window" where data is sent like a window being "slid" open and when an ACK is returned the window is "slid" back and so on.

Transport Layer 4 - UDP-User Datagram Protocol is used to provide streaming "connection-less" communications such as audio, video, telephony media where reliability or QoS is presumed to be available and waiting for ACK processes found in TCP would hinder the communications session.

Layer 4 is also critical as firewalls generally operate at this level and for effective network design, knowledge of key TCP/UDP "ports" is required. Here are some common TCP-UDP "Ports of Call." Ports are used to designate user access to features-software on host servers and mainframes. For example, according to Cisco (click here for article), "Dynamically closing the media ports has been observed to cause temporary blocking of a large number of TelePresence audio and video media packets, since the codecs typically send SIP BYE messages before actually stopping the media flows."

Just as a ship entering the harbor, it has to know what port it will dock (connect) to. Some common ports for tele/presence and SIP are:

20 - FTP - file transfer protocol data

21 - FTP - file transfer protocol control

23 - SSH - secure sockets shell

25 - SMTP - simple mail transfer program - email

35 - Private print server

38 - Remote access protocol

53 - DNS - domain name service

67-68 - DHCP - dynamic host configuration protocol

69 - TFTP - trivial file transfer protocol

162 - SNMP - simple network management protocol

80-8080 - HTTP

443 - HTTPS - hypertext transfer protocol secure

109 - POP - post office protocol

161 - SNMP - simple network management protocol

5060-5061 - SIP - session initiation protocol

6970 - Cisco tele/presence IP phone configuration files and images

16384-16388 to 32677 - Cisco media streams

Note: From Ports 0 and 1023 are for public use.

There are 65,535 ports total, assigned by ICANN.

Network Layer 3 - is the IP-Internet Protocol packet before and with the MPLS "label" attached or "tagged" on as it was originally called. MPLS consists of four elements, label bits, experimental bits, a stack bit and TTL-Time-To-Live bits which indicate the number of Label Switch Routers passed. To begin with, IP-Internet Protocol packets may have a number of labels or "tags" attached to them. MPLS-Multi-Protocol Label Switching is just one type of label. In a Provider Provisioned Virtual Private Network known as PWE3 or PPVPN, there may be more than one label. Here are some terms associated with labeling:

- Push - add a label

- Swap - replace the label

- Pop - remove the label

Datalink Layer 2 - Ethernet is the most common LAN-local area network protocol designed for sharing devices inside an office. Ethernet is a standardized protocol managed by the IEEE 802.3 (10 MBPS) - 802.3u (100 MBPS) - 802.3ae (1 & 10 gigabits). For example, to provide QoS, 802.1Q. is 4-byte packet called TCI-Tagged Control Information which is inserted (added) to Ethernet frames that includes VLAN - 12 bits assignment and the 802.1p QoS. 802.1p consists of 3 bits within an 802.1Q header, and the resulting 8 possible QoS values (0-7) signal the type of traffic that the Ethernet frame contains - from background to network critical. However, to understand the inner workings of Ethernet is important to how effectively QoS will be. Ethernet uses CSMA-CD-CS approach to data communications. Carrier Sense means everyone (node) listens, MA - Multiple Access - every node (endpoint) is treated "democratically"and CD - Collision Detect or truncated binary exponential backoff (wait) and try again. Analogous to a human conversation or "cocktail party" - everyone listens, then talk, if both talking, both pause, then one starts again - Ethernet is a passive-aggressive network. Another way to look at it "like crossing a street" - listen, look, if busy then wait, if not busy wait for a semi-random time, then transmit. This means that there is an unpredictable access window or wait time and unpredictable performance or no QoS. Ethernet was originally designed to share printers on a LAN where there was no need for QoS just reliable communications.

Physical Layer 1 - Serialization - multiplexer digital equipment processing delays is the result of digital signal processing of the DataLink Layer 2 data frame onto the Physical Layer 1 network interface. Serialization rate is the size of the data frame (packet) divided by the clocking speed (bandwidth rate) of the network interface. Related to serialization is LFI-Link Fragmentation and Interleaving, used to fragment large frames into smaller "same-size" packets so end-to-end delay can be modeled or predicted. That is, managing serialization can reduce delays and resulting jitter by reducing traffic size that may be delayed behind large non-delay sensitive data traffic.

Here are some but certainly not all of the various types of delay (queuing) that can occur in a Tele/Presence and SIP network:

1 - Coding-CODEC-compression/decompression/decoding - DSP-Digital Signal processing - compression and analog-to-digital processing

2 - Queuing delay - network administrator router priority queues/router configurations

3 - Variable packet sizes - different packet sizes such as variable length video packets

4 - Packetization - encapsulation of Layer 5 RTP/SRTP -- Layer 4 UDP -- Layer 3 IP -- Layer 2 Ethernet/ATM/PPP/ISDN -- Layer 1 OC-1/SONET

5 - Serialization - multiplexer digital processing delays

6 - Network "Flight Time" Propagation - speed of light divided by distance or ~60 ms-milliseconds across the continental U.S. plus multi-hop (router) delays - average number of routers passed is six domestically. This can amount to 120+ ms of delay (Cisco articles suggest common delays of 138-210 ms). However, above 150 ms, users get annoyed and confused thinking the connection has been disconnected or lost. The speed of radio waves, electricity and light is 300,000,000 meters per second or 186,000 miles per second in free space. For example, the distance between Boston and San Diego is 3,000 miles. Without any delay, RTT-Round Trip Time for voice or data transmission would take 32 ms. On fiber optic systems, the speed is half of light or 64 ms. Here are some of the types of delays found in various network tests. Actual delay results may be considerably higher. CODEC-COmpression-DECompression (computer processing) of different voice compression protocols such as G.723.1: 40 ms, G.729: 10-20 ms, G.728: 2-5 ms. Router processing hand-offs are approximately 10 ms at each end router (hop). Adding in ~240 millisecond delays for satellite transmission generally makes satellites undesirable for real-time communications.

7 - Memory Buffers - memory overflow and out-of-order packet processing/reprocessing including jitter and de-jitter buffers.

8 - Bandwidth - there are two types of bandwidth:

- LAN-local area network. While LAN speeds are often Ethernet of 100 megabits or more, the effective throughput of Ethernet is 50% or less and can be considerable less if a large number of users are on the same LAN segment and VLAN-virtual LAN configuration is not implemented. This suggests that while it appears there is lot of bandwidth on the LAN, if not configured properly, it may not be enough. Remember Ethernet has no QoS and providing QoS on LANs is expensive.

- WAN-wide area network. WAN bandwidth for voice (ROM-rough order of magnitude) is 100 kbps-kilobits per second for each voice channel (80 kbps including RTP overhead plus 20 kbps or 20% for spare). For video, including one audio channel is 2 mbps-megabit per second per video screen (in Cisco 720p is 1628 kbps to 15,307 kbps for 1080p for three screens - in Polycom 832 Kbps for 720p and 1024 Kbps for 1090p). Cisco recommends 20% extra spare bandwidth (over-provisioning). The reason for the extra bandwidth can found in issues 1-7 above and that video packet sizes vary proportionately to the degree of movement of the participants moving about (e.g. waving their hands consumes more bandwidth than just talking). Most importantly, the existing WAN infrastructure is often already at capacity and adding thousands of voice users along with even a few video rooms will likely result in not just additional bandwidth but an upgrade to all aspect of the network. However, 2 mbps is just a planning guide, not an absolute.

Note: All vendors argue that video tele/presence systems can use a lot less bandwidth, however, having excess bandwidth provides great assurance that when the network faces a "traffic pileup" there may be enough for it to work effectively.

In the next few examples of TCP/IP in action, you can see that there are functions at EACH Layer that can improve, inhibit or deny access and QoS. In other words, managing QoS is a complex process. This means that there may be likely multiple professionals to get involved in this process.

We hope this tutorial was of benefit to you and your design efforts.