http://www.networkworld.com/community/node/37865
The 11 application classes are as follows:
- Routing
- Voice
- Interactive Video
- Streaming Video
- Mission-Critical Data
- Call Signalling
- Transactional Data
- Network Management
- Bulk Data
- Best Effort
- Scavenger
interactive video
The interactive video class represents two-way real-time video. Cisco’s Unified Video Advantage (CUVA) client and Cisco Unified Personal Communicator (CUPC) both represent video to the desktop solutions that would be mapped to the interactive video class. Cisco’s 7985 series video phone would also be mapped to this application class. Cisco’s telepresence solution has much more stringent requirements than video to the desktop applications and is deserving of its own application class. RFC 4594 details three video classes to differentiate streaming video (broadcast), video to the desktop (multimedia conferencing) and telepresence video (real-time interactive). For more details on telepresence QoS requirements, read the QoS section of the Telepresence 2.0 SRND. Third party videoconferencing solutions (Polycom, Tanberg, Sony, etc.) would be mapped to either the interactive video or real-time interactive video class based on the application requirements. Video to the desktop applications normally have bandwidth requirements in line with the H.263 and H.264 video codecs (384kbps + 20% overhead). The delay and jitter requirements of interactive video are in line with real-time voice, but telepresence has more stringent delay and jitter requirements detailed in the Telepresence 2.0 SRND. Most real-time voice and video traffic are based on the UDP transport layer of the OSI-Rm. UDP based traffic is connectionless; if a packet is lost it will never be retransmitted. The bandwidth, delay, and jitter requirements for both voice and interactive video are as follows:
Bandwidth: Dependent upon audio and/or video codec (460kbps for 384kbps H.264 video)
Delay: 150ms (one way)
Jitter (delay variation): 20ms (jitter should be equal to the sampling rate)
Streaming video
Streaming video (broadcast) is one-way video. Broadcast television and pre-recorded distance learning video are two types of streaming video. This type of video is sometimes referred to as on-demand video. Streaming video does not have stringent delay and jitter requirements because the hard drive and memory (DRAM) of the receiving station provides a very large de-jitter (playout) buffer to ensure playback of streaming video is smooth and gapless. www.youtube.com is an example of a streaming video application. To ensure high quality playback, I click the pause button when accessing a youtube video and I wait until the playback bar under the video is red from left to right. The PC is downloading the video packets from Youtube and putting them in the de-jitter buffer (temporary hard drive space).
Delay and jitter is not of much concern when you have a large de-jitter buffer and 5 minutes to wait for the de-jitter buffer to fill up. Most streaming video applications are also TCP based which will re-transmit lost packets. The large de-jitter buffer can re-order the packets when they arrive.
mission-critical data
The next application class is mission-critical data. Mission critical data is normally reserved for one transactional data application that is of higher importance to the company than the other transactional data applications and data application such as instant messaging, telnet, remote desktop, Citrix, and highly interactive groupware applications. Mission critical data applications are usually of a more interactive nature than other important data applications. Most mission critical data applications are TCP based, but the interactive nature of the traffic requires more care than other data traffic.
Call signaling
The call signaling class will include the protocols used in IP telephony to setup and teardown telephone and video conversations. These protocols include, but are not limited to H.323, MGCP, SIP, and SCCP. The port numbers for these protocols can be find by issuing the “show ip nbar port-map†command on a Cisco router. This application class should also include CTI-QBE (Computer Telephony Integration – Quick Buffer Encoding) if any TAPI (Telephony Application Programming Interface) or JTAPI (Java TAPI) integrations exist. CTIQBE uses TCP 2748 and is not included in Auto QoS Voip or Auto QoS Enterprise. CUCM deployments using the clustering over the WAN model, must identify additional protocols in the call signaling class. Most CUCM ICCS (inter-cluster communications signaling) is marked as CS3 by default, but I prefer to not rely on defaults. The ports used in CUCM can be very large, but I have included a link in the references section of this blog. The main protocol used for ICCS between CUCM servers is SDL (signal distribution layer) which uses TCP port 8002 and 8003.
transactional data
The transactional data class includes enterprise resource planning (ERP) applications such as Peoplesoft, SAP, Oracle, SQL server, and Citrix. The transactional data class is very similar to the mission critical data class, but the traffic in the transactional data class is normally of a slightly lower priority and not interactive in nature.
network management
The network management data class includes protocols like simple network management protocol (SNMP) and service assurance agent (SAA) probes. SAA probes have been associated with the IOP SLA measurement features of Cisco IOS 12.4. I have included a link to the documentation of the 12.4 IOS IP SLA Configuration guide. The SLA configuration guide provides a useful way to measure the delay, jitter, and packet loss of the 11 different application classes in QoS.
Bulk data
Bulk data applications include bursty applications which transfer large amounts of data. Bulk data applications are TCP based which makes the traffic very resilient. Lost packets are retransmitted and out of order packets are reordered in the TCP/IP stack. The bulk data application class normally includes HTTP, E-Mail, FTP, and backup traffic. Backup traffic represents a traffic type that may be of high priority during non production hours, but not allowed during business hours. Time-based access control lists allow the administrator to distinguish backup applications based on the time of day and/or day of the week.
best effort
The best effort data class will include all of the applications that have not been identified by other application classes. This traffic is normally sent on a best effort (BE) basis with or without a bandwidth guarantee. The modular QoS CLI (MQC) will only allow the administrator to provision 75% of the interface bandwidth by default to ensure applications not classified and marked by one of the other application classes receive a share of the bandwidth and the traffic streams are not starved. The 75% rule does not guarantee 25% bandwidth to the best effort class though. The best effort class will compete for the remaining 25% of bandwidth with all of the other application classes that need more bandwidth than that which was provisioned. The only exception to the 25% remaining bandwidth contention is the priority queue used for voice over IP media and videoconferencing. The priority queue traffic is implicitly policed to the amount of bandwidth provisioned and a 200ms burst which is based on the configured priority queue bandwidth.
Scavenger data
The Scavenger data class includes peer to peer file sharing applications, Internet worms, and other nasty stuff on the network. The scavenger class concept relies on policing technology at the access layer switches. Most networks have been, or are planning to upgrade to Gigabit Ethernet technology for desktop and laptop LAN connectivity. The average utilization of these connections is normally less than 5% or 5Mbps. Any traffic above 5Mbps that is not classified and marked by the access switch classification and marking policy should be marked into the scavenger class. The scavenger traffic includes suspect traffic that may be dangerous to the network. Scavenger traffic is given a very small bandwidth guarantee ensuring this traffic does not consume resources during periods of high congestion. The use of this data class represents a preemptive measure to protect your network before the signatures of viruses or worms have been identified by security vendors.
REFERENCES
IP SLA Configuration Guide (IOS 12.4)
http://www.cisco.com/en/US/docs/ios/12_4/ip_sla/configuration/guide/hsla…
CUCM 7.0 TCP and UDP Port Usage:
http://www.cisco.com/en/US/docs/voice_ip_comm/cucm/port/7_0/CCM_7.0PortL…
Cisco Telepresence SRND:
http://www.cisco.com/en/US/docs/solutions/Enterprise/Video/TP-Book.html
IETF RFCs
http://www.ietf.org