QoS Scheduling in Cable and Broadband Wireless Networks Mohammed Hawa and David W. Petr Information and Telecommunication Technology Center (ITTC), University of Kansas Project Funded by Sprint Corporation

Presentation Outline What is QoS and QoS Management? Difference between QoS in wired and wireless networks. A scheduling architecture to support QoS in broadband wireless access networks. Infrastructure MAC Protocol Support for UGS, rtPS, nrtPS and BE Traffic Scheduler Features and Advantages Discussion Design goals are: simplicity, efficiency and support for various types of traffic including CBR, VBR and BE. © Information & Telecommunication Technology Center (ITTC), EECS, University of Kansas

Quality of Service (QoS) Packet switched networks (e.g., Internet) were designed to provide best effort service. A QoS architecture introduces tools to treat packets differently, thus one flow receives better performance on the expense of others. Provides guaranteed services to end users (better than best effort). QoS guarantees can be characterized by: Delay, delay jitter, bandwidth and error rate. The most common parameters to describe QoS guarantees include: delay, delay jitter, bandwidth and error rate. © Information & Telecommunication Technology Center (ITTC), EECS, University of Kansas

QoS Management Supporting QoS in packet switched networks requires collaboration of many components: Admission Control: Limits number of flows admitted into the network so that each individual flow obtains its desired QoS. Scheduling: Which packet gets transmitted first on the output link significantly impacts QoS guarantees for different flows. Scheduling affects delay, jitter and loss rate. Allows protection against misbehaving flows. Loss, delay and delay jitter are results of queueing effects within the network.Queueing is, of course, affected by scheduling. To provide service differentiation, First come first served (FIFO) queueing is not really suited. Instead we have to use Fair Queueing algortihms (such as WFQ, SCFQ, etc). © Information & Telecommunication Technology Center (ITTC), EECS, University of Kansas

QoS Management (Cont.) Buffer Management: Control the buffer size and decide which packets to drop. Controls packet loss rate. Many packet drop strategies including weighted Random Early Detection (RED). Congestion Control: Prevent, handle and recover from network congestion scenarios. Challenging feedback problem because Internet traffic shows self-similar behavior. © Information & Telecommunication Technology Center (ITTC), EECS, University of Kansas

IETF QoS Architectures IntServ: Resource reservation per flow using RSVP. Similar to ATM virtual connections. Problems with scalability. DiffServ: Aggregation of flows into per-hop behavior groups. Expedited forwarding and Assured forwarding. A good wireless QoS architecture should integrate with both IETF standards. IETF is the Internet Engineering Task Force. Those standards are suggested for the wired part of the Internet. RSVP is a Resource Reservation Protocol. Intserv is suited mostly for access networks. Diffserv is suited for the core network. Any wireless QoS should integrate with both of those IETF architectures. We would like to integrate with Intsev because wireless is usually an access network, and we want to integrate with Diffserv because it scales better. © Information & Telecommunication Technology Center (ITTC), EECS, University of Kansas

Wireless Networks The Internet is expanding to the wireless realm, specially in recent years: Wireless Local Area Networks (WLANs) based on the IEEE 802.11 standard, called Wi-Fi™. WLAN Hot Spots by companies such as: Boingo, Surf and Sip, T-Mobile HotSpot and Wayport. Broadband Wireless Access Networks (BWAs) based on IEEE 802.16, called WiMAX. Supported by Intel, Nokia, and Fujitsu. Need to expand QoS to the wireless side. Cell phones are now competing with regular wired phones (PSTN). Maybe wireless access will compete with wired Ethernet, etc. IEEE 802.11 is designed to emulate a wireless Ethernet. It uses the unlicensed industrial frequency band to enable multiple computers and portable devices to connect to one or more wireless hubs, thus gaining access to the Internet. For example, IEEE 802.11b allows for the wireless transmission of approximately 11 Mbps of raw data at indoor distances from several dozen to several hundred feet and outdoor distances of several to tens of miles as it uses the unlicensed 2.4 GHz band. The IEEE 802.11a uses the 5 GHz band, and can handle 54 Mbps at typically shorter distances. IEEE 802.11b networks are now heavily deployed as public short-range wireless access networks, such as those found at airports, hotels, conference centers, coffee shops and restaurants. Several companies (such as Boingo, Surf and Sip, T-Mobile HotSpot, and Wayport) currently offer paid hourly, session-based, or unlimited monthly access via their deployed networks around the U.S. and internationally. Major component and equipment manufacturers such as Intel, Nokia, and Fujitsu have recently indicated they will support WiMAX (Worldwide Interoperability for Microwave Access), which promotes the IEEE 802.16a standard for wide-area broadband wireless access. The IEEE 802.16a standard was developed as a wireless competitor to broadband Cable and Digital Subscriber Line (DSL) last-mile access networks. It has a range of up to about 30 miles with data transfer speeds of up to 70 Mbps. Third generation (3G) cellular systems are also in the beginning processes of providing high data rates to cellular phone subscribers. Although 3G is now doubtful for Internet access as opposed to IEEE 802.11. Maybe 3G will be used for niche data applications, while IEEE 802.11 is used for main Internet access (similar to what happened to Iridium by cellular technology). © Information & Telecommunication Technology Center (ITTC), EECS, University of Kansas

Wireless QoS Observations Several new technical challenges: Quality of the wireless channel is typically different for different users, and randomly changes with time (on both slow and fast time scales). Wireless bandwidth is usually a scarce resource that needs to be used efficiently (can not overprovision the wireless link). Excessive amount of interference and higher error rates are typical. Mobility complicates resource allocation. When the mobile moves, the mobile need reservations in the new cell along the new path. © Information & Telecommunication Technology Center (ITTC), EECS, University of Kansas

The MAC Protocol For wireless networks it is natural to integrate the QoS architecture with the MAC protocol. The MAC protocol coordinates communication over the shared wireless medium (and the shared HFC cable medium as well). IEEE 802.16 is the most talked about standard for broadband wireless access (BWA) systems, and is based on DOCSIS. DOCSIS is the de facto standard for delivering broadband services over HFC networks. MAC is Medium Access Control layer. This is because If the operation of the MAC protocol is to be optimized, a complete synchronization of the MAC functions (including scheduling, random access and data transmission) need to be achieved. Hence, our scheduling architecture has to flawlessly cooperate with the wireless MAC protocol to make the most efficient use of the shared wireless resources. DOCSIS stands for Data Over Cable Service Interface Specifications protocol. Versions 1.0 and 1.1 of DOCSIS were completed by 1999, and version 2.0 was introduced in early 2002. IEEE 802.16 was completed in December 2001. IEEE 802.16 is a consolidation of two proposals, one of which was based on DOCSIS. The QoS features in DOCSIS and IEEE 802.16 are identical due to the similarities between HFC and BWA systems. Although our scheduler supports both DOCSIS and IEEE 802.16, our discussion will be directed more toward the DOCSIS standard to avoid duplicity in technical terms and because of the wide availability of the DOCSIS standard at the time of writing. © Information & Telecommunication Technology Center (ITTC), EECS, University of Kansas

DOCSIS Introduction A Cable Modem Termination System (CMTS) controls many terminating Cable Modems (CMs). Upstream and downstream channels are separated using FDD. Each upstream channel is further divided into a stream of fixed-size time minislots. DOCSIS MAC utilizes a request/grant mechanism to coordinate transmission between multiple CMs. © Information & Telecommunication Technology Center (ITTC), EECS, University of Kansas

DOCSIS Operation Requests arrive at the CMTS not only through contention, but also through unicast requests and piggybacking. These are the three possible request mechanisms. © Information & Telecommunication Technology Center (ITTC), EECS, University of Kansas

QoS in DOCSIS To support QoS, DOCSIS 1.1 introduces the concept of a service flow. IEEE 802.16 defines identical upstream service flow types. Upstream Service Flow Types in DOCSIS and IEEE 802.16: Unsolicited Grant Service (UGS) Real-Time Polling Service (rtPS) Non Real-Time Polling Service (nrtPS) Best Effort (BE) The different Service Flow Types in DOCSIS are similar to the different Connection Types in ATM: UGS is similar to CBR, rtPS and nrtPS are similar to rt-VBR and nrt-VBR. Best Effort is similar to that of the Internet traffic. © Information & Telecommunication Technology Center (ITTC), EECS, University of Kansas

Real-Time Service Flows Unsolicited Grant Service (UGS): Supports real-time traffic (Voice over IP). Offers fixed size unsolicited data grants (transmission opportunities) on a periodic basis. Real-Time Polling Service (rt-PS): Supports real-time flows that generate variable size data packets on a periodic basis (MPEG). Offers periodic unicast request opportunities. The CMs specify the size of the desired data grants. In UGS, contention and piggyback requests are prohibited. Key service parameters are: Nominal Grant Interval, Tolerated Grant Jitter. In rtPS, contention and piggyback requests are prohibited. Key service parameters are: Nominal Polling Interval, Tolerated Poll Jitter, Minimum Reserved Traffic Rate. © Information & Telecommunication Technology Center (ITTC), EECS, University of Kansas

Non Real-Time Service Flows Non Real-Time Polling Service (nrt-PS): Supports flows that require variable size data grants on a regular basis (high bandwidth FTP). Offers infrequent unicast polls plus contention and piggybacking. Best Effort (BE): The CM uses contention and piggybacking only. Key service parameters for nrt-PS and BE: Minimum Reserved Traffic Rate. Traffic Priority. For nrt-PS, Key service parameters are: Nominal Polling Interval plus the Minimum Reserved Traffic Rate and Traffic Priority. © Information & Telecommunication Technology Center (ITTC), EECS, University of Kansas

The New CMTS Scheduler © Information & Telecommunication Technology Center (ITTC), EECS, University of Kansas

Scheduler Architecture Requests arrive at the CMTS. Through contention, unicast requests and piggybacking. Requests are translated into upstream data grants. Data grants are scheduled on a frame-by-frame basis by building a corresponding allocation MAP: The hardware block responsible for creating the MAP is represented by a server. Each data grant (or unicast request opportunity) is treated as a packet. Actual transmission of the corresponding data packet takes place in the next frame. Data grants are queued in three types of buffers: Type 1, Type 2 and Type 3 buffers. UGS packets cannot tolerate excessive delays, so their processing is decoupled from all other flows. A separate block maintains UGS information and uses it to periodically generate data grants that feed the Type 1 queue. Type 1 grants receive a semi-preemptive priority. Since the characteristics of rtPS data grants are different than UGS, they are treated differently. The scheduler handles two portions of rtPS traffic: Periodic upstream unicast request opportunities: Treated in a similar way to UGS traffic. Actual data grants: Treated differently. The rtPS data grants are fed to either a Type 2 or a Type 3 queue based on whether the corresponding service flow has a minimum bandwidth reservation or not. We use WFQ (or a simpler variant) to handle Type 2 and Type 3 queues. The WFQ weight assigned to each queue is based on its minimum bandwidth reservation. nrtPS is still treated in the same was as rtPS traffic (i.e., generated data grants are fed to a Type 2 or Type 3 queue based on whether a minimum bandwidth reservation is made or not). nrtPS flows can be assigned different priority levels while rtPS has only one implicit priority level. Use a priority-enhanced WFQ to make sure that higher-priority grants always incur less delay. Adopt a strict non-preemptive priority discipline in serving data grants from Type 3 queue before being handed to the WFQ global server. BE is treated in the same way as nrtPS except that no unicast requests are scheduled for any BE flows. © Information & Telecommunication Technology Center (ITTC), EECS, University of Kansas

Extra Features Dynamic Contention Minislot Allocation: An appropriate number of contention request minislots is allocated in each frame period to reduce collisions and to shorten contention resolution. A Buffer Management mechanism based on RED was also suggested. Dynamic Contention Minislot Allocation improves the MAC performance. © Information & Telecommunication Technology Center (ITTC), EECS, University of Kansas

Scheduler Advantages Easy to implement in hardware thus gaining a performance advantage over software-based alternatives. Takes advantage of the Tolerated Jitter parameter for UGS to fit as many packets as possible in the upstream frame thus avoiding fragmentation and being more efficient. Lends itself to easier and straightforward performance analysis via classical queuing theory techniques. Easier to implement in hardware because it uses a single-pass as opposed to a multi-pass approach. © Information & Telecommunication Technology Center (ITTC), EECS, University of Kansas

Future Work Stochastic analysis of FQ algorithms have not been given much attention due to difficulty in tracing the dynamics of fair queuing algorithms. Two methods of scheduler analysis: Deterministic (leaky-bucket assumptions) and Stochastic (stochastic arrival models). Stochastic analysis is important in predicting the behavior of the scheduler under normal operating conditions. Analysis led to delay bounds (paper) Simulation led to performance study and comparison of different FQ policies (paper) Validation of the analysis and comparing the architecture performance to other architectures will be done through simulation using OPNET. © Information & Telecommunication Technology Center (ITTC), EECS, University of Kansas

Summary Introduced QoS concepts in both the wired and wireless parts of the network. Summary of DOCSIS and IEEE 802.16 standards and their QoS features. Introduced a new scheduling architecture to integrate QoS within the MAC layer of such protocols. Discussed the features of the architecture and its advantages. Suggested possible future work. © Information & Telecommunication Technology Center (ITTC), EECS, University of Kansas

Discussion Thank you! © Information & Telecommunication Technology Center (ITTC), EECS, University of Kansas