320 likes | 413 Views
Quality of Service I: Definitions, Services and Architectures. A NDREW T . C AMPBELL Dept. of Electrical Engineering Columbia University http://comet.columbia.edu/campbell campbell@comet.columbia.edu. LECTURE 9. 6 November, 1998. Reality check. Project proposal
E N D
Quality of Service I: Definitions, Services and Architectures ANDREW T. CAMPBELL Dept. of Electrical Engineering Columbia University http://comet.columbia.edu/campbell campbell@comet.columbia.edu LECTURE 9 6 November, 1998
Reality check • Project proposal • New Intel Lab open for work • Quiz this Tuesday 1.5 hrs. • Take home programming assignment • tutorial Thursday on it • released Friday on Web/email to class • returned Sunday midnight (mail it to mk)
QOS I and II • QOS II - a survey • Basics; language, definitions, survey • Architectures: services and mechanisms • QOS II - differential services
Observations • Networks, distributed computing the advent of multimedia and the web call for stronger QOS paradigms than todays best effort • What is the QOS that fits? • Is there one QOS for all? • What does QOS mean to the user and the network? • The emergence of end-to-end QOS: a unified approach • The failure of ATM and IEFT model - so where now? • 10 years of research, many fundamental problems resolved but QOS is still in a major state of flux, why?
What is Quality of Service? • Means many things to many people • multimedia driver, audio and video coding, operating systems, communication protocols, networks, scheduling and traffic management issues, ORB QOS, architectures, mechanisms, and on and on
Different notions of Quality of Service • Different notion of QOS • user view point is perceptive (e.g., MOS testing) • application-level QOS • end-system based QOS • network based QOS • simple network [Keshav,92] definition from QOS workshop • ‘network quality sufficient to satisfy users needs, however they may expressed’
utility bandwidth Application-level QOS example
End-system QOS issues • Operating system issues • Real-time scheduling, apps, threads • resources • memory, CPU and IO are resources • end-system communication issues • transport • NICs
Network-based QOS • QOS metrics • end-to-end delay • variation in end-to-end delay (jitter) • packet/cell loss • bandwidth • shared resources • NICs, links, switches • Historically outside control of user until more recently
Todays notion of Network QOS • One-demand establishment of end-to-end QOS • ‘circuit emulation’ approach • ATM network traffic classes and QOS • IETF integrated services • Other views • best effort ‘QOS’ • Adaptive QOS • Differential services QOS • Components • specify, contract and deliver QOS • admission control and services disciples key
Notion of a flow and micro-flow • Important service abstraction with QOS • Characteristics • production, transmission and consumption of a single media source (e.g., audio, video, real-time data) governed by a single statement of QOS • simplex, unicast/multicast • generally require admission control and resource reservation and support for heterogeneous QOS • micro-flows • short-lived flows that have similar characteristics and QOS demands (e.g., web transaction of real-time image)
End-to-end control • Consider streaming media from a web server to a client • QOS should apply to all flows (e.g., audio and video) from the server, across the network to clients at the point of delivery • ‘open loop’ control issues • end-to-end admission control and resource reservation • coordination of disk and thread scheduling in server • flow/rate control in end-system • packet/cell scheduling in network • monitoring and maintenance of delivered quality
Who is doing QOS research? • Distributed systems community • QOS in CORBA for example • Still in early phase • Operating systems community • substantial work done • classic scheduling, e.g., earliest deadline first • still work to do • Networking community • pioneering work • huge amount of work done • recycle going on
QOS architecture • End-to-end in nature • Unifying approach across all layers • Generalized Framework • QOS principles • QOS specification • QOS mechanisms
QOS principles • integration principle • each resource module (e.g., CPU, memory, NIC, switches, links) traversed must provide QOS configurability, resource assurances and maintenance of flows/sessions/ micro-flows • separation principle • transport, control/signaling and management • transparency principle • app shielded from complexity of underlying QOS control and management, and reduces need to embed mechanisms in app • multiple time-scales principle • guides the division of functionality between architectural modules and pertains to modeling control and management • performance principle • techniques and rules for constructing QOS driven comms, minimizing layered multiplexing, structuring protocols, etc.
QOS specification • Capturing application-level QOS and management policy • Different at each layer (apps, transport, network) • Applications • specifies what is required rather than how to achieve it • Encompasses • flow synchronization (e.g., lip synch) • flow performance (e.g., delay, jitter, loss, bandwidth) • level of service (e.g., deterministic, predicative, best effort) • QOS management policy (e.g., QOS violations, actions) • cost of service (e.g., pricing policy dictates behavior)
Services and mechanisms • Services • e.g., continuous media, real-time transaction, unreliable/reliable comms • Mechanisms • help to deliver the QOS for the service and should be configurable based on the specification and mapping functions • Characteristics and time-scales • quasi-static resource management (QOS provisioning) • establish, re-negotiate, tear-down • dynamic resource management (control and mgmt) • deals with media transfer phase
QOS provision Mechanisms • QOS mapping • admission testing • resource reservation protocols
QOS control mechanisms • Flow shaping • Flow scheduling • Flow policing • Flow control • Flow synchronization
QOS management Mechanisms • QOS monitoring • QOS maintenance • QOS degradation • QOS availability • QOS scalability
QOS architectures • A number of QOS-archietctures have been proposed • standards (IETF, ISO) • telecommunications (TINA, XRM) • computer communications (IntServ, QOS-A) • Operating systems and networking coming together • Examples • IBM Hiedleberg QOS model • Columbia’s XRM (xbind realization) • UPEN OMEGA • Int-serv Architecture • Lancaster QOS-A • ISO QOS Framework • UCB Tenet Architecture
Columbia XRM • Five distinct planes • management functions (N-plane): cover OSI functional area for network and systems management • traffic control function comprises • resource control (M-plane) • cell scheduling, call admission, call routing • process scheduling, memory management • control (C-plane) • connection management, call re-negotiation • telebase which resides in the (D-plane) • collectively represents distributed global state that is accessed and shared among all planes (e.g., routing state, call state, resource availability)
XRM’s theoretical underpinings • ATM network oriented • call setup and traffic classes supporting deterministic and probabilistic QOS guarantees • Capacity regions used for network and end-system • schedulable region for network multiplexing point that is used to build end-to-end calls with QOS • traffic classes and hyper-region representation • multimedia capacity region in end-systems • e.g., mpeg flows also hyper-region representation
Schedulable region • Real-time bin packing operation • abstraction for cells (i.e fixed sized packets) with certain traffic class characteristics • traffic classes characterized by • cell loss probability, bandwidth, delay constraints • abstraction for a multiplexing point with QOS guarantees that supports 4 cell level traffic classes • circuit emulation, voice and video, data and and network management • resources controlled • switching bandwidth and link capacity • in order to efficiently satisfy QOS requirements of the cell-level, scheduling and buffer management algorithms dynamically allocate communication bandwidth and buffer space
IEFT Integrated Services Arch. • Flow spec • a number of traffic classes came and went • predicated service, controlled load, guaranteed delay • bandwidth (no explicit loss metric) • delay (jitter considered to be resolved at end system) • Signaling • RSVP based on multicast heterogenious receiver model • Architecture • packet classifier • admission controller • scheduler • signaling
Failure of ATM and IETF models • RSVP and Int-serv ran out of steam for a number of reasons • concerns of scalability RSVP/ state management/ aggregation issues for router and high speed • ATM’s ‘circuit emulation’ model may not fit • new softer notion of QOS called differential services • in-band control • longer time resource management • telcos engineering capacity planning • Many issues to solve - jury out • ATM is the only network model for QOS but it languishes • signaling a mess and its not ubiquitous and hyped • QOS model very complex - what does cell loss probability mean to an app and customer
Lancaster QOS Architecture (QOS-A) • Hybrid architecture between telecomms and Internet model • Incorporated the notion of ‘integrated QOS’ across layers via planes • APIs with QOS particular focus on end-system transport and support for continuous media • Adaptation (action,events) • QOS mechanisms pushed into the transport as oppose to into the application as in the case of RTP
Next week • Jim Kurose’s Open Issues paper • Four approaches to providing QOS guarantee in the network • Tightly controlled approach • Approximate approach • bounding approach • observation approach • Van Jacobson’s premium services • one differential services proposal