24.11.2012 Views

Traffic Management for the Available Bit Rate (ABR) Service in ...

Traffic Management for the Available Bit Rate (ABR) Service in ...

Traffic Management for the Available Bit Rate (ABR) Service in ...

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Notice that max-m<strong>in</strong> allocation is both fair and e cient. It is fair <strong>in</strong> <strong>the</strong> sense<br />

that all sources get an equal share on every l<strong>in</strong>k provided that <strong>the</strong>y can use it. It is<br />

e cient <strong>in</strong> <strong>the</strong> sense that each l<strong>in</strong>k is utilized to <strong>the</strong> maximum load possible.<br />

When we take <strong>the</strong> m<strong>in</strong>imum cell rate (MCR) of sources <strong>in</strong>to account, <strong>the</strong>re are<br />

several possible optimality criteria. O<strong>the</strong>r criterion such as weighted fairness have<br />

been proposed to determ<strong>in</strong>e optimal allocation of resources over and above MCR<br />

[32].<br />

Abraham and Kumar [1] develop a natural extension of <strong>the</strong> concept of max-m<strong>in</strong><br />

fair rate allocation to <strong>the</strong> case of <strong>ABR</strong> sessions with non-zero MCRs. Speci cally,<br />

<strong>the</strong> feasibility condition <strong>in</strong>cludes <strong>the</strong> fact that every VC's rate is at least its MCR�<br />

<strong>the</strong> max-m<strong>in</strong> criteria is <strong>the</strong> same: <strong>the</strong> network is considered to be <strong>in</strong> a state of max-<br />

m<strong>in</strong> fairness if it is impossible to <strong>in</strong>crease <strong>the</strong> rate of any session, while ma<strong>in</strong>ta<strong>in</strong><strong>in</strong>g<br />

feasibility, without decreas<strong>in</strong>g <strong>the</strong> rate of sessions whose rate is equal or smaller. The<br />

characterization <strong>in</strong> terms of rate vectors is also <strong>the</strong> same, i.e., a rate vector is max-<br />

m<strong>in</strong> fair is it is lexicographically <strong>the</strong> largest among all feasible rate vectors. The<br />

authors also develop centralized and distributed algorithms to achieve this max-m<strong>in</strong><br />

allocation.<br />

F<strong>in</strong>ally, it should be po<strong>in</strong>ted out that all de nitions of fairness assume that<br />

<strong>the</strong> tra c sources always have data to send (i.e., are <strong>in</strong> nite sources). For tra c<br />

which is \bursty" (i.e., conta<strong>in</strong>s active and idle periods), <strong>the</strong> concept of fairness is<br />

ill-de ned. As a heuristic, <strong>the</strong> de nitions should be rephrased <strong>in</strong> terms of <strong>the</strong> through-<br />

puts achieved by sources. Source throughput is measured over a long time <strong>in</strong>terval<br />

(cover<strong>in</strong>g many idle and active <strong>in</strong>tervals) and not approximated as a series of <strong>in</strong>stan-<br />

taneous rate allocations. In o<strong>the</strong>r words, \fairness" is a long-term goal. While we<br />

45

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!