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

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

st l<strong>in</strong>k, VC 15 is limited to a throughput of less than 1/15 <strong>the</strong> l<strong>in</strong>k rate. VC 16 and<br />

VC 17 should, <strong>the</strong>re<strong>for</strong>e, each converge to a little less than 7/15 of <strong>the</strong> second l<strong>in</strong>k<br />

rate. The con guration is called an upstream con guration because <strong>the</strong> bottleneck<br />

l<strong>in</strong>k is <strong>the</strong> rst l<strong>in</strong>k (upstream l<strong>in</strong>k). A WAN con guration (1000 Km l<strong>in</strong>ks) is used<br />

<strong>in</strong> this situation to illustrate <strong>the</strong> scalability of ERICA to long distances.<br />

Figure 6.17 shows that <strong>the</strong> orig<strong>in</strong>al ERICA algorithm was unfair <strong>in</strong> this situation,<br />

and gure 6.18 shows that ERICA, after <strong>the</strong> modi cation discussed <strong>in</strong> section 6.2, is<br />

fair. As seen <strong>in</strong> gure 6.18, <strong>the</strong> modi ed algorithm converges to max-m<strong>in</strong> allocations.<br />

Regardless of <strong>the</strong> <strong>in</strong>itial load factor value, after a short transient period, all sources<br />

contend<strong>in</strong>g <strong>for</strong> bandwidth are allocated equal rates, and <strong>the</strong> two curves <strong>in</strong> gure<br />

6.18(b) (number of cells received at <strong>the</strong> dest<strong>in</strong>ation) have <strong>the</strong> same slope (compare<br />

with gure 6.17(b)). The transient responseisslightly worse than <strong>the</strong> orig<strong>in</strong>al ERICA<br />

algorithm due to <strong>the</strong> temporary over-allocation needed to equalize <strong>the</strong> shares, but <strong>the</strong><br />

steady state per<strong>for</strong>mance is as good as with <strong>the</strong> orig<strong>in</strong>al ERICA algorithm.<br />

6.22.5 Transient and Steady State Per<strong>for</strong>mance<br />

To test <strong>the</strong> transient response of <strong>the</strong> system, we use a modi ed two source con-<br />

guration. The con guration is similar to <strong>the</strong> two source con guration because two<br />

sources share <strong>the</strong> same l<strong>in</strong>k, but one of <strong>the</strong> sources is only active from 10 ms to<br />

20 ms while <strong>the</strong> o<strong>the</strong>r source is active throughout. Besides illustrat<strong>in</strong>g <strong>the</strong> transient<br />

response of <strong>the</strong> system, this con guration also illustrates <strong>the</strong> e ect of <strong>the</strong> \fairshare<br />

rst" algorithm discussed <strong>in</strong> section 6.3. That algorithm (see section 6.3) prevents a<br />

low rate VC to rise above F airShare. This VC takes an extra round trip compared<br />

to <strong>the</strong> basic ERICA because it rst comes to F airShare be<strong>for</strong>e ris<strong>in</strong>g fur<strong>the</strong>r. The<br />

192

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

Saved successfully!

Ooh no, something went wrong!