### Table 9 Three-Tier PDS

"... In PAGE 63: ... Finally, such a system would take advantage of the best available subsystems for each tier. Table9 summarizes our proposed three-tier PDS based on expected development of technologies at the times of expected deployment. Figure 3-10 provides a notional picture of the three-tier proposal.... ..."

### Table 1 lists a number of potential prioritization policies that a customer might select based on its network or business needs as- suming the three-tiered priority scheme shown in Figure 2. This

2006

"... In PAGE 3: ...Table 1 lists a number of potential prioritization policies that a customer might select based on its network or business needs as- suming the three-tiered priority scheme shown in Figure 2. This a1 a1 a2 a2 a3a4a3 a3a4a3 a3a4a3 a5 a5 a5 a6a4a6 a6a4a6 a6a4a6 a7a4a7 a7a4a7 a7a4a7 prioritized traffic Low Medium High (3prios) Prioritizer Bad COI COI Good traffic a8a4a8 a8a4a8 a9a4a9 a9a4a9 a10a4a10 a10a4a10 a10a4a10 a11 a11 a12 a12 a12 a12 a13 a13 a13 discarded (100MBit/s) accepted traffic Policer neither in good nor bad coi only in bad coi only in good coi Classifier COI traffic classified a14a4a14 a14a4a14 a14a4a14 a15a4a15 a15a4a15 a15a4a15 traffic a16a4a16a4a16 a16a4a16a4a16 a16a4a16a4a16 a17a4a17a4a17 a17a4a17a4a17 a17a4a17a4a17 in good and bad cois Figure 2: Traf c classi er model Policy COI Prio a18 Traf c source Higha18 good COI goodcoi Good Low a18 remainder Higha18 remainder badcoi Bad Low a18 bad COI Higha18 good COI a19 bad COI 3prios Good Med a18 remainder Bad Low a18 bad COI a19 good COI puregoodcoi Good Higha18 good COI a19 bad COI Bad Low a18 remainder purebadcoi Good Higha18 remainder Bad Low a18 bad COI a19 good COI none none Meda18 All Table1 : Simulated customer policies list is by no means complete, rather we chose a number of simple policies that can be readily implemented to investigate the utility of our approach. The goodcoi and badcoi policies represent what a customer or ISP might do if they had access to either the good COI or the bad COI, but not both (e.... In PAGE 6: ... The prioritizer then prioritizes the traf c classes according to the policy under test. The example in Figure 2 shows the 3prios policy, which places traf c into three priority classes (see Table1 ). Note that 3prios is just one possible policy; the number of priority classes as well as the mapping from COI classes to priority classes varies from policy to policy.... ..."

Cited by 3

### Table 1. Experimental results for various con gurations of the three-tiered (D-S-C), resource allocator (R-C) and hypercube (HC) speci cations

"... In PAGE 6: ... In order to compare strategies it is possible to select the strategy used (rather than let TopSPIN choose the most e cient). 5 Experimental Results Table1 gives experimental results applying our techniques to three families of Promela speci cations. For each speci cation, we give the number of model states without symmetry reduction (orig), with full symmetry reduction (red), and using the strategy chosen by TopSPIN (best).... ..."

### Table 4: Three-Tiered Accessibility Ranking of Popular Web Sites

2001

"... In PAGE 4: ... Inaccessible: Sites with 2 and above Priority 1 accessibility problems. Table4 lists the fifty universities terms of their accessibility rankings based on the three categories defined above. A qualitative representation of the obtained significant correlation between accessibility and usability is provided in Table 6.... ..."

Cited by 1

### Table 5: Three-Tiered Usability Ranking of Popular Web Sites

2001

"... In PAGE 4: ... A qualitative representation of the obtained significant correlation between accessibility and usability is provided in Table 6. Sites listed in Table5 include only those sites that ranked in the top or bottom tier (Tier 1 or Tier 3, respectively) both in terms of usability and accessibility. Tier 1 Highly Accessible (Approved) Tier 2 (Not Approved: 1 P1 error) Tier 3 (Not Approved: 2 or more P1 errors) University of California Irvine Univ.... ..."

Cited by 1

### Table 1: Risk calculations of a meta-analysis which fails to adjust for publication bias, R0( ?), and the augmentation method, RA( ?), on simulated data in three tiers.

1997

"... In PAGE 9: ... 3.2 Results of the Simulations Table1 shows the results of the risk calculations. Table 2 shows the posterior mean and a 95% credible interval (CI) for in each parameter/suppression combination when = 0.... In PAGE 9: ... Table 3 gives these results for = 0:4. Table1 indicates that the posterior of resulting from a standard Bayes meta-analysis is concentrated near the true when there are no studies missing. In those cases where publication bias is present, the meta-analysis adjusted for publication bias decreases the ratio of the risks.... In PAGE 9: ... This is partly due to the augmentation algorithm generally underestimating the true log RR in these cases and partly due to the fact that the credible intervals for the standard Bayesian meta-analysis are more narrow than those of the augmentation algorithm. However, Table1 shows that RA( ?) is more consistent across the suppression schemes than R0( ?). The R0( ?) posterior risk increases as more studies are suppressed, which is the expected e ect of publication bias.... ..."

Cited by 1

### Table 1: Risk calculations of a meta-analysis which fails to adjust for publication bias, R0( ?), and the augmentation method, RA( ?), on simulated data in three tiers.

"... In PAGE 10: ... 3.2 Results of the Simulations Table1 shows the results of the risk calculations. Table 2 shows the posterior mean and a 95% credible interval (CI) for in each parameter/suppression combination when = 0.... In PAGE 10: ... Table 3 gives these results for = 0:4. Table1 indicates that the posterior of resulting from a standard Bayes meta-analysis is slightly more concentrated near the true when there are no studies missing as compared to the data augmentation technique. This is partly due to the augmentation algorithm generally underestimating the true log RR in these cases and partly due to the fact that the credible intervals for the standard Bayesian meta-analysis are more narrow than those of the augmentation algorithm.... In PAGE 10: ... In those cases where publication bias is present, the meta- analysis adjusted for publication bias generally decreases the ratio of the risks and favors the augmentation method. In addition, Table1 shows that RA( ?) is more consistent across the suppression schemes than R0( ?). The R0( ?) posterior risk increases as more studies are suppressed, which is the expected e ect of publication bias.... ..."

### Table 3: Results of a meta-analysis which fails to adjust for publication bias (\Std quot;) and the augmentation algorithm (\Aug quot;) on simulated data in three tiers. The mean and 95% credible limits come from the marginal posterior of . The true is 0.40.

"... In PAGE 10: ... Table 2 shows the posterior mean and a 95% credible interval (CI) for in each parameter/suppression combination when = 0. Table3 gives these results for = 0:4. Table 1 indicates that the posterior of resulting from a standard Bayes meta-analysis is slightly more concentrated near the true when there are no studies missing as compared to the data augmentation technique.... In PAGE 10: ... The aug- mentation algorithm apos;s credible intervals are wider than those of the standard meta-analysis, presumably because that analysis accounts for an additional source of uncertainty. In those data sets where = 0:4 ( Table3 ), the overall number of studies suppressed was smaller compared to those where = 0 since there were fewer studies in I2 and I3. The standard Bayesian meta-analyses and the augmentation method covered the true in all... ..."

### Table 3: Results of a meta-analysis which fails to adjust for publication bias (\Std quot;) and the augmentation algorithm (\Aug quot;) on simulated data in three tiers. The mean and 95% credible limits come from the marginal posterior of . The true is 0.40.

"... In PAGE 10: ... Table 2 shows the posterior mean and a 95% credible interval (CI) for in each parameter/suppression combination when = 0. Table3 gives these results for = 0:4. Table 1 indicates that the posterior of resulting from a standard Bayes meta-analysis is slightly more concentrated near the true when there are no studies missing as compared to the data augmentation technique.... In PAGE 10: ... The augmentation algorithm apos;s credible intervals are wider than those of the standard meta-analysis, presumably because that analysis accounts for an additional source of uncertainty. In those data sets where = 0:4 ( Table3 ), the overall number of studies suppressed was smaller compared to those where = 0 since there were fewer studies in I2 and I3. The standard Bayesian meta-analyses and the augmentation method covered the true in all cases.... ..."

### Table 2: Results of a meta-analysis which fails to adjust for publication bias (\Std quot;) and the augmentation algorithm (\Aug quot;) on simulated data in three tiers. The mean and 95% credible limits come from the marginal posterior of . The true is 0.

"... In PAGE 10: ...2 Results of the Simulations Table 1 shows the results of the risk calculations. Table2 shows the posterior mean and a 95% credible interval (CI) for in each parameter/suppression combination when = 0. Table 3 gives these results for = 0:4.... In PAGE 10: ... The R0( ?) posterior risk increases as more studies are suppressed, which is the expected e ect of publication bias. In those data sets where = 0 ( Table2 ), suppressing studies in a standard meta-analysis results in an elevated point estimate of . As more studies are suppressed, the bias of the standard meta-analysis increases.... ..."