"... In PAGE 12: ...Table 5: PlanetLab results with moderate capacity distribution tions performed in the burst mode here to about 2500. Figure 8 shows the variation of average node degrees, message loads and the relative selection frequencies with time in a high-churn moderate capacity experiment , and Table5 summarizes both the high-churn and low-churn experiments. While the node-degree curve in the high churn case is not completely stable, due to the high churn, all the values nevertheless adhere reasonably closely to the desired 5:10:20 ratio.... ..."

"... In PAGE 41: ...Table 11: Capacity and performance (scale V) Table11 suggests that the synaptic storage problem is four orders of magnitude more chal- lenging than achieving adequate performance. It also suggests that achieving comparable computational bandwidth is perhaps feasible, but that comparable storage capacity will re- quire a technological breakthrough.... ..."

"... In PAGE 11: ... Over- all, these experiments demonstrate that Swaplinks is ro- bust to various kinds of network churn under widely dif- ferent capacity distributions, and that it manages to re- tain its fine-grained sensitivity to the desired heterogene- ity under these conditions. Table4 summarizes KRB results from the last 175 sec- onds of the flash-crowd and the mass departure simula- tions for only the moderate capacity distribution under high churn. The flash-crowds and mass departures oc- cur at the same times as those reported in the Swaplinks experiments.... ..."

"... In PAGE 8: ... This, cou- pled with the fact that Swaplinks also realizes capacity- wise selection distribution (Figure 1(c)), demonstrates that the selection mechanism realizes the desired distri- bution. Table1 gives a summary of results from all of the Swaplinks experiments in this section by averaging each value over the second half of the experiment time. The duration of high-churn experiments here is around 930 2007 USENIX Annual Technical Conference USENIX Association... In PAGE 11: ... Figure 6 shows that while there is a temporary deteri- oration in all the metrics of interest for a short duration of time immediately after the entry of the flash-crowd, the system quickly recovers to re-establish desired be- havior. The Swaplinks graph in fact generally benefits from nodes entering the system, since this pushes the average degree distribution across the graph towards the ideal value; a comparison of Table 3 with Table1 shows that the average values for the degree and relative selec- tion frequencies in fact improve as a result of the arrival of the flash-crowd! Figure 7 and Table 3 show results of the Swaplinks mass departure experiments. The mass departures oc- cur at 649 seconds after system start, and burst selections are performed at 719 seconds after system start.... In PAGE 11: ... The message loads and the selection frequencies recover to re-approach the desired 1:2:4 split of message loads and selection fre- quencies. The extreme capacity results from Table 3 also look encouraging: the degrees and the relative se- lection frequencies are similar to the high (stable) churn, extreme-capacity results shown earlier in Table1 . Over- all, these experiments demonstrate that Swaplinks is ro- bust to various kinds of network churn under widely dif- ferent capacity distributions, and that it manages to re- tain its fine-grained sensitivity to the desired heterogene- ity under these conditions.... ..."

"... In PAGE 8: ...5, the performance is affected by the ratio of ALU latency to the instruction issue cycle time. So we made three assumptions on operation latency as shown in Table2 . Note that actual denotes the average latency of a synthesized circuit with the process we use.... In PAGE 8: ... future : This is intended to reflect the latency which is reduced by scaling with the minimum feature size in future technologies. Table2 does not contain the load / store operation la- tency. The load / store operation consists of Address Calc (add) and Data Cache Access .... ..."

"... In PAGE 8: ... Consequently, deeper pipelines are required as a fixed-capacity microarchitecture is scaled to smaller technologies. In Table6 , we show the parameters for the capacity scaling ex- periments. Because the access penalties are held nearly constant, the capacities of the structures decrease dramatically as smaller technologies and higher clock rates are used.... In PAGE 8: ... In such cases, we in- crease the access penalty slightly, permitting a structure that is large enough to be useful. The access penalties are shown in the sub- scripts in Table6 . Note that for technologies smaller than 130nm, no structure can be accessed in less than two cycles for the f8 and fSIA frequency scales.... In PAGE 9: ... One example of this effect is evident for capacity scaling at fSIA clock rates when moving from 130nm to 100nm. In Table6 , the branch predictor, L1 I-Cache, and L1 D- Cache all become noticeably larger but one cycle slower. A similar effect occurs for fSIA capacity scaling at the transition from 70nm to 50nm.... ..."

"... In PAGE 3: ... In the case of market orientation , a mean summated score was computed containing the three market orientation sub-scales in order to derive the composite scale index labeled MARKOR. The scale validation was accomplished by the analysis of item intercorrelations, and item-to-total scale correlations ( Table1 ). The scale has acceptable reliability coefficient and the alpha exceeds the standard level ( 0.... ..."

"... In PAGE 12: ...List of Tables Table1 - Wireless communication standards.... In PAGE 18: ... These standards were adopted by different parties focusing on different technologies, user applications, and geographic regions [Rap96]. Table1 lists several popular standards. As the recent advances in very large scale integrated circuits (VLSI) technology make complex systems reliable and affordable, the popularity of wireless communications increases dramatically.... In PAGE 20: ... Table1 (cont.) - Wireless communications standards Standard Type Frequency (MHz) Multiple Access Number of Channel Channel Space Modu- lation IEEE 802.... ..."