### Table 2. Lossy compression results: percentage improvement over traditional scheme at various average RMSE values.

### Table 2a. Traditional loading and two generalized loadings on the various indices, in descending order of magnitude, for PC1. Loading

2005

"... In PAGE 49: ...Table2 b. Traditional loading and two generalized loadings on the various indices, in descending order of magnitude, for PC2.... ..."

Cited by 1

### Table 2b. Traditional loading and two generalized loadings on the various indices, in descending order of magnitude, for PC2. Loading

2005

"... In PAGE 48: ...Table2 a. Traditional loading and two generalized loadings on the various indices, in descending order of magnitude, for PC1.... ..."

Cited by 1

### Table 1: Traditional estimates and bounds.

1999

"... In PAGE 11: ...xpression on the items in the set. Let v1; : : : ; vn be a uniform random sample of the multiset fx1; : : : ; xmg. We wish to estimate the aggregate (AVG, SUM, and COUNT) on all m values based on this sample of n values. Table1 summarizes the traditional estimates and the bounds for AVG, SUM and COUNT with no predicates, where p is the desired confidence probability. Shown are upper bounds for t such that Pr(je ? j t) p, where is the precise result to an aggregate, and e is an estimate based on n samples.... In PAGE 11: ... The last column indicates whether or not a bound is guaranteed with probability p or holds with probability p only under large sample assumptions [HHW97, Haa97]. Comparing the bounds in Table1 , we see that among the two bounds using ^ , the Chebychev (estimated ) bound is better than the CLT bound whenever n gt; 1=(z2 p(1?p)). Since n must be sufficiently large for either approximation to hold, the Chebychev bound is better unless the desired error probability is inversely proportional to n.... In PAGE 12: ... We report an estimate and a bound based on the ej. We can apply any of the methods in Table1 to obtain the chunk estimators, ej, and the confidence bounds on the estimators. Since each chunk estimator is based on only a subsample, the confidence in a single chunk estimator is less than if it were based on the entire sample.... In PAGE 13: ... Thus the best choice for k depends on the relationship of and t in Equation 2 as a function of k, and the desired confidence p = 1 ? qk. In the remainder of this section, we highlight our results analyzing and comparing the effects of applying the various methods in Table1 , and determining the optimal number of chunks. Table 2 summarizes our analysis on the use of Chebychev for Equation 2 in conjunction with various values for p, with and without chunking.... In PAGE 14: ... The bounds are shown for Chebychev (known ) . Alternatively, as in Table1 , we can obtain bounds for Chebychev (estimated ) by plugging in ^ for in Table 2, where ^ is computed over all the sample points, not just those in one chunk. We can also obtain bounds for Chebychev (conservative) by plugging in (MAX ? MIN)=2 for .... In PAGE 15: ...s queries without joins (i.e., as single-table queries). There are several popular methods (see Table1 ) for obtaining error bounds for approximate answers to (single-table) aggregation queries. We have presented a detailed analysis that demonstrates the precise trade-offs among these methods, as well as a method based on subsampling which we call chunking .... In PAGE 20: ... Figure 5 plots the error bounds for the PropJoin allocation scheme for a summary size of 2%. It shows the 90% confidence bounds of three of the five techniques in Table1 , namely, Hoeffding, Chebychev (estimated ), and Chebychev (conservative).10 These bounds are compared with bounds based on chunk statistics.... ..."

Cited by 116

### Table 1: Traditional estimates and bounds.

1999

"... In PAGE 11: ... We wish to estimate the aggregate (AVG, SUM, and COUNT) on all a174 values based on this sample of a146 values. Table1 summarizes the traditional estimates and the bounds for AVG, SUM and COUNT with no predicates, where a182 is the desired confidence probability. Shown are upper bounds for a130 such that a183a94a184a3a29a144a122 a158 a90a178a185 a122 a80 a130a61a32 a118 a182 , where a185 is the precise result to an aggregate, and a158 is an estimate based on a146 samples.... In PAGE 11: ... The last column indicates whether or not a bound is guaranteed with probability a182 or holds with probability a182 only under large sample assumptions [HHW97, Haa97]. Comparing the bounds in Table1 , we see that among the two bounds using a164 a165 , the Chebychev (estimated a165 ) bound is better than the CLT bound whenever a146a192a191 a1a12a10 a29a55a162 a111 a163 a29 a1a91a90 a182 a32a61a32 . Since a146 must be sufficiently large for either approximation to hold, the Chebychev bound is better unless the desired error probability is inversely proportional to a146 .... In PAGE 12: ... We report an estimate and a bound based on the a158 a156 . We can apply any of the methods in Table1 to obtain the chunk estimators, a158 a156 , and the confidence bounds on the estimators. Since each chunk estimator is based on only a subsample, the confidence in a single chunk estimator is less than if it were based on the entire sample.... In PAGE 13: ... Thus the best choice for a24 depends on the relationship of a204 and a130 in Equation 2 as a function of a24 , and the desired confidence a182 a72 a1a194a90 a157 a25 . In the remainder of this section, we highlight our results analyzing and comparing the effects of applying the various methods in Table1 , and determining the optimal number of chunks. Table 2 summarizes our analysis on the use of Chebychev for Equation 2 in conjunction with various values for a182 , with and without chunking.... In PAGE 14: ... The bounds are shown for Chebychev (known a165 ) . Alternatively, as in Table1 , we can obtain bounds for Chebychev (estimated a165 ) by plugging in a164 a165 for a165 in Table 2, where a164 a165 is computed over all the sample points, not just those in one chunk. We can also obtain bounds for Chebychev (conservative) by plugging in a29 MAX a90 MINa32 a10 a14 for a165 .... In PAGE 15: ...s queries without joins (i.e., as single-table queries). There are several popular methods (see Table1 ) for obtaining error bounds for approximate answers to (single-table) aggregation queries. We have presented a detailed analysis that demonstrates the precise trade-offs among these methods, as well as a method based on subsampling which we call chunking .... In PAGE 20: ... Figure 5 plots the error bounds for the PropJoin allocation scheme for a summary size of a14 a149 . It shows the 90% confidence bounds of three of the five techniques in Table1 , namely, Hoeffding, Chebychev (estimated a165 ), and Chebychev (conservative).10 These bounds are compared with bounds based on chunk statistics.... ..."

Cited by 116

### Table 6: Evaluation of Prevailing Directions of Enterprises apos; Production Developmentin

"... In PAGE 14: ...3 Directions of development of the surveyed enterprises The proposed variants of the answers were based on din0berent strategies of the enterprises apos; market behavior, from the most conservative | extending the range of traditional prod- ucts | to a radical change of the previous specialization and the development of other types of production and non- production activities. The results of the analysis of the answers received are as follows n28see Table6 in Appendix 2n29: 1. Despite the extremely unfavorable conditions of various economic activities, the majority of the enterprises try to preserve their traditional specialization, extending the range of products manufactured, searching for new sales markets, as well as manufacturing new types of products within the existing specialization.... ..."

### Table 1. Contrasting traditional and modern societies

"... In PAGE 2: ... The United States receives more refugees than all other countries put together (Martin, Larkin amp; Nathanson, 2000). In Table1... In PAGE 4: ... As to religious cosmology, people in traditional societies submit to and depend on the direction of a supreme entity in their daily lives, whereas indi- viduals in modern counterparts tend to place more importance on science and rationality. In contrasting primary values, it can be observed in Table1 that in developing countries, sentiment and tradition take precedence over rationality and efficiency. Residential settings of people in poor nations are generally rural, whereas in the United States, for example, most people live in urban areas.... In PAGE 4: ... Easy access to communication and transportation enables people to contact friends and associates any time anywhere. Also, listed in Table1 is literacy. In traditional societies, educational opportunities are not available to children as they are in the United States, France, England, Canada, and other Western countries, where the state requires parents to educate their children.... In PAGE 5: ... It is easy to comprehend why cultures collide when people migrate from traditional to modern societies. It should be pointed out that the contrasts shown in Table1 are general. Individuals from various parts of the world dream of becoming French, English, American, or a citizen of some other country in the developed world.... ..."

### Table 2: Qualitative comparison of various hashing functions: traditional, XOR, prime modulo (pMod), prime displace- ment (pDisp), skewed associative cache with circular-shift and XOR (Skewed) [18, 4, 19], and our skewed associative cache with prime displacement (Skewed + pDisp).

2004

"... In PAGE 6: ... To ensure inter-bank dispersion, a different prime number for each bank is used. Table2 compares the various hashing functions based on the following: a17 when the ideal balance is achieved, a17 whether they satisfy the sequence invariance prop- erty, a17 whether a simple hardware implementation exists, and a17 whether they place restrictions on the replacement al- gorithm. The major disadvantage of the traditional hashing is that it achieves the ideal balance only when the stride amount a10 is odd, where a12 a28a54a57a89 a45 a13a10 a21 a4 a4a3a73a5a48a7 a47 a35 a10 .... In PAGE 7: ... Later, we will show that skewed caches do degrade the performance of some applications. Finally, although not described in Table2 , some other hashing functions, such as all XOR-based functions and random-h [14, 15, 16, 20, 8, 7, 26, 11], are not sequence invariant, and therefore do not achieve the ideal concentra- tion. 4.... ..."

Cited by 12

### Table 1: Internet services vs. traditional applications.

2002

"... In PAGE 4: ... Although they share some character- istics with traditional desktop applications and traditional high-availability software, Internet services are built from a unique combination of building blocks, they have differ- ent requirements, and they are deployed in a different environment. Table1 summarizes these differences. Continuous 24x7 availability is demanded by a globally distributed user base that increasingly sees these services as essential parts of the world apos;s communications infrastruc- ture.... In PAGE 4: ... Continuous 24x7 availability is demanded by a globally distributed user base that increasingly sees these services as essential parts of the world apos;s communications infrastruc- ture. As suggested in Table1 , however, large-scale Internet services present an availability... In PAGE 28: ...in one of the service groups. Because the service had neither a remote console nor remote power supply control to those servers, an operator had to physically travel to the coloca- # Prob type Online -test Red Online -flt/ld Config Isol Pre- flt/ld Restrt Pre- test- 1 FE-OP XX X 2 FE-OP X 3 FE-OP X 4 FE-OP X 5 BE-OP XX 6 BE-SW 7 EXT- SW X 8 NET- HW XX 9 EXT-OP X 10 FE-SW 11 BE-HW X 12 FE-SW XX XX 13 FE-OP X 14 FE-SW X 15 FE-HW X 16 FE-HW XXX 17 FE-SW X 18 NET- OL XX X 19 BE-HW X TOTAL 19332211 Table1 0: Potential benefit from using in Online various proposed techniques for avoiding or mitigating ser- vice failures. Nineteen problems were examined (rather than sixteen as in Section 4.... In PAGE 29: ... Some- times redundancy must come in the form of orthogonal redundancy mechanisms, such as an alternate path to controlling an existing set of machines. #problem typeexp/mon-TTR exp/mon-TTD 1 FE-OP 2 FE-OP X 3 FE-OP 4 FE-OP X 5 BE-OP 6 BE-SW XX 7 EXT-SW X 8 NET-HW XX 9 EXT-OP X 10 FE-SW X 11 BE-HW 12 FE-SW 13 FE-OP X 14 FE-SW XX 15 FE-HW 16 FE-HW XX 17 FE-SW 18 NET-OL 19 BE-HW TOTAL 88 Table1 1: How better exposing component failures and better monitoring would have improved Time to Detection and Time to Repair for nineteen service failures from Online. These are the... In PAGE 30: ... News postings to local moderated news- groups are received from users by the front-end news daemon, converted to email, and then sent to a special email server. Delivery of the email on that server triggers execution of a script that verifies the validity of the user posting the message; if the sender is not a Technique Implementation cost Potential reliability cost Performance impact Redundancy low low very low Isolation moderate low moderate Restart low low low Pre-deployment fault/load test high zero zero Pre-deployment correctness testing medium to high zero zero Online fault/load testing high high moderate to high Online correctness testing medium to high low to moderate low to moderate Tools for checking configurations medium zero zero Better exposing failures/monitoring medium low (false alarms) low Table1... ..."