In large distributed networks of computers, it is often the case that a subset of machines wants to cooperate to perform a task. Before they can do so, these machines need to learn of the existence of each other. In this paper we are interested in distributed algorithms whereby machines in a network learn of other machines in the network by making queries to machines they already know. The algorithms should be efficient both in terms of the time required and in terms of the total network communication required until all machines have discovered all other machines. We propose a very simple algorithm called Name-Dropper whereby all machines learn about each other within O(log 2 n) rounds, where n is the number of machines in the network. The total number of connections required is O(n log 2 n), and the total number of pointers which must be communicated is O(n 2 log 2 n). Each of the preceding bounds is optimal to within polylogarithmic factors. Contact author: Mor Harc...

It is known that there exists a PCP characterization of NP where the verifier makes 3 queries and has a one-sided error that is bounded away from 1; and also that 2 queries do not suffice for such a characterization. Thus PCPs with 3 queries possess non-trivial verification power and motivate the task of determining the lowest error that can be achieved with a 3-query PCP. Recently, Hastad (1997) has shown a tight characterization of NP by constructing a 3-query PCP verifier with "error" arbitrarily close to 1/2. Unfortunately this verifier makes two-sided error and Hastad makes essential use of this feature. One-sided error, on the other hand, is a natural notion to associate with a proof system, since it has the desirable property that every rejected proof has a short counterexample. The question of determining the smallest error for which there exists a 3-query PCP verifier making one-sided error and accepting an NP-complete language, however, remained open. We resolve this question by showing that NP has a 3-query PCP with a one-sided error that is arbitrarily close to 1/2. This characterization is tight, i.e., the error cannot be lower. This result is in seeming contradiction with the results of Trevisan (1997) and Zwick (1998) who show that in order to recognize an NP-complete language, the error probability of a PCP verifier making 3 non-adaptive queries and having one-sided error must be at least 5/8. We get around this bottleneck by designing an adaptive 3-query PCP for NP. Our result yields the first tight analysis of an adaptive PCP; and reveals a previously unsuspected separation between the powers of adaptive and non-adaptive PCPs. Our design and analysis of adaptive PCPs can be extended to higher number of queries as well and we give an example of such a proof system with 5 queries. Our adaptive verifiers yield proof systems whose error probabilities match those of previous constructions, while also achieving one-sidedness in the error. This raises new questions about the power of adaptive PCPs, which deserve further study.

We present a Monte Carlo algorithm for testing multivariate polynomial identities over any field using fewer random bits than other methods. To test if a polynomial \\(P(x_1, ..., x_n)\\) is zero, our method uses \\(\\sum_{i=1}^n \\log \\lceil d_i+1 \\rceil\\) random bits, where \\(d_i\\) is the degree of \\(x_i\\) in \\(P\\), to obtain any inverse polynomial error in polynomial time. The algorithm applies to polynomials given as a black box or in some implicit representation such as a straight-line program. Our method works by evaluating P at truncated formal power series representing square roots of irreducible polynomials over the field. This approach is similar to that of Chen and Kao (STOC '97), but with the advantage that the techniques are purely algebraic and apply to any field. We also prove a lower bound showing that the number of random bits used by our algorithm is essentially optimal in the black-box model.

We describe a family of caching protocols for distrib-uted networks that can be used to decrease or eliminate the occurrence of hot spots in the network. Our protocols are particularly designed for use with very large networks such as the Internet, where delays caused by hot spots can be severe, and where it is not feasible for every server to have complete information about the current state of the entire network. The protocols are easy to implement using existing network protocols such as TCP/IP, and require very little overhead. The protocols work with local control, make efficient use of existing resources, and scale gracefully as the network grows. Our caching protocols are based on a special kind of hashing that we call consistent hashing. Roughly speaking, a consistent hash function is one which changes minimally as the range of the function changes. Through the development of good consistent hash functions, we are able to develop caching protocols which do not require users to have a current or even consistent view of the network. We believe that consistent hash functions may eventually prove to be useful in other applications such as distributed name servers and/or quorum systems.