We study the online problem of holding a number of idle threads on an application server, which we have ready for processing new requests. The problem stems from the fact that both creating/deleting and holding threads is costly, but future requests and completion times are unpredictable. We propose a practical scheme of barely random discrete algorithms with competitive ratio arbitrarily close to e/(e−1), where e ≈ 2.718 is Euler’s number. The competitive ratio is sharply concentrated for any input. The results are generalizations of the well-known result for the rent-to-buy problem. Key words: multithreading, spin-block problem, online algorithms, randomization, implementation issues ∗ ded abstract appeared in the Proceedings of the 10th International Symposium on Algorithms and Computation ISAAC’99, Chennai/India, Lecture Notes in Computer Science 1741 (Springer), pp. 27-36. 1