On Fri, May 2, 2008 at 9:12 AM, fschmidt <fschmidt@xxxxxxxxx> wrote: > > An implementation of clustering without locking would start by comparing the > index to the table from the beginning to find the first mismatch. Rows > before the mismatch are fine, and can be left alone. From here on, go > through the index and rewrite each row in order. This will put the rows at > the end of the table in cluster order. When done, vacuum the table. This > will result in a clustered table without any locking needed. Those few > records that were updated while clustering was happening will be out of > order, but that should only be a few. In your proposal, a large amount of dead tuple can be generated in both table and index. This is a serious problem that spoils the effect of clustering. -- Fujii Masao NIPPON TELEGRAPH AND TELEPHONE CORPORATION NTT Open Source Software Center
