Adaptive Data Access Methods

In this chapter, we present an adaptive method we proposed to improve efficiency of wireless data access. The proposed method is based on the observation that the index tree based methods preserve good access time as well as stable overall performance and that the hashing method exhibits good tuning time. By combining these two techniques, the new method takes the advantages of both techniques. As a result, it exhibits greater flexibility and better performance.

As previously shown, the tuning time of hashing depends on Nc, which in turn depends on the number of collisions. The number of collisions normally depends on how good the hashing function is. Thus, deriving a good hashing function is crucial for good tuning time. However, because of the heterogeneity of broadcast data in different applications, this is usually difficult to achieve. Furthermore, the hashing function itself is included in every data bucket. This obviously increases the broadcast cycle, and thus, the access time. In our method, hashing is used only to partition the broadcast data into a number of partitions. B+ tree technique is then used to index each partition. The hashing function is only stored at the beginning of each partition. Since the number of partitions is a small number compared to the number of all data items, the overhead introduced is much smaller than that in hashing based method. We now show how hashing and index tree techniques are combined in our method:

• First level hashing: Generate hashing value (h1) for key attribute of each data item using a hashing function H1. The hashing value generated at this step is similar to the hash value in simple hashing method.

  Second level hashing: Use another hashing function (H2) to produce p second level hashing values (h2) based on the value of h1 generated in the first step. This process will partition the broadcast data into p parts. All data items having the same second level hashing value will be in the same partition.

  Generating the index tree: Within each partition, generate an index tree on the key attributes of the data items in the partition.