AccessMyLibrary provides FREE access to millions of articles from top publications available through your library.
Create a link to this page
Copy and paste this link tag into your Web page or blog:
What kind of processing is right for you?
Choosing a database, LAN, or Internet/intranet server is an increasingly complex process. Only a few years ago, you bought a box from traditional server vendors like NetFrame Systems or Tricord. Today's server market is in transition. The Intel Pentium Pro servers offer a new lower-cost entry point for multiprocessor server buy-in and provide some innovative features that ease implementation (like on-board glue logic for building four-processor systems), but they have some limitations. Among the vendors who offer these systems are Compaq Computer Corporation, IBM Corporation, and Advanced Logic Research Inc. (ALR). These PC servers range from US$10,000 to $50,000, while middle range systems from vendors like Tricord and NetFrame run from US$40,000 to $150,000 for eight processor servers. And at the high end, you can expect to pay up to $1 million for a server with more than 12 processors from Sequent or Pyramid Technology Corporation.
Today's servers can be loosely divided into single processor, symmetric multiprocessing (SMP), and massively parallel processing (MPP) systems. A further split occurs when you consider the processors themselves. While most vendors offer a proprietary box that generally runs some derivative of UNIX, most action today in SMP systems wraps around multiple Intel Pentium processors.
The server market: Moving in parallel with databases
As data warehouses have matured, so have parallel databases, supporting parallel query techniques, as well as users' needs for powerful parallel processing hardware. Many vendors focus on a parallel query technique called intraquery parallelism. Intraquery techniques let your software application parallelize individual queries. Within the query, you can have pipeline parallelism and partition parallelism. Parallel join pipelines execute a series of partitioned joins to process a query. Today's dominant parallel hardware architecture is a massive parallel processing machine consisting of an assortment of nodes, which are SMP computers with several processors.
Multiprocessing systems use three different architectures: shared memory, shared disk, and shared nothing. An SMP shared memory system gives all processors access to shared memory. A shared disk system has its own memory, but has access to all disk-based data. This architecture is used in system clusters. Shared nothing systems provide each processor with its own memory and disk access. MPP systems are shared nothing systems. Some users complain that shared memory systems bog down during interprocess communications, reducing their scalability. Lock contention can limit the scalability of shared disk systems, and it's …