******************************************************************************* * IBM DB2 Intelligent Miner for Data * Version 6.1.0 * for Sun Solaris * Fixpak U469938 ******************************************************************************* * * * This FixPak contains fixes that apply only to the IBM DB2 Intelligent Miner * for Data, Version 6.1.0, for Sun Solaris servers. * See the section "Installation" below for details on how to install this * FixPak. ******************************************************************************* -------------------------------------------------------------------------------- Contents: The following fixes are included in this Program Temporary Fix (PTF): COMPONENT PROBLEM DESCRIPTION -------------------------------------------------------------------------------- Stepwise Polynomial Regression (SPR) Handling of missing values, KLG live table Demographic Clustering Too many messages Data Access UBS problem Client server API Client/server problems Tree Classifier Progress indicator Neural Clustering Confidence, output width Neural Clustering Sun build for Neural model deployment Parallel MPI finalize Server Support UDTs again Tree Classifier Suppress generation of large distributions Support for named pipes Support for named pipes (also see below) Associations Performance problem -------------------------------------------------------------------------------- -------------------------------------------------------------------------------- Installation: You must have installed the IBM DB2 Intelligent Miner for Data successfully. No other PTFs are required. To install the PTF package, follow these steps: - Close any client of IBM DB2 Intelligent Miner for Data that might be executing - Type the following commands in the same directory where the PTF.PKG file resides: * 'pkgadd -d PTF_U469938.PKG IMiner' to update the Intelligent Miner base package * 'pkgadd -d PTF_U469938.PKG IMinerTK' to update the Intelligent Miner technical toolkit * 'pkgadd -d PTF_U469938.PKG IMinerP' to update the Intelligent Miner parallel support - You need not reboot your system. Named pipe support for IBM DB2 Intelligent Miner V6.1 Basically, the usage of named pipes is similar to the usage of flat files. See 'Using the Intelligent Miner for Data' for information on how to use flat files. ADDITIONAL HINTS AND TIPS 1. Check if the named pipe '/tmp/mypipe' was already created on your system by another application. If it is not yet available, create a named pipe by using the following command: mkfifo /tmp/mypipe where mypipe is a name of your choice. 2. Use the IM GUI to create a data object and a mining object that fits to your needs. The wizard that assists you to create the data object automatically detects that you have specified a named pipe. It does not wait for the pipe to be filled. A warning message will be displayed informing you, that you must specify the record length and the field description correctly although you cannot see the data. Alternatively, you can follow these steps in the data definition phase: a) Write the data once to the file mypipe in the tmp directory b) Use the file mypipe to create the data object c) Remove the file mypipe from the tmp directory d) Create a named pipe called mypipe in the tmp directory 3. Fill the pipe. For example: when your data producing process 'myprocess' writes data to stdout, you can call it by using the following command myprocess > /tmp/mypipe 4. Start the mining run by using the GUI or the command 'idmeruns'. Restriction: Make sure that only one mining run at a time reads a certain named pipe. Note: You can also start the mining run first and fill the pipe afterwards. In this case the mining run waits for the pipe to be filled and closed. APPENDIX: DESIGN TOPICS This appendix covers additional background information Objectives On SOLARIS, kernels and GUI must accept the data to be mined from named pipes. A process apart from the Intelligent Miner for Data writes data to a named pipe. If you want that the mining kernels use named pipes instead of flat files, you must specify the named pipe in the Intelligent Miner data object. Properties of named pipes • When you start a mining kernel (the reader), a pipe is opened and continues reading as long as the data source (the writer) does not close the pipe. • fopen(), fread(), fwrite() behave the same way on pipes and files. • Any necessary buffering between writer and reader is done automatically by the operating system. • If necessary, the writer waits on the reader until data is read. This is handled automatically by the operating system. • rewind() is not possible on named pipes. • In contrast to files, a named pipe cannot be read by multiple processes at the same time. Impact on IM components • You can open named pipes only once for processing by the kernel by using fopen(). Do not rewind(), do not fopen() by other processes (idmerlen, server). • Make sure that one pass of reading is enough for test and application mode. Save the data in a temporary file for training mode, so that the original data must be read only once. Now you can remove the temp file. -------------------------------------------------------------------------------- Warning: If you want to install additional components of the IBM DB2 Intelligent Miner for Data at a later point of time, make sure to reinstall all PTF packages. If you do not reinstall the PTFs, your IBM DB2 Intelligent Miner for Data might become unusable, because you have a mixture of old and fixed files. --------------------------------------------------------------------------------