IBM SPSS Analytic Server 2.0 Release Notes

Content

IBM SPSS Analytic Server 2.0 and fixes are available.

Contents

• Description
• System requirements and installation
• Known issues for Analytic Server

Description

IBM SPSS Analytic Server is a solution for big data analytics that combines IBM SPSS technology with big data systems and allows you to work with familiar IBM SPSS user interfaces to solve problems on a previously unattainable scale.

System requirements and installation

For information on system requirements and installation, see the installation documents.

For information on downloading the product, see the Analytic Server 2.0 Download Document, the Analytic Server 2.0 Interim Fix 1 Download Document, and the Analytic Server 2.0 Interim Fix 2 Download Document and the Analytic Server 2.0 Interim Fix 2 Download Document.

For information on using the product, see the Knowledge Center.

Known issues for Analytic Server

At time of publication, the following issues were known.

Problem: The Sort and PSM model building algorithms sometimes produce a large number of reduce jobs, which negatively impacts performance.
Solution: Analytic Server 2.0 Interim Fix 1 provides a fix for this.

Problem: Analytic Server does not support Hortonworks HDP 2.x.
Solution: Analytic Server 2.0 Interim Fix 1 provides a fix for this.

Problem: Analytic Server is not supported running on Microsoft Windows.
Solution: A future fix pack for Analytic Server 2.0 is expected to provide a fix for this.

Problem: In a 2.0 to 2.0 copy migration, Analytic Server can only migrate configuration settings to and from Hadoop distributions that use HDFS. Configuration settings to or from GPFS and MAPRFS are not supported.
Solution: A future fix pack for Analytic Server 2.0 is expected to provide a fix for this.

Problem: When performing a "manual" configuration with BigInsights 3.0 in the Administration Tool, the list of files to collect is incorrect.
Solution: Analytic Server 2.0 Interim Fix 1 provides a fix for this.

Problem: There are two issues when accessing the Analytic Server Console from Safari on iOS.

1. The tenant status dropdown list does not work
2. You cannot update the data model for File-based data sources

Solution: Use another browser when performing these actions.

Problem: In a High Availability environment, if an Analytic Server server node fails unexpectedly (for example server crash), in some cases the in-progress jobs are not resumed.
Solution: Re-run that job.

Problem: In a High Availability environment, Analytic Server servers hang occasionally upon server shutdown.
Solution: Kill the server manually.

Problem: On installations with BigInsights 3.0 with GPFS, MapReduce jobs will sometimes fail with "Failure : com.spss.analyticframework.api.exceptions.ComponentException: No message found for AEQFC3063E"
Solution: In the config.properties file in the {AS_SERVER_ROOT}/configuration directory, set mmr.thread=1.

Problem: On a MapR Hadoop distribution, Analytic Server installation will fail with not enough permissions to write on HDFS.
Solution: Make sure that the user has the necessary permissions to write on MapR filesystem, and the local user has the same UID as the user on the MapR machines.

Problem: On Linux, your Installation Manager may crash when installing or updating offerings. Additionally, you may not be able to initially install Installation Manager in GUI mode. The problem is an issue with the Cario library used when rendering the GUI.
Solution: Add the following lines to the IBMIM.ini file (under /eclipse).

-Dorg.eclipse.swt.internal.gtk.useCairo=false
-Dorg.eclipse.swt.internal.gtk.cairoGraphics=false  
 

Problem: Installing Analytic Server on BigInsights with GPFS on a non-GPFS node fails.
Solution: Run the installer on one of the GPFS nodes.

Note that console and silent mode are not affected by this.

Problem: On BigInsights with Adaptive MapReduce, Analytic Server jobs fail when one of the cluster nodes are down.
Solution: Update the Adaptive MapReduce environment with the following information in order to not reuse the JVM in case one or more nodes from the cluster are down.
1. Log on to the name node
2. Edit $EGO_CONDIR/../../soam/profiles/enabled/MapReduce6.1.1.xml
3. In the file, update
<Method name="Invoke">
              ......
               <Return actionOnSI="keepAlive" ......>

to
<Method name="Invoke">
              ......
               <Return actionOnSI="restartService" ......>

4. Run the following command:
 soamreg $EGO_CONDIR/../../soam/profiles/enabled/MapReduce6.1.1.xml -f

The following issues pertain to Analytic Server 2.0.0.1

Problem: Migration to BigInsights 4.0 fails with "Access time for hdfs is not configured".
Solution: Prior to migration, set dfs.namenode.accesstime.precision to 3600000 in the hdfs-site.xml file in your BigInsights 4.0 installation.

Problem: When configuring Analytic Server to use Hive complex types with a BigInsights 4.0 installation, jackson-core-asl-1.9.13.jar and jackson-mapper-asl-1.9.13.jar are used instead of jackson-core-asl-1.8.8.jar and jackson-mapper-asl-1.8.8.jar; however, these files cannot be found in {HIVE_HOME}/hcatalog/share/hcatalog.
Solution: Look in the /usr/iop/4.0.0.0/hbase/lib/ or /usr/iop/4.0.0.0/hadoop/lib/ directory of your BigInsights 4.0 Hive server, and copy them to the /auxlib directory on the BigInsights 4.0 Hive server.

Problem: Installing Analytic Server High Availability with BigInsights 4.0 succeeds, but jobs run against Analytic Server fail.
Solution: Apply the WebSphere Application Server 8.5.5.4 fix TFPI33526.

Problem: Creating an external table in Hive 0.14 for an Oracle NoSQL data source fails.
Solution: When you create the Hive table, the SERDEPROPERTIES parameter list must include the following parameter: "kv.json.serde.class"="org.apache.hive.hcatalog.data.JsonSerDe".

Problem: For Oracle NoSQL 3.x.x, the HiveKVStorageHandler.jar given in the Installation and Configuration Guide does not work.
Solution: Download the jar from https://github.com/dvasilen/HiveKVStorageHandler3/raw/master/dist/HiveKVStorageHandler.jar.

Problem: Installing Analytic Server on BigInsights with GPFS on a non-GPFS node fails.
Solution: Run the installer on one of the GPFS nodes.

Problem: Installing Analytic Server on BigInsights 4.0 with GPFS or Adaptive MapReduce fails.
Solution: GPFS and Adaptive MapReduce are available, but we strongly recommend that you obtain assistance from support in order to correctly configure them.

Problem: Hadoop jobs fail when run by a user other than the one that installed Analytic Server.
Solution:

1. Create /etc/krb5.conf with correct domain and realm parameters on each Analytic Server machine. A sample krb5.conf for TESTAS.COM realm follows:
  ------------------------------------    
       [libdefaults]
        default_realm = TESTAS.COM        
        dns_lookup_realm = false
        dns_lookup_kdc = false
        ticket_lifetime = 24h
        renew_lifetime = 7d
        forwardable = true

       [realms]
        TESTAS.COM = {
         kdc = kdchost.testas.com
         admin_server = kdcadminsvr.testas.com:755
        }

       [domain_realm]
        .testas.com = TESTAS.COM
 ----------------------------

2. Create user accounts for all users planned for Analytic Server in the Kerberos user repository. In case of Ldap, configure the Kerberos server to use the Ldap user repository. Note: After installing Analytic Server, the same user accounts need to used in the server.xml file.
3. Create Linux OS users for each of the user accounts created in the previous step on all Analytic Server machines and Hadoop nodes. Make sure that UID for these users matches on all machines; for example, you can test this with the kinit command.
4. Create a user home folder on HDFS for all users used in Analytic Server. For example if we use 'testuser1', then there should be home folder created for testuser1 like '/user/testuser1' on HDFS. testuser1 should have read & write permission to this folder.
5. Allow read access to all users created in the steps above to the parent folder used for analytic-root on HDFS. For example if the Analytic Server installer uses '/user/spss/analytic-root', then '/user/spss' should be given read access to all users.

Problem: When using Kerberos mode, you receive the error "com.spss.analyticframework.api.exceptions.ComponentException: Data compression initialization error: null."
Solution:
1. Stop Analytic Server ({AS_ROOT}/bin/stop.sh)
2. Edit the file {AS_SERVER_ROOT}/configuration/config.properties and search and replace seqreadwriteinjector with binreadwriteinjector
3. Start Analytic Server ({AS_ROOT}/bin/start.sh)

Known issues with IBM SPSS Modeler

At time of publication, the following issues when working with SPSS Analytic Server and SPSS Modeler were known.

Problem: On Hadoop 2.0 or higher versions, jobs that include R nodes may fail.
Solution: Make sure that the user that installed Analytic Server is added to the no password sudoers list.

Problem: When a split field is present, tree models built locally in Modeler are slightly different from tree models built by Analytic Server on HDFS, and thus produce different scores.
Solution: The algorithms in both cases are valid; the algorithms used by Analytic Server are simply newer. Given the fact that that tree algorithms tend to have many heuristic rules, the difference between the two components is normal. Looking at the predicted values, we see only about a 3% difference between the two models.

Problem: The Means node cannot produce a standard error or 95% confidence interval.
Solution: This is a limitation of the current version.

Problem: When using the Select node with the discard option, fields with null values are discarded in the result set. For example: if the criteria is to discard rows where where OCCUPATION = "Retired" all rows where OCCUPATION = "Retired" AND OCCUPATION = null are discarded.
Solution: The work around is to modify the selection criteria to add "not(field = undef)". For example: update the selection criteria to ((OCCUPATION = "Retired) and not(OCCUPATION = undef)). The result set will contain rows where the OCCUPATION field is null.

Problem: The Data Audit node cannot produce the mode for continuous fields.
Solution: This is a limitation of the current version.

Problem: Real time scoring against an Analytic Server cluster fails.
Solution: Remove the leader node from the load balancer for the cluster. The leader is the first node you start.