•Access log: The access log contains detailed information about client connections to the directory. It provides beneficial troubleshooting information. Access logging is enabled by default.

•Error log: The error log contains detailed messages about errors and events that the directory experiences during normal operations. Error logging is enabled by default.

•Audit log: The audit log contains detailed information about changes made to each database and to server configuration. By default, audit logging is disabled.

In a typical production environment, access log should be turned off; otherwise, it can cause excessive disk I/O that affects performance — even with buffering. For improved directory server performance, HP recommends hosting the access log file on a lightly-loaded disk.

In the performance test environment, turning off the access log yielded an increase in the search throughput rate by 4%. For test results, see Table 6.

Performance measurements

This section describes performance testing for HP-UX Directory Server version 8.1 under a controlled environment with databases containing 100, 10K, 100K, 250K, 500K and 1M entries. The directory entries were inetorgPerson entries generated by the dbgen script (located in the /opt/dirsrv/bin directory on a system installed with HPDS 8.1). Performance data was measured for exact search on cn only.

Purpose

•Find out how the number of CPUs affects the performance.

•Find out how thread numbers (nsslapd-threadnumber) affects the performance.

•Find out how the dbcache size (nsslapd-dbcachesize)affects the performance.

•Find out how the cache size in terms of memory (nsslapd-cachememsize)and the cache size in terms of entries it can hold (nsslapd-cachesize)affect the performance.

•Find out how logging affects the performance.

•Find out the performance differences between a SSL connection and a non-SSL connection.

Test results

Data collection 1: (Different number of CPUs)

This set of data is collected to show the performance differences resulting from different numbers of CPUs, as measured on the Montvale-based test configuration @1.66GHz /CPUs.

•# of entries: 500k entries

•128 client threads

•HPDS parameter settings: nslapd-dbcachesize and nslapd-cachememsize are big enough to cache all the entries, and nsslapd-threadnumber is set to 6.

From Table 1, we can see that as the number of CPUs doubles, the performance nearly doubles.

Table 1: HPDS 8.1 performance in relation to numbers of CPUs

Server

 

Searches per

 

 

second

Montvale-based test configuration 1

CPU (2 cores) @1.6GHz

9471.60

Montvale-based test configuration 2

CPUs (4 cores) @1.6GHz

18650.35

12