HP-UX Directory Server 8.1 Performance Tuning and Sizing Guidelines

15
Table 5: Performance in relation to nsslapd-cachesize/nsslapd-cachememsize for
various database sizes
Number of Entries
in the Database
Searches per second with
cachememsize 100% or max,
cachesize -1,
dbcachesize 100% or max
Searches per second with
cachememsize default,
cachesize 1,
dbcachesize 100% or max
100 19597.22 12049.68
10,000 19169.68 11314.75
100,000 18983.23 11399.68
250,000 18545.38 11056.25
500,000 18487.29 11048.27
1,000,000 18398.92 10817.41
From Table 5, you can see from the second column that if all the entries are cached in the entry cache
(cachesize of -1, unlimited entries), and if all the pages are cached in the dbcache, then the search
throughput changes minimally, no matter how big the database is. So, the number of database
entries does not affect performance as long as all entries are cached. However, as shown from the
third column, if only a very small percentage of entries can be held in cache (cachesize of 1), the
search throughput for different sizes of databases also varies minimally. So, if dbcachesize is big
enough to hold all the pages, then the average exact search throughput for any size of database with
HPDS 8.1 will be between around 18398 to 19597 operations/second, depending on the number of
entries present in the database (the size of the entry cache). If the cache size is restricted to minimize
the number of entries in cache, the average search throughput drops considerably, ranging from
around 10817 to 12049 operations/second.
Data collection 5: (Access log on compared to Access log off)
This set of data is collected to show how logging can affect the performance.
• Server: Montvale-based test configuration 8 CPUs @1.6GHz each
• Number of entries: 500K
• Number of client threads: 128
• HPDS parameter settings: nsslapd-dbcachesize and nsslapd-cachememsize are big
enough to cache all the entries, and nsslapd-threadnumber is set to 6.
Table 6 shows that by turning the access log off, the exact search throughput can increase about 4%
in a Montvale-based test bed.
Table 6: Performance in relation to nsslapd-accesslog setting: (on or off)
Access log on or off Searches per second
ON 17127.78
OFF 18463.38
Data collection 6: (SSL connection enabled compared to SSL connection disabled)
This set of data is collected to show the performance impact if SSL connection is enabled or disabled.
• Server: Montvale-based test configuration 8 CPUs @1.6GHz each
• Number of entries: 500K
• Number of client threads: 128
• HPDS parameter settings: assume nsslapd-dbcachesize and nsslapd-cachememsize
are big enough to cache all the entries. nsslapd-threadnumber is set to 6.
• Special setting in client: “Bind for every operation” option is selected.