Data collection 2: (Different values of nsslapd-threadnumber)

This set of data is collected to show how the tuning of HPDS worker threads affects the performance.

•Number of entries: 100,000 entries

•Number of client threads: 128

•HPDS settings

onsslapd-dbcachesize: 100,000,000 (100MB) o nsslapd-cachememsize: 1,000,000,000 (1GB) o nsslapd-accesslog-logging-enabled: off

o nsslapd-threadnumber: Table 2 shows how values of the nsslapd- threadnumber parameter affect search performance.

Table 2: Performance in relation to nsslapd-threadnumber

nsslapd-threadnumber

Searches per second

2

13194.78

3

18563.93

4

19148.13

6

18680.52

8

18490.97

12

17517.95

16

17379.27

24

17206.30

32

17175.17

64

16345.50

96

16178.92

128

15300.62

From Table 2, we can see that search throughput is highest when the number of worker threads is the same as the number of CPUs.

Data collection 3: (Different values of nsslapd-dbcachesize)

This set of data is collected to show how nsslapd-dbcachesizeaffects the performance:

•Number of entries: 1,000,000 entries

•Number of client threads: 128

•HPDS settings

onsslapd-dbcachesize: Table 3 shows the search performance against various values of the nsslapd-dbcachesizeparameter.

onsslapd-cachememsize: 3,000,000,000 (2GB) o nsslapd-cachesize:-1

o nsslapd-accesslog-logging-enabled: off

o nsslapd-threadnumber: 4

Table 3: Performance in relation to nsslapd-dbcachesize

nsslapd-dbcachesize

Searches per second

500,000

14861.93

1,000,000

15082.67

5,000,000

15194.85

10,000,000

15231.68

50,000,000

18243.45

100,000,000

18348.92

200,000,000

17743.15

13