[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: HDB compared with LMDB



For write rates see 
https://wiki.zimbra.com/wiki/OpenLDAP_MDB_vs_HDB_performance


> On Sep 17, 2015, at 8:06 AM, Quanah Gibson-Mount <quanah@zimbra.com> wrote:
> 
> Tools that do not provide distributed capabilities are generally worthless for benching marking LDAP as the client itself is the bottleneck. 
> 
> For read rates see 
> https://mishikal.wordpress.com/2013/05/16/openldap-a-comparison-of-back-mdb-and-back-hdb-performance/
> 
> --Quanah
> 
> 
>> On Sep 17, 2015, at 6:26 AM, Philip Colmer <philip.colmer@linaro.org> wrote:
>> 
>> We're currently using OpenLDAP 2.4.38 on our production server using
>> HDB as the database type. I wanted to upgrade to the latest version
>> and take advantage of LMDB as the database type so I've built a second
>> server and transferred the data.
>> 
>> Before making that server the production server, we're running
>> performance comparisons, using ldclt[1] as a stress test tool.
>> 
>> I don't know if anyone else has got any experience of using this tool;
>> the output we're getting when running the tool doesn't show
>> significant differences in performance, but I'm not sure if it is just
>> the test type we're running or the fact that, actually, we shouldn't
>> be expecting significant differences ...
>> 
>> I've included the output below. I've removed the details of our
>> servers from the command line.
>> 
>> Can someone who has made a transition from HDB to LMDB give an
>> indication of whether or not we should see a performance difference on
>> LDAP queries and, if yes, by what sort of factor?
>> 
>> If you are familiar with ldclt, should we be running different tests
>> to get a better insight into how the two servers compare?
>> 
>> Thanks.
>> 
>> Philip
>> 
>> [1] http://docs.oracle.com/cd/E19957-01/816-6400-10/ldclt.html
>> 
>> 2.4.38 with HDB:
>> ldclt -f uid=testXXXXX -e esearch,random -r0 -R99999 -I 32
>> ldclt version 4.23
>> ldclt[1979]: Starting at Thu Sep 17 11:28:51 2015
>> ldclt[1979]: Average rate:  109.70/thr  ( 109.70/sec), total:   1097
>> ldclt[1979]: Average rate:  114.80/thr  ( 114.80/sec), total:   1148
>> ldclt[1979]: Average rate:  115.00/thr  ( 115.00/sec), total:   1150
>> ldclt[1979]: Average rate:  111.90/thr  ( 111.90/sec), total:   1119
>> ldclt[1979]: Average rate:  115.20/thr  ( 115.20/sec), total:   1152
>> ldclt[1979]: Average rate:  114.80/thr  ( 114.80/sec), total:   1148
>> ldclt[1979]: Average rate:  113.70/thr  ( 113.70/sec), total:   1137
>> ldclt[1979]: Average rate:  111.20/thr  ( 111.20/sec), total:   1112
>> ldclt[1979]: Average rate:  114.80/thr  ( 114.80/sec), total:   1148
>> ldclt[1979]: Global average rate: 1021.10/thr  (113.46/sec), total:  10211
>> 
>> 2.4.41 with MDB:
>> ldclt -f uid=testXXXXX -e esearch,random -r0 -R99999 -I 32
>> ldclt version 4.23
>> ldclt[1967]: Starting at Thu Sep 17 11:26:55 2015
>> ldclt[1967]: Average rate:  109.70/thr  ( 109.70/sec), total:   1097
>> ldclt[1967]: Average rate:  111.00/thr  ( 111.00/sec), total:   1110
>> ldclt[1967]: Average rate:  112.20/thr  ( 112.20/sec), total:   1122
>> ldclt[1967]: Average rate:  113.00/thr  ( 113.00/sec), total:   1130
>> ldclt[1967]: Average rate:  112.70/thr  ( 112.70/sec), total:   1127
>> ldclt[1967]: Average rate:  112.50/thr  ( 112.50/sec), total:   1125
>> ldclt[1967]: Average rate:  113.00/thr  ( 113.00/sec), total:   1130
>> ldclt[1967]: Average rate:  112.80/thr  ( 112.80/sec), total:   1128
>> ldclt[1967]: Average rate:  112.30/thr  ( 112.30/sec), total:   1123
>> ldclt[1967]: Average rate:  112.90/thr  ( 112.90/sec), total:   1129
>> ldclt[1967]: Global average rate: 1122.10/thr  (112.21/sec), total:  11221
>