John Lewis wrote:
I am wondering if anybody has benchmarked OpenLDAP as a generic network object storage service. I know Networked Key Value Databases are getting more popular because they seem to scale laterally better.
Probably nothing recent. Back in the CORBA days it was probably a frequent event.
Most of the so called NoSQL databases use some kind of JSON over HTTPS network transport. I like the idea of using something LDAP based because of the IETF standardized interface lowers the risk of sudden changes to the interface forcing a program change, kind of like how Microsoft broke Skype calls for Blog Talk Radio.
LDAP isn't a Key/Value database; it could be used as one but there would be a lot of unnecessary overhead - much like using SQL as only a Key/Value database. The value in using LDAP comes from mapping an application's higher level data model onto LDAP entries. Not just individual keys and blob values.
I know Howard Chu has been reporting a lot of awful but true things about MongoDB over the last couple of years. https://twitter.com/hyc_symas/status/664197817498509312 https://twitter.com/hyc_symas/status/740339320645689344 https://twitter.com/hyc_symas/status/817696625313349632
What I would really want to see is some numbers comparing OpenLDAP to networked key value databases that are in production such as Reddit's two table Postgres https://kev.inburke.com/kevin/reddits-database-has-two-tables/ or Redis or even HBase from Hadoop.
Could probably rig up a YCSB test without too much trouble. Frankly though I've got very low confidence in any java-based benchmark utility. Our experience with java-based SLAMD over the years has shown that it vastly underreports OpenLDAP's performance simply because SLAMD/java are not efficient enough to drive high enough loads at slapd.