NoSQL with HBase
Judging by the last numbers I’ve seen, it looks like both NoSQL and MyNoSQL are quite loved in France — according to current stats, France comes in the 4th place. So, I thought that MyNoSQL’s French-reading audience will appreciate this nice roundup article (linked from the title) by Michael Figuiere on NoSQL and HBase (French), that also includes references to the lessons learnt by running HBase in production for 2 years.
On associe en général l’utilisation de HBase et de MapReduce de manière générale avec des configurations matérielles de type commodity hardware (matériel courant non spécifique aux serveurs). Lars George utilise pour sa part des machines octo cœurs, afin d’assurer une puissance de calcul importante sur ses tâches distribuées tout en conservant un nombre de machines raisonnable.
[…]
L’administration d’une base de données HBase au quotidien est plus proche de celle d’un middleware Java que de celle d’une base de données relationnelle classique. Nul besoin d’un DBA spécialisé donc.
[…]
La plupart des pratiques habituelles doivent être repensées lors de la mise en oeuvre d’une telle base de données.
For the other readers, I”ve just included a couple of my notes:
- the system referred in the post is storing tens of Terra of data
- HBase have seen some good performance improvements over the last versions that brought it closer to Cassandra performance
- HBase administration has more in common with (Java) middleware administration than RDBMS
- tuning HBase is mostly done at the JVM level; HBase configuration is much much simpler than Oracle or even MySQL
- for multi-criteria or full-text searches, Lucene (with Katta or other distributed configurations) is still the preferred solution
via: http://blog.xebia.fr/2009/11/18/devoxx-jour-1-nosql-avec-hbase/print/
