Dotomi creates and delivers display advertising where the ad creatives and media placements are dynamically adapted in real time at the user and impression level. Dotomi has developed strategic, direct relationships with over 100 retail brands, including over forty brands from the Internet Retailer Top 100.
Dotomi wanted to deploy HBase over Hadoop to create a distributed, highly available and highly reliable data store. Dotomi needed to store about 90TB of data, including data stored for redundancy purposes. Two mission critical needs were identified in the selection process. First, their cluster environment needed to be truly redundant with no single points of failure. Second, their query response times had to be fast especially for peak traffic scenarios.
Jim Scott, lead architect/engineer, for Dotomi is deeply rooted in Hadoop as cofounder of the Chicago Hadoop users group and initiator of Nokia/Navteq’s first Hadoop project several years ago. With prior Cloudera experience, he expected to select their distribution for Dotomi earlier this year after preaching the benefits of Hadoop to his current employer. Instead, MapR was picked for several factors. Jim and his team expect to leverage MapReduce version 2 and favored MapR’s NFS-based plug and play architecture that exposes native HDFS allowing them to easily migrate data to the new file system versus the arduous task of reinstalling it. Additionally, only MapR provided for no single point of failure — a selling point that Jim said is well worth the money and an important differentiator.
Dotomi has deployed MapR distribution for Hadoop in a high availability cluster environment. They run Apache HBase included in the distribution and obtain superior response rates. They also intend to supplement their existing environment with a third cluster to run real time data analytics with such tools as OpenTSDB on HBase. With MapR they are confident that they should be able to optimize it for high performance.