The shipping of Oracle’s Big Data Appliance earlier this month could pressure major rivals like IBM, Hewlett-Packard and SAP to come up with Hadoop offerings that tightly bundle hardware, software and other tools, analysts say.
On the day the appliance shipped, Oracle announced that its new offering will run Cloudera’s Apache Hadoop implementation and management software.
The tightly bundled Big Data Appliance includes Sun server hardware — 18 Linux-based x86 Sun servers with 216 processor cores, 864GB of working memory and 648TB of raw disk storage — along with an Oracle NoSQL database, an open-source distribution of R statistical software and a copy of Oracle’s Java HotSpot Virtual Machine.
The appliance provides 40Gbps InfiniBand connectivity among the nodes, a rarity in Hadoop deployments, many of which use Ethernet to connect the nodes. Multiple racks can be tethered together in a cluster configuration, noted George Lumpkin, vice president of data warehousing product management at Oracle.
Lumpkin added that the company this month also started shipping Big Data Connector drivers for exchanging data between the Big Data Appliance and several other Oracle products, including Oracle Database 11g, the Exadata data warehousing and online transaction processing appliance and the new Exalytics business intelligence appliance.
Oracle unveiled the Big Data Appliance, along with the NoSQL database, at its OpenWorld user conference last fall.
David Menninger, an analyst with Ventana Research, said the offering will likely prompt IT managers to look more closely at bundled systems that could be “one-stop shops” for corporate data needs.
The Oracle system is designed to manage and analyze data sets, such as telemetry data, click-stream data or other log data, that are too large or are otherwise unsuitable for keeping in databases, Menninger said.
James Kobielus, an analyst at Forrester Research, said if IT managers do in fact take to the bundled Hadoop system, the other top tech vendors, many of which have already spent considerable sums on big data research and development, could be pushed to follow suit. “[Multiple bundled systems would be] good for the market. It gives customers a choice of commercially available products,” he added.
Most enterprise Hadoop deployments have so far been custom built by internal IT engineers — companies had to buy the hardware, license the software and integrate it on their own, Kobielus noted.
Kobielus said the growing popularity of the open-source Apache Hadoop framework should in time help to further increase the number of packaged big data options available to IT managers.
Menninger said Oracle’s unexpected decision to use the Cloudera implementation of Hadoop could prove beneficial to corporate users — and Oracle. When Oracle first announced the appliance, it was expected to run an Oracle distribution of Hadoop, he said.
The selection of an established implementation means that the vendor won’t have to “fight to establish another distribution,” Menninger added. “If Oracle had created its own distribution, it would have given the competitors a potential weakness to exploit.”