![XD logo.png](http://boulderbibraintrust.org/brain_trust_blog/XD%20logo.png)
Founded in 2004,
XtremeData presents their ideas and products about "Big Data" with
Ravi Chandran, CEO/Founder and
Geno Valente, Vice President of Sales and Marketing. Ravi has a background in parallel processing machine in medical imaging systems. Geno comes from a digital engineering background especially using Field Programmable Gate Arrays (FPGA) for low latency equity trading. The management team has strong connections with Knightsbridge Consulting, now part of HP.
![XD BI-DA.png](http://boulderbibraintrust.org/brain_trust_blog/XD%20BI-DA-thumb-150x67.png)
Geno started with an overview of their dbX analytics appliance. To be successful in the data appliance market, he emphasized that you need three areas of
expertise: computer
architecture, database engine internals, and domain knowledge of the
analytics space.
XtremeData has all three. The computing sub-system for one in four
medical CT scanners worldwide was created by the engineering team that
is currently at
XtremeData. He distinguished between Business Intelligence and Data Analysis (Analytics), which causes considerable discussion whether this distinction was useful to the industry. This chasm between the two categories was motivated by the table at the right. (click for full resolution) There should be a (loosely coupled) closed-loop between the Enterprise systems with the data analysis system, with strategies generated in the analysis impacts the operational systems via application projects.
![XD rack.png](http://boulderbibraintrust.org/brain_trust_blog/XD%20rack-thumb-100x184.png)
We continued with a technical description of their dbX analytics appliance, priced at $20K per TB of uncompressed user data and scales from 1 TB to 3.8 PB.
![XD arch.png](http://boulderbibraintrust.org/brain_trust_blog/XD%20arch-thumb-100x111.png)
Their architectural differentiation is the heavy reliance on their FPGA chip that performs a broad spectrum of SQL operations (select/filter, partition, join group, aggregate, distribute). Competitors from this respect are Netezza and Kickfire. Ravi argued that they leverage hardware-assist from the bottom levels of the query parse tree to the higher levels, thus achieving high performance. Deep question by Neil about doing a median calculation had Ravi to explain their internal sort mechanism. FPGA chipsallow 10x performance with 1/3 the power consumption. MapReduce functionality is inherent in the dbX architecture but is implemented as User Defined Functions with C/C++, instead as a separate API. In addition, custom functions can be embedded in the FPGA for special customer requests.
![XD config.png](http://boulderbibraintrust.org/brain_trust_blog/XD%20config-thumb-100x19.png)
After the break, we looked at the various product configuration and key industry trends in hardware storage, servers, network, and database engines. Applications for dbX are: bioinformatics to march and find genomic sequences and financial analysis of US consumer credit data. We got into the processing details that showed good load balancing at each stage.
They share the announcement of the partnership with Cray to produce a Personal Data Warehouse (PDW) appliance with 3 nodes of 5 TB of uncompressed user data, deployed in an office environment. KXEN has their software running in this type of box. We then had a good (confidential) discussion on market positioning of the PDW.
My Take...
One aspect that impressed me about XtremeData is their intellectual property in FPGA technology. They have a patent on using FPGA as a CPU on standard blade servers ... which means that XtremeData can do much more than SQL processing on Big Data. Some applications in massive image recognition are amazing and could revolutionize the business of major corporations. So, XtremeData is a company to know if your company has complex and specialized analytics on Big Data.