Title: XtremeData Parallel Database Appliance and In-Socket FPGA Accelerator

Speaker: Joe Ward, Geno Valente and Vellore Ravichandran, XDI Inc.

Date/Time: Tuesday, March 10, Wednesday, March 11, 2009, 1:00 pm – 3:00 pm

Location: CSRI Building, Room 90 (Tuesday), CSRI Room 279 (Wednesday)

Brief Abstract: Tuesday: XDI's Database Appliance
XtremeData, Inc., of Schaumburg, IL (http://www.xtremedatainc.com) has developed a parallel DataBase Analytics Appliance.  This new product is ready for proof-of-concept evaluations now.  Product roll-out is scheduled for Q1 2009.

The system has been designed from first principles to provide scalable and sustainable performance.  It incorporates all components of a traditional database analytics environment such as an RDBMS engine, computing servers, and a storage and interconnect network into an integrated appliance.  The system leverages COTS technology for servers, storage & interconnect, and couples these to a hardware-accelerated software stack based on the popular open-source PostgreSQL database to provide extremely high performance.

The Appliance can sustain 1TB/min of complex SQL processing with support for 20TB of user data all within the footprint of a single 19-inch tower. This performance has been achieved by coupling a re-engineered open-source RDBMS engine to SQL processing on a re-configurable silicon chip.

Wednesday: In-Socket FPGA Acceleration
XtremeData uses FPGAs for hardware acceleration to provide dramatic performance improvement while cutting power consumption compared to traditional HPC systems.  XtremeData ISA¹s (In Socket Accelerators) fit directly into processor sockets on a motherboard (on Dual or Quad Opteron or Xeon Systems) and can increase performance of certain applications up to 200x over conventional approaches.

The features of the in-socket accelerator include the following
capabilities:

• Direct link to the CPU with ³master² privileges to decrease latency and increase performance of certain applications by 200x
• FPGA uses all motherboard resources meant for CPU:
         ­ Intel Processors: Front Side Bus links, memory interface, power supply, heat-sink
         ­ AMD Processors: HyperTransport Links, memory interface, power supply, heat-sink
• VHDL/Verilog (Altera Quartus) and C/C++ (through third party
  compilers) programming methodologies supported
          ­ IP libraries available for many applications

CSRI POC: Andy Wilson, (505) 844-1089