Balancing flexibility and a myriad of computational demands with the need for efficiency in web scale data centers drove Microsoft data center researchers and colleagues to examine hardware alternatives in a project known as Catapult. Delivered at the 41st International Symposium on Computer Architecture (ISCA) this month, a paper, titled A Reconfigurable Fabric for Accelerating Large Scale Data Center Services, describes an effort to combine programmable hardware and software that uses field programmable gate arrays (FPGA) to deliver performance improvements of as much as 95 percent.
Building and operating data centers that can be flexible and efficient for hyperscale needs has been a challenge for a select handful of companies. Five years ago Google’s Luiz Andre Barroso and Urs Holzle published a paper about looking at a data center as a computer, denoting the company’s use of warehouse-scale machines in its approach to data center infrastructure.
After announcing its $1.1 billion data center expansion in Iowa, Microsoft Director of Data Center Services Kevin Williams said that the new site would “house a Generation 5 facility incorporating fully-integrated software applications that are built as distributed systems into every aspect of the physical environment — from the server design to the building itself — and drive systems integration for greater reliability, scalability, efficiency and sustainability.”
Catapult is a reconfigurable fabric embedded into each half rack of 48 servers in the form of a small board with a medium sized FPGA and local DRAM attached to each server. The Microsoft approach of using FPGAs will enable programmable hardware to specifically benefit Microsoft’s search algorithms. In the evaluation deployment outlined in the paper, Catapult’s interconnected nodes linked by high-bandwidth connections were tested on a collection of 1,632 servers to measure its efficacy in accelerating the workload of a production web-search service.
Also making use of FPGAs to chase the hyperscale data center market, Intel recently said that it would produce a hybrid chip that bolts an FPGA onto its high-end Xeon E5 server chip. Specialized chips that can be programmed have enabled optimized workloads for things like bitcoin mining. Semiconductor company Altera makes reconfigurable logic with on-chip memory and DSP blocks for a software defined data center. The company announced it is working with Microsoft Research and Bing to accelerate portions of the web search engine.
“We designed a platform that permits the software in the cloud, which is inherently programmable, to partner with programmable hardware,” said Microsoft Researcher Doug Burger. “You can move functions into custom hardware, but rather than burning them into fixed chips [application-specific integrated circuits], we map them to Altera FPGAs, which can run hardware designs but can be changed by reconfiguring the FPGA. We’ve demonstrated a ‘programmable hardware’ enhanced cloud, running smoothly and reliably at large scale.
“This portends a future where systems are specialized dynamically by compiling a good chunk of demanding workloads into hardware,” Burger went on. “I would imagine that a decade hence, it will be common to compile applications into a mix of programmable hardware and programmable software.”
If all goes well in research trials, Bing plans to roll out FPGA-accelerated servers to process customer searches in one of its data centers starting in early 2015.