Google for years has assembled its own server hardware in order to operate at scale efficiently. Now it appears the company has customized networking silicon.
Google software engineer Jacques Pienaar on Tuesday contributed backend source code to the open source LLVM compiler project in support of a processor identified as "Lanai."
Pienaar said that Lanai has 32 32-bit processors, with two used for fixed values and four used for program state tracking. The hardware lacks floating point support, meaning it isn't well-suited for mathematical calculations.
Google did not respond to a request for comment.
In response to a question seeking further details on the LLVM mailing list, Google software engineer Chandler Carruth said, "This is internal hardware for us, so there's not a lot we can share, and you can't really grab a version of the hardware."
Networking company Myricom, acquired by CSP in 2013, makes a network adapter with a chip called Lanai that's designed for parallel processing. In the past, the chip has been referred to as "LANai," a variation in capitalization that suggests support for both networking and artificial intelligence functions. CSP did not respond to a request for comment.
Google, meanwhile, has acquired a considerable portion of Myricom's engineering talent. Nan Boden, once CEO of Myricom, is now a director of engineering at Google, along with at least eight former colleagues, including former Myricom CTO Jakov Seizovic.
A 2002 paper from the the French Institute for Research in Computer Science and Automation (INRIA) describes the LANai processor as suitable for a network architecture designed around an "oversimplified, dumb, and performant core, with potential 'intelligence' at the edges of the network thanks to programmable processors."
Presumably, Google's custom networking processors can be used for network interconnection to increase data center efficiency through software-defined functions. They may also provide some security benefit, given concerns about the extent to which national intelligence agencies covet knowledge of vulnerabilities in commercial network hardware. Myricom currently suggests its networking adapters are well-suited to efficient packet sniffing and video streaming.
A report in The Register indicates Google has been developing low-latency code for its network hardware to accelerate memcache workloads. Memcache, a distributed memory caching system, stores data in memory to reduce the number of times data needs to be read from a database.
Given the scale at which Google operates, even small optimizations can lead to significant savings. This explains the company's interest in custom hardware.
According to Bloomberg, Google purchases as many as 300,000 chips for its custom servers every quarter and has encouraged Qualcomm's efforts to enter the server chip market, currently dominiated by Intel.
Rising stars wanted. Are you an IT professional under age 30 who's making a major contribution to the field? Do you know someone who fits that description? Submit your entry now for InformationWeek's Pearl Award. Full details and a submission form can be found here.